Version bump

- Create a known_hosts file if it doesn't exist
  Otherwise ssh-keygen exits
- Move some common rsync code to common_start_setup
- Build the project before deploying it

Cargo.toml

@@ -1,7 +1,7 @@
 [package]
 name = "solana"
 description = "Blockchain, Rebuilt for Scale"
-version = "0.7.0-beta"
+version = "0.7.0-rc.6"
 documentation = "https://docs.rs/solana"
 homepage = "http://solana.com/"
 readme = "README.md"
@@ -37,10 +37,6 @@ path = "src/bin/fullnode-config.rs"
 name = "solana-genesis"
 path = "src/bin/genesis.rs"
 
-[[bin]]
-name = "solana-mint"
-path = "src/bin/mint.rs"
-
 [[bin]]
 name = "solana-drone"
 path = "src/bin/drone.rs"
@@ -85,6 +81,7 @@ futures = "0.1.21"
 clap = "2.31"
 reqwest = "0.8.6"
 influx_db_client = "0.3.4"
+dirs = "1.0.2"
 
 [dev-dependencies]
 criterion = "0.2"

README.md

@@ -84,17 +84,24 @@ Now start the server:
 $ ./multinode-demo/leader.sh
 ```
 
-To run a performance-enhanced fullnode on Linux,
-[CUDA 9.2](https://developer.nvidia.com/cuda-downloads) must be installed on
-your system:
-```bash
-$ ./fetch-perf-libs.sh
-$ SOLANA_CUDA=1 ./multinode-demo/leader.sh
-```
-
 Wait a few seconds for the server to initialize. It will print "Ready." when it's ready to
 receive transactions.
 
+Drone
+---
+
+In order for the below test client and validators to work, we'll also need to
+spin up a drone to give out some test tokens.  The drone delivers Milton
+Friedman-style "air drops" (free tokens to requesting clients) to be used in
+test transactions.
+
+Start the drone on the leader node with:
+
+```bash
+$ ./multinode-demo/drone.sh
+```
+
+
 Multinode Testnet
 ---
 
@@ -104,15 +111,18 @@ To run a multinode testnet, after starting a leader node, spin up some validator
 $ ./multinode-demo/validator.sh ubuntu@10.0.1.51:~/solana 10.0.1.51
 ```
 
-To run a performance-enhanced fullnode on Linux,
+To run a performance-enhanced leader or validator (on Linux),
 [CUDA 9.2](https://developer.nvidia.com/cuda-downloads) must be installed on
 your system:
 ```bash
 $ ./fetch-perf-libs.sh
-$ SOLANA_CUDA=1 ./multinode-demo/leader.sh ubuntu@10.0.1.51:~/solana 10.0.1.51
+$ SOLANA_CUDA=1 ./multinode-demo/leader.sh
+$ SOLANA_CUDA=1 ./multinode-demo/validator.sh ubuntu@10.0.1.51:~/solana 10.0.1.51
+
 ```
 
 
+
 Testnet Client Demo
 ---
 

benches/bank.rs

@@ -32,7 +32,7 @@ fn bench_process_transaction(bencher: &mut Bencher) {
 
             let rando1 = KeyPair::new();
             let tx = Transaction::new(&rando0, rando1.pubkey(), 1, last_id);
-            assert!(bank.process_transaction(&tx.clone()).is_ok());
+            assert!(bank.process_transaction(&tx).is_ok());
 
             // Finally, return the transaction to the benchmark.
             tx
@@ -58,5 +58,9 @@ fn bench(criterion: &mut Criterion) {
     });
 }
 
-criterion_group!(benches, bench);
+criterion_group!(
+    name = benches;
+    config = Criterion::default().sample_size(2);
+    targets = bench
+);
 criterion_main!(benches);

benches/ledger.rs

@@ -32,5 +32,9 @@ fn bench(criterion: &mut Criterion) {
     });
 }
 
-criterion_group!(benches, bench);
+criterion_group!(
+    name = benches;
+    config = Criterion::default().sample_size(2);
+    targets = bench
+);
 criterion_main!(benches);

benches/signature.rs

@@ -16,5 +16,9 @@ fn bench(criterion: &mut Criterion) {
     });
 }
 
-criterion_group!(benches, bench);
+criterion_group!(
+    name = benches;
+    config = Criterion::default().sample_size(2);
+    targets = bench
+);
 criterion_main!(benches);

ci/buildkite-snap.yml

@@ -1,4 +1,4 @@
 steps:
   - command: "ci/snap.sh"
-    timeout_in_minutes: 20
+    timeout_in_minutes: 40
     name: "snap [public]"

ci/buildkite.yml

@@ -1,4 +1,38 @@
 steps:
+  - command: "ci/docker-run.sh rust ci/test-stable.sh"
+    name: "stable [public]"
+    env:
+      CARGO_TARGET_CACHE_NAME: "stable"
+    timeout_in_minutes: 30
+  - command: "ci/shellcheck.sh"
+    name: "shellcheck [public]"
+    timeout_in_minutes: 20
+  - command: "ci/docker-run.sh solanalabs/rust-nightly ci/test-nightly.sh"
+    name: "nightly [public]"
+    env:
+      CARGO_TARGET_CACHE_NAME: "nightly"
+    timeout_in_minutes: 30
+  - command: "ci/test-stable-perf.sh"
+    name: "stable-perf [public]"
+    env:
+      CARGO_TARGET_CACHE_NAME: "stable-perf"
+    timeout_in_minutes: 20
+    retry:
+      automatic:
+      - exit_status: "*"
+        limit: 2
+    agents:
+      - "queue=cuda"
+  - command: "ci/pr-snap.sh"
+    timeout_in_minutes: 20
+    name: "snap [public]"
+  - wait
+  - command: "ci/publish-crate.sh"
+    timeout_in_minutes: 20
+    name: "publish crate [public]"
+  - command: "ci/hoover.sh"
+    timeout_in_minutes: 20
+    name: "clean agent [public]"
   - trigger: "solana-snap"
     branches: "!pull/*"
     async: true
@@ -7,5 +41,4 @@ steps:
       commit: "${BUILDKITE_COMMIT}"
       branch: "${BUILDKITE_BRANCH}"
       env:
-        BUILDKITE_TAG: "${BUILDKITE_TAG}"
-        PARENT_BUILDKITE_TAG: "${BUILDKITE_TAG}"
+        TRIGGERED_BUILDKITE_TAG: "${BUILDKITE_TAG}"

ci/docker-rust-nightly/Dockerfile

@@ -1,3 +1,6 @@
 FROM rustlang/rust:nightly
 
-RUN cargo install --force clippy cargo-cov
+RUN cargo install --force clippy cargo-cov && \
+    echo deb http://ftp.debian.org/debian stretch-backports main >> /etc/apt/sources.list && \
+    apt update && \
+    apt install -y llvm-6.0

ci/docker-snapcraft/Dockerfile

@@ -2,6 +2,6 @@ FROM snapcraft/xenial-amd64
 
 # Update snapcraft to latest version
 RUN apt-get update -qq \
- && apt-get install -y snapcraft \
+ && apt-get install -y snapcraft daemontools \
  && rm -rf /var/lib/apt/lists/* \
  && snapcraft --version

ci/refresh-testnet.sh

@@ -1,68 +0,0 @@
-#!/bin/bash
-#
-# Refreshes the Solana software running on the Testnet full nodes
-#
-# This script must be run by a user/machine that has successfully authenticated
-# with GCP and has sufficient permission.
-#
-
-if [[ -z $SOLANA_METRICS_CONFIG ]]; then
-  echo Error: SOLANA_METRICS_CONFIG environment variable is unset
-  exit 1
-fi
-
-# Default to --edge channel.  To select the beta channel:
-#   export SOLANA_METRICS_CONFIG=--beta
-if [[ -z $SOLANA_SNAP_CHANNEL ]]; then
-  SOLANA_SNAP_CHANNEL=--edge
-fi
-
-vmlist=(testnet-solana-com:us-west1-b) # Leader is hard coded as the first entry
-
-echo "--- Available validators"
-gcloud compute instances list --filter="labels.testnet-mode=validator"
-while read -r vmName vmZone status; do
-  if [[ $status != RUNNING ]]; then
-    echo "Warning: $vmName is not RUNNING, ignoring it."
-    continue
-  fi
-  vmlist+=("$vmName:$vmZone")
-done < <(gcloud compute instances list --filter="labels.testnet-mode=validator" --format 'value(name,zone,status)')
-
-
-echo "--- Refreshing"
-mode=leader+drone
-for info in "${vmlist[@]}"; do
-  vmName=${info%:*}
-  vmZone=${info#*:}
-  echo "Starting refresh for $vmName"
-
-  (
-    echo "--- Processing $vmName in zone $vmZone as $mode"
-    cat > "autogen-refresh-$vmName.sh" <<EOF
-      set -x
-      sudo snap remove solana
-      sudo snap install solana $SOLANA_SNAP_CHANNEL --devmode
-      sudo snap set solana mode=$mode metrics-config=$SOLANA_METRICS_CONFIG
-      snap info solana
-      sudo snap logs solana -n200
-EOF
-    set -x
-    gcloud compute scp --zone "$vmZone" "autogen-refresh-$vmName.sh" "$vmName":
-    gcloud compute ssh "$vmName" --zone "$vmZone" \
-      --ssh-flag="-o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null -t" \
-      --command="bash ./autogen-refresh-$vmName.sh"
-  ) > "log-$vmName.txt" 2>&1 &
-  mode=validator
-done
-
-echo "Waiting..."
-wait
-
-for info in "${vmlist[@]}"; do
-  vmName=${info%:*}
-  cat "log-$vmName.txt"
-done
-
-echo "--- done"
-exit 0

ci/snap.sh

@@ -7,7 +7,11 @@ if [[ -z $BUILDKITE_BRANCH ]] || ./ci/is-pr.sh; then
   DRYRUN="echo"
 fi
 
-if [[ -z "$BUILDKITE_TAG" ]]; then
+# BUILDKITE_TAG is the normal environment variable set by Buildkite.  However
+# when this script is run from a triggered pipeline, TRIGGERED_BUILDKITE_TAG is
+# used instead of BUILDKITE_TAG (due to Buildkite limitations that prevents
+# BUILDKITE_TAG from propagating through to triggered pipelines)
+if [[ -z "$BUILDKITE_TAG" && -z "$TRIGGERED_BUILDKITE_TAG" ]]; then
   SNAP_CHANNEL=edge
 else
   SNAP_CHANNEL=beta
@@ -33,6 +37,12 @@ fi
 
 set -x
 
+echo --- checking for multilog
+if [[ ! -x /usr/bin/multilog ]]; then
+  echo "multilog not found, install with: sudo apt-get install -y daemontools"
+  exit 1
+fi
+
 echo --- build
 snapcraft
 

ci/test-nightly.sh

@@ -27,6 +27,6 @@ ls -l target/cov/report/index.html
 if [[ -z "$CODECOV_TOKEN" ]]; then
   echo CODECOV_TOKEN undefined
 else
-  bash <(curl -s https://codecov.io/bash) -x 'llvm-cov gcov'
+  bash <(curl -s https://codecov.io/bash) -x 'llvm-cov-6.0 gcov'
 fi
 

ci/testnet-deploy.sh

@@ -0,0 +1,165 @@
+#!/bin/bash -e
+#
+# Deploys the Solana software running on the testnet full nodes
+#
+# This script must be run by a user/machine that has successfully authenticated
+# with GCP and has sufficient permission.
+#
+cd "$(dirname "$0")/.."
+
+# TODO: Switch over to rolling updates
+ROLLING_UPDATE=false
+#ROLLING_UPDATE=true
+
+if [[ -z $SOLANA_METRICS_CONFIG ]]; then
+  echo Error: SOLANA_METRICS_CONFIG environment variable is unset
+  exit 1
+fi
+
+# Default to edge channel.  To select the beta channel:
+#   export SOLANA_SNAP_CHANNEL=beta
+if [[ -z $SOLANA_SNAP_CHANNEL ]]; then
+  SOLANA_SNAP_CHANNEL=edge
+fi
+
+case $SOLANA_SNAP_CHANNEL in
+edge)
+  publicUrl=master.testnet.solana.com
+  publicIp=$(dig +short $publicUrl | head -n1)
+  ;;
+beta)
+  publicUrl=testnet.solana.com
+  publicIp="" # Use default value
+  ;;
+*)
+  echo Error: Unknown SOLANA_SNAP_CHANNEL=$SOLANA_SNAP_CHANNEL
+  exit 1
+  ;;
+esac
+
+resourcePrefix=${publicUrl//./-}
+vmlist=("$resourcePrefix":us-west1-b) # Leader is hard coded as the first entry
+validatorNamePrefix=$resourcePrefix-validator-
+
+echo "--- Available validators for $publicUrl"
+filter="name~^$validatorNamePrefix"
+gcloud compute instances list --filter="$filter"
+while read -r vmName vmZone status; do
+  if [[ $status != RUNNING ]]; then
+    echo "Warning: $vmName is not RUNNING, ignoring it."
+    continue
+  fi
+  vmlist+=("$vmName:$vmZone")
+done < <(gcloud compute instances list --filter="$filter" --format 'value(name,zone,status)')
+
+wait_for_node() {
+  declare pid=$1
+
+  declare ok=true
+  wait "$pid" || ok=false
+  cat "log-$pid.txt"
+  if ! $ok; then
+    echo ^^^ +++
+    exit 1
+  fi
+}
+
+if ! $ROLLING_UPDATE; then
+  count=1
+  for info in "${vmlist[@]}"; do
+    nodePosition="($count/${#vmlist[*]})"
+    vmName=${info%:*}
+    vmZone=${info#*:}
+    echo "--- Shutting down $vmName in zone $vmZone $nodePosition"
+    gcloud compute ssh "$vmName" --zone "$vmZone" \
+      --ssh-flag="-o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null" \
+      --command="echo sudo snap remove solana" &
+
+    if [[ $((count % 10)) = 0 ]]; then
+      #  Slow down deployment to avoid triggering GCP login
+      #  quota limits (each |ssh| counts as a login)
+      sleep 3
+    fi
+
+    count=$((count + 1))
+  done
+
+  wait
+fi
+
+echo "--- Refreshing leader for $publicUrl"
+leader=true
+pids=()
+count=1
+for info in "${vmlist[@]}"; do
+  nodePosition="($count/${#vmlist[*]})"
+
+  vmName=${info%:*}
+  vmZone=${info#*:}
+  echo "Starting refresh for $vmName $nodePosition"
+
+  (
+    SECONDS=0
+    echo "--- $vmName in zone $vmZone $nodePosition"
+    commonNodeConfig="\
+      rust-log=$RUST_LOG \
+      default-metrics-rate=$SOLANA_DEFAULT_METRICS_RATE \
+      metrics-config=$SOLANA_METRICS_CONFIG \
+    "
+    if $leader; then
+      nodeConfig="mode=leader+drone $commonNodeConfig"
+      if [[ -n $SOLANA_CUDA ]]; then
+        nodeConfig="$nodeConfig enable-cuda=1"
+      fi
+    else
+      nodeConfig="mode=validator leader-address=$publicIp $commonNodeConfig"
+    fi
+
+    set -x
+    gcloud compute ssh "$vmName" --zone "$vmZone" \
+      --ssh-flag="-o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null -t" \
+      --command="\
+        set -ex; \
+        logmarker='solana deploy $(date)/$RANDOM'; \
+        sudo snap remove solana; \
+        logger \$logmarker; \
+        sudo snap install solana --$SOLANA_SNAP_CHANNEL --devmode; \
+        sudo snap set solana $nodeConfig; \
+        snap info solana; \
+        echo Slight delay to get more syslog output; \
+        sleep 2; \
+        sudo grep -Pzo \"\$logmarker(.|\\n)*\" /var/log/syslog \
+      "
+    echo "Succeeded in ${SECONDS} seconds"
+  ) > "log-$vmName.txt" 2>&1 &
+  pid=$!
+  # Rename log file so it can be discovered later by $pid
+  mv "log-$vmName.txt" "log-$pid.txt"
+
+  if $leader; then
+    echo Waiting for leader...
+    # Wait for the leader to initialize before starting the validators
+    # TODO: Remove this limitation eventually.
+    wait_for_node "$pid"
+
+    echo "--- Refreshing validators"
+  else
+    #  Slow down deployment to ~20 machines a minute to avoid triggering GCP login
+    #  quota limits (each |ssh| counts as a login)
+    sleep 3
+
+    pids+=("$pid")
+  fi
+  leader=false
+  count=$((count + 1))
+done
+
+echo --- Waiting for validators
+for pid in "${pids[@]}"; do
+  wait_for_node "$pid"
+done
+
+echo "--- $publicUrl sanity test"
+USE_SNAP=1 ci/testnet-sanity.sh $publicUrl
+
+exit 0

ci/testnet-sanity.sh

@@ -0,0 +1,17 @@
+#!/bin/bash -e
+#
+# Perform a quick sanity test on the specific testnet
+#
+
+cd "$(dirname "$0")/.."
+
+TESTNET=$1
+if [[ -z $TESTNET ]]; then
+  TESTNET=testnet.solana.com
+fi
+
+echo "--- $TESTNET: wallet sanity"
+multinode-demo/test/wallet-sanity.sh $TESTNET
+
+echo --- fin
+exit 0

multinode-demo/client.sh

@@ -7,25 +7,27 @@ here=$(dirname "$0")
 # shellcheck source=multinode-demo/common.sh
 source "$here"/common.sh
 
-leader=${1:-${here}/..}  # Default to local solana repo
+leader=$1
+if [[ -z $leader ]]; then
+  if [[ -d "$SNAP" ]]; then
+    leader=testnet.solana.com # Default to testnet when running as a Snap
+  else
+    leader=$here/.. # Default to local solana repo
+  fi
+fi
 count=${2:-1}
 
 rsync_leader_url=$(rsync_url "$leader")
 
 set -ex
 mkdir -p "$SOLANA_CONFIG_CLIENT_DIR"
-if [[ ! -r "$SOLANA_CONFIG_CLIENT_DIR"/leader.json ]]; then
-  (
-    set -x
-    $rsync -vPz "$rsync_leader_url"/config/leader.json "$SOLANA_CONFIG_CLIENT_DIR"/
-  )
-fi
+$rsync -vPz "$rsync_leader_url"/config/leader.json "$SOLANA_CONFIG_CLIENT_DIR"/
 
 client_json="$SOLANA_CONFIG_CLIENT_DIR"/client.json
-if [[ ! -r $client_json ]]; then
-  $solana_mint <<<0 > "$client_json"
-fi
+[[ -r $client_json ]] || $solana_keygen -o "$client_json"
+
+$solana_client_demo \
+  -n "$count" \
+  -l "$SOLANA_CONFIG_CLIENT_DIR"/leader.json \
+  -k "$SOLANA_CONFIG_CLIENT_DIR"/client.json \
 
-# shellcheck disable=SC2086 # $solana_client_demo should not be quoted
-exec $solana_client_demo \
-  -n "$count" -l "$SOLANA_CONFIG_CLIENT_DIR"/leader.json -m "$SOLANA_CONFIG_CLIENT_DIR"/client.json

multinode-demo/common.sh

@@ -3,21 +3,41 @@
 # Disable complaints about unused variables in this file:
 # shellcheck disable=2034
 
+# shellcheck disable=2154 # 'here' is referenced but not assigned
+if [[ -z $here ]]; then
+  echo "|here| is not defined"
+  exit 1
+fi
+
 rsync=rsync
+leader_logger="cat"
+validator_logger="cat"
+drone_logger="cat"
+
 if [[ -d "$SNAP" ]]; then # Running inside a Linux Snap?
   solana_program() {
     declare program="$1"
-    if [[ "$program" = wallet ]]; then
-      # TODO: Merge wallet.sh functionality into solana-wallet proper and
-      #       remove this special case
+    if [[ "$program" = wallet || "$program" = client-demo ]]; then
+      # TODO: Merge wallet.sh/client.sh functionality into
+      #       solana-wallet/solana-demo-client proper and remove this special case
       printf "%s/bin/solana-%s" "$SNAP" "$program"
     else
       printf "%s/command-%s.wrapper" "$SNAP" "$program"
     fi
   }
   rsync="$SNAP"/bin/rsync
+  multilog="$SNAP/bin/multilog t s16777215"
+  leader_logger="$multilog $SNAP_DATA/leader"
+  validator_logger="$multilog t $SNAP_DATA/validator"
+  drone_logger="$multilog $SNAP_DATA/drone"
+  # Create log directories manually to prevent multilog from creating them as
+  # 0700
+  mkdir -p "$SNAP_DATA"/{drone,leader,validator}
+
   SOLANA_METRICS_CONFIG="$(snapctl get metrics-config)"
+  SOLANA_DEFAULT_METRICS_RATE="$(snapctl get default-metrics-rate)"
   SOLANA_CUDA="$(snapctl get enable-cuda)"
+  RUST_LOG="$(snapctl get rust-log)"
 
 elif [[ -n "$USE_SNAP" ]]; then # Use the Linux Snap binaries
   solana_program() {
@@ -37,13 +57,18 @@ else
     declare features=""
     if [[ "$program" =~ ^(.*)-cuda$ ]]; then
       program=${BASH_REMATCH[1]}
-      features="--features=cuda,erasure"
+      features="--features=cuda"
     fi
     if [[ -z "$DEBUG" ]]; then
       maybe_release=--release
     fi
     printf "cargo run $maybe_release --bin solana-%s %s -- " "$program" "$features"
   }
+  if [[ -n $SOLANA_CUDA ]]; then
+    # Locate perf libs downloaded by |./fetch-perf-libs.sh|
+    LD_LIBRARY_PATH=$(cd "$here" && dirname "$PWD"):$LD_LIBRARY_PATH
+    export LD_LIBRARY_PATH
+  fi
 fi
 
 solana_client_demo=$(solana_program client-demo)
@@ -53,7 +78,7 @@ solana_fullnode=$(solana_program fullnode)
 solana_fullnode_config=$(solana_program fullnode-config)
 solana_fullnode_cuda=$(solana_program fullnode-cuda)
 solana_genesis=$(solana_program genesis)
-solana_mint=$(solana_program mint)
+solana_keygen=$(solana_program keygen)
 
 export RUST_LOG=${RUST_LOG:-solana=info} # if RUST_LOG is unset, default to info
 export RUST_BACKTRACE=1
@@ -73,29 +98,29 @@ configure_metrics() {
   for param in "${metrics_params[@]}"; do
     IFS='=' read -r -a pair <<< "$param"
     if [[ "${#pair[@]}" != 2 ]]; then
-      echo Error: invalid metrics parameter: "$param"
+      echo Error: invalid metrics parameter: "$param" >&2
     else
       declare name="${pair[0]}"
       declare value="${pair[1]}"
       case "$name" in
       host)
         export INFLUX_HOST="$value"
-        echo INFLUX_HOST="$INFLUX_HOST"
+        echo INFLUX_HOST="$INFLUX_HOST" >&2
         ;;
       db)
         export INFLUX_DATABASE="$value"
-        echo INFLUX_DATABASE="$INFLUX_DATABASE"
+        echo INFLUX_DATABASE="$INFLUX_DATABASE" >&2
         ;;
       u)
         export INFLUX_USERNAME="$value"
-        echo INFLUX_USERNAME="$INFLUX_USERNAME"
+        echo INFLUX_USERNAME="$INFLUX_USERNAME" >&2
         ;;
       p)
         export INFLUX_PASSWORD="$value"
-        echo INFLUX_PASSWORD="********"
+        echo INFLUX_PASSWORD="********" >&2
         ;;
       *)
-        echo Error: Unknown metrics parameter name: "$name"
+        echo Error: Unknown metrics parameter name: "$name" >&2
         ;;
       esac
     fi
@@ -105,17 +130,23 @@ configure_metrics
 
 tune_networking() {
   # Reference: https://medium.com/@CameronSparr/increase-os-udp-buffers-to-improve-performance-51d167bb1360
-  [[ $(uname) = Linux ]] && (
-    set -x
-    # TODO: Check values and warn instead, it's a little rude to set them here.
-    sudo sysctl -w net.core.rmem_max=26214400 1>/dev/null 2>/dev/null
-    sudo sysctl -w net.core.rmem_default=26214400 1>/dev/null 2>/dev/null
-  )
+  if [[ $(uname) = Linux ]]; then
+    (
+      set -x +e
+      # test the existence of the sysctls before trying to set them
+      # go ahead and return true and don't exit if these calls fail
+      sysctl net.core.rmem_max 2>/dev/null 1>/dev/null &&
+          sudo sysctl -w net.core.rmem_max=26214400 1>/dev/null 2>/dev/null
+
+      sysctl net.core.rmem_default 2>/dev/null 1>/dev/null &&
+          sudo sysctl -w net.core.rmem_default=26214400 1>/dev/null 2>/dev/null
+    ) || true
+  fi
 }
 
 SOLANA_CONFIG_DIR=${SNAP_DATA:-$PWD}/config
 SOLANA_CONFIG_PRIVATE_DIR=${SNAP_DATA:-$PWD}/config-private
-SOLANA_CONFIG_CLIENT_DIR=${SNAP_USER_DATA:-$PWD}/config-client-client
+SOLANA_CONFIG_CLIENT_DIR=${SNAP_USER_DATA:-$PWD}/config-client
 
 rsync_url() { # adds the 'rsync://` prefix to URLs that need it
   declare url="$1"

multinode-demo/drone.sh

@@ -36,6 +36,7 @@ set -ex
 mkdir -p "$SOLANA_CONFIG_DIR"
 $rsync -vPz "$rsync_leader_url"/config/leader.json "$SOLANA_CONFIG_DIR"/
 
-# shellcheck disable=SC2086 # $solana_drone should not be quoted
-exec $solana_drone \
-  -l "$SOLANA_CONFIG_DIR"/leader.json -m "$SOLANA_CONFIG_PRIVATE_DIR"/mint.json
+set -o pipefail
+$solana_drone \
+  -l "$SOLANA_CONFIG_DIR"/leader.json -k "$SOLANA_CONFIG_PRIVATE_DIR"/mint.json \
+2>&1 | $drone_logger

multinode-demo/leader.sh

@@ -25,9 +25,8 @@ fi
 
 tune_networking
 
-# shellcheck disable=SC2086 # $program should not be quoted
-exec $program \
-  -l "$SOLANA_CONFIG_DIR"/leader.json \
-  < <(shopt -s nullglob && cat "$SOLANA_CONFIG_DIR"/genesis.log \
-          "$SOLANA_CONFIG_DIR"/tx-*.log) \
-  > "$SOLANA_CONFIG_DIR"/tx-"$(date -u +%Y%m%d%H%M%S%N)".log
+set -xo pipefail
+$program \
+  --identity "$SOLANA_CONFIG_DIR"/leader.json \
+  --ledger "$SOLANA_CONFIG_DIR"/ledger.log \
+2>&1 | $leader_logger

multinode-demo/myip.sh

@@ -1,59 +0,0 @@
-#!/bin/bash
-
-function myip()
-{
-  # shellcheck disable=SC2207
-  declare ipaddrs=(
-    # query interwebs
-    $(curl -s ifconfig.co)
-    # machine's interfaces
-    $(ifconfig |
-          awk '/inet addr:/ {gsub("addr:","",$2); print $2; next}
-               /inet6 addr:/ {gsub("/.*", "", $3); print $3; next}
-               /inet(6)? / {print $2}'
-    )
-  )
-
-  if (( ! ${#ipaddrs[*]} ))
-  then
-      echo "
-myip: error: I'm having trouble determining what our IP address is...
-  Are we connected to a network?
-
-"
-      return 1
-  fi
-
-
-  declare prompt="
-Please choose the IP address you want to advertise to the network:
-
-   0) ${ipaddrs[0]} <====== this one was returned by the interwebs...
-"
-
-  for ((i=1; i < ${#ipaddrs[*]}; i++))
-  do
-    prompt+="   $i) ${ipaddrs[i]}
-"
-  done
-
-  while read -r -p "${prompt}
-please enter a number [0 for default]: " which
-  do
-    [[ -z ${which} ]] && break;
-    [[ ${which} =~ [0-9]+ ]] && (( which < ${#ipaddrs[*]} )) && break;
-    echo "Ug. invalid entry \"${which}\"...
-"
-    sleep 1
-  done
-
-  which=${which:-0}
-
-  echo "${ipaddrs[which]}"
-
-}
-
-if [[ ${0} == "${BASH_SOURCE[0]}" ]]
-then
-    myip "$@"
-fi

multinode-demo/remote_leader.sh

@@ -0,0 +1,14 @@
+#!/bin/bash -e
+
+[[ -n $FORCE ]] || exit
+
+chmod 600 ~/.ssh/authorized_keys ~/.ssh/id_rsa
+
+PATH="$HOME"/.cargo/bin:"$PATH"
+
+./fetch-perf-libs.sh
+
+# Run setup
+USE_INSTALL=1 ./multinode-demo/setup.sh -p
+USE_INSTALL=1 SOLANA_CUDA=1 ./multinode-demo/leader.sh >leader.log 2>&1 &
+USE_INSTALL=1 ./multinode-demo/drone.sh >drone.log 2>&1 &

multinode-demo/remote_nodes.sh

@@ -0,0 +1,185 @@
+#!/bin/bash
+
+command=$1
+ip_addr_file=
+remote_user=
+ssh_keys=
+
+shift
+
+usage() {
+  exitcode=0
+  if [[ -n "$1" ]]; then
+    exitcode=1
+    echo "Error: $*"
+  fi
+  cat <<EOF
+usage: $0 <start|stop> <-f IP Addr Array file> <-u username> [-k ssh-keys]
+
+Manage a GCE multinode network
+
+ start|stop    - Create or delete the network
+ -f file       - A bash script that exports an array of IP addresses, ip_addr_array.
+                 Elements of the array are public IP address of remote nodes.
+ -u username   - The username for logging into remote nodes.
+ -k ssh-keys   - Path to public/private key pair that remote nodes can use to perform
+                 rsync and ssh among themselves. Must contain pub, and priv keys.
+
+EOF
+  exit $exitcode
+}
+
+while getopts "h?f:u:k:" opt; do
+  case $opt in
+  h | \?)
+    usage
+    ;;
+  f)
+    ip_addr_file=$OPTARG
+    ;;
+  u)
+    remote_user=$OPTARG
+    ;;
+  k)
+    ssh_keys=$OPTARG
+    ;;
+  *)
+    usage "Error: unhandled option: $opt"
+    ;;
+  esac
+done
+
+set -e
+
+# Sample IP Address array file contents
+# ip_addr_array=(192.168.1.1 192.168.1.5 192.168.2.2)
+
+[[ -n $command ]] || usage "Need a command (start|stop)"
+[[ -n $ip_addr_file ]] || usage "Need a file with IP address array"
+[[ -n $remote_user ]] || usage "Need the username for remote nodes"
+
+ip_addr_array=()
+# Get IP address array
+# shellcheck source=/dev/null
+source "$ip_addr_file"
+
+build_project() {
+  echo "Build started at $(date)"
+  SECONDS=0
+
+  # Build and install locally
+  PATH="$HOME"/.cargo/bin:"$PATH"
+  cargo install --force
+
+  echo "Build took $SECONDS seconds"
+}
+
+common_start_setup() {
+  ip_addr=$1
+
+  # Killing sshguard for now. TODO: Find a better solution
+  # sshguard is blacklisting IP address after ssh-keyscan and ssh login attempts
+  ssh "$remote_user@$ip_addr" " \
+    set -ex; \
+    sudo service sshguard stop; \
+    sudo apt-get --assume-yes install rsync libssl-dev; \
+    mkdir -p ~/.ssh ~/solana ~/.cargo/bin; \
+  " >log/"$ip_addr".log
+
+  # If provided, deploy SSH keys
+  if [[ -n $ssh_keys ]]; then
+    {
+      rsync -vPrz "$ssh_keys"/id_rsa "$remote_user@$ip_addr":~/.ssh/
+      rsync -vPrz "$ssh_keys"/id_rsa.pub "$remote_user@$ip_addr":~/.ssh/
+      rsync -vPrz "$ssh_keys"/id_rsa.pub "$remote_user@$ip_addr":~/.ssh/authorized_keys
+      rsync -vPrz ./multinode-demo "$remote_user@$ip_addr":~/solana/
+    } >>log/"$ip_addr".log
+  fi
+}
+
+start_leader() {
+  common_start_setup "$1"
+
+  {
+    rsync -vPrz ~/.cargo/bin/solana* "$remote_user@$ip_addr":~/.cargo/bin/
+    rsync -vPrz ./fetch-perf-libs.sh "$remote_user@$ip_addr":~/solana/
+    ssh -n -f "$remote_user@$ip_addr" 'cd solana; FORCE=1 ./multinode-demo/remote_leader.sh'
+  } >>log/"$1".log
+
+  leader_ip=$1
+  leader_time=$SECONDS
+  SECONDS=0
+}
+
+start_validator() {
+  common_start_setup "$1"
+
+  ssh -n -f "$remote_user@$ip_addr" "cd solana; FORCE=1 ./multinode-demo/remote_validator.sh $leader_ip" >>log/"$1".log
+}
+
+start_all_nodes() {
+  echo "Deployment started at $(date)"
+  SECONDS=0
+  count=0
+  leader_ip=
+  leader_time=
+
+  mkdir -p log
+
+  for ip_addr in "${ip_addr_array[@]}"; do
+    if ((!count)); then
+      # Start the leader on the first node
+      echo "Leader node $ip_addr, killing previous instance and restarting"
+      start_leader "$ip_addr"
+    else
+      # Start validator on all other nodes
+      echo "Validator[$count] node $ip_addr, killing previous instance and restarting"
+      start_validator "$ip_addr" &
+      # TBD: Remove the sleep or reduce time once GCP login quota is increased
+      sleep 2
+    fi
+
+    ((count = count + 1))
+  done
+
+  wait
+
+  ((validator_count = count - 1))
+
+  echo "Deployment finished at $(date)"
+  echo "Leader deployment too $leader_time seconds"
+  echo "$validator_count Validator deployment took $SECONDS seconds"
+}
+
+stop_all_nodes() {
+  SECONDS=0
+  local count=0
+  for ip_addr in "${ip_addr_array[@]}"; do
+    ssh-keygen -R "$ip_addr" >log/local.log
+    ssh-keyscan "$ip_addr" >>~/.ssh/known_hosts 2>/dev/null
+
+    echo "Stopping node[$count] $ip_addr. Remote user $remote_user"
+
+    ssh -n -f "$remote_user@$ip_addr" " \
+    set -ex; \
+    sudo service sshguard stop; \
+    pkill -9 solana-; \
+    pkill -9 validator; \
+    pkill -9 leader; \
+    "
+    sleep 2
+    ((count = count + 1))
+    echo "Stopped node[$count] $ip_addr"
+  done
+  echo "Stopping $count nodes took $SECONDS seconds"
+}
+
+if [[ $command == "start" ]]; then
+  build_project
+  stop_all_nodes
+  start_all_nodes
+elif [[ $command == "stop" ]]; then
+  stop_all_nodes
+else
+  usage "Unknown command: $command"
+fi

multinode-demo/remote_validator.sh

@@ -0,0 +1,17 @@
+#!/bin/bash -e
+
+[[ -n $FORCE ]] || exit
+
+chmod 600 ~/.ssh/authorized_keys ~/.ssh/id_rsa
+
+PATH="$HOME"/.cargo/bin:"$PATH"
+
+touch ~/.ssh/known_hosts
+ssh-keygen -R "$1" 2>/dev/null
+ssh-keyscan "$1" >>~/.ssh/known_hosts 2>/dev/null
+
+rsync -vPrz "$1":~/.cargo/bin/solana* ~/.cargo/bin/
+
+# Run setup
+USE_INSTALL=1 ./multinode-demo/setup.sh -p
+USE_INSTALL=1 ./multinode-demo/validator.sh "$1":~/solana "$1" >validator.log 2>&1

multinode-demo/setup.sh

@@ -71,6 +71,9 @@ done
 
 leader_address_args=("$ip_address_arg")
 validator_address_args=("$ip_address_arg" -b 9000)
+leader_id_path="$SOLANA_CONFIG_PRIVATE_DIR"/leader-id.json
+validator_id_path="$SOLANA_CONFIG_PRIVATE_DIR"/validator-id.json
+mint_path="$SOLANA_CONFIG_PRIVATE_DIR"/mint.json
 
 set -e
 
@@ -78,25 +81,27 @@ echo "Cleaning $SOLANA_CONFIG_DIR"
 rm -rvf "$SOLANA_CONFIG_DIR"
 mkdir -p "$SOLANA_CONFIG_DIR"
 
+rm -rvf "$SOLANA_CONFIG_PRIVATE_DIR"
+mkdir -p "$SOLANA_CONFIG_PRIVATE_DIR"
+
+$solana_keygen -o "$leader_id_path"
+$solana_keygen -o "$validator_id_path"
 
 if $node_type_leader; then
-  rm -rvf "$SOLANA_CONFIG_PRIVATE_DIR"
-  mkdir -p "$SOLANA_CONFIG_PRIVATE_DIR"
+  echo "Creating $mint_path with $num_tokens tokens"
+  $solana_keygen -o "$mint_path"
 
-  echo "Creating $SOLANA_CONFIG_DIR/mint.json with $num_tokens tokens"
-  $solana_mint <<<"$num_tokens" > "$SOLANA_CONFIG_PRIVATE_DIR"/mint.json
-
-  echo "Creating $SOLANA_CONFIG_DIR/genesis.log"
-  $solana_genesis < "$SOLANA_CONFIG_PRIVATE_DIR"/mint.json > "$SOLANA_CONFIG_DIR"/genesis.log
+  echo "Creating $SOLANA_CONFIG_DIR/ledger.log"
+  $solana_genesis --tokens="$num_tokens" < "$mint_path" > "$SOLANA_CONFIG_DIR"/ledger.log
 
   echo "Creating $SOLANA_CONFIG_DIR/leader.json"
-  $solana_fullnode_config "${leader_address_args[@]}" > "$SOLANA_CONFIG_DIR"/leader.json
+  $solana_fullnode_config --keypair="$leader_id_path" "${leader_address_args[@]}" > "$SOLANA_CONFIG_DIR"/leader.json
 fi
 
 
 if $node_type_validator; then
   echo "Creating $SOLANA_CONFIG_DIR/validator.json"
-  $solana_fullnode_config "${validator_address_args[@]}" > "$SOLANA_CONFIG_DIR"/validator.json
+  $solana_fullnode_config --keypair="$validator_id_path" "${validator_address_args[@]}" > "$SOLANA_CONFIG_DIR"/validator.json
 fi
 
 ls -lh "$SOLANA_CONFIG_DIR"/

multinode-demo/start_nodes.sh

@@ -1,107 +0,0 @@
-#!/bin/bash
-
-ip_addr_file=$1
-remote_user=$2
-ssh_keys=$3
-
-usage() {
-  echo -e "\\tUsage: $0 <IP Address array> <username> [path to ssh keys]\\n"
-  echo -e "\\t <IP Address array>: A bash script that exports an array of IP addresses, ip_addr_array. Elements of the array are public IP address of remote nodes."
-  echo -e "\\t <username>:         The username for logging into remote nodes."
-  echo -e "\\t [path to ssh keys]: The public/private key pair that remote nodes can use to perform rsync and ssh among themselves. Must contain pub, priv and authorized_keys.\\n"
-  exit 1
-}
-
-# Sample IP Address array file contents
-# ip_addr_array=(192.168.1.1 192.168.1.5 192.168.2.2)
-
-if [[ -z "$ip_addr_file" ]]; then
-  usage
-fi
-
-if [[ -z "$remote_user" ]]; then
-  usage
-fi
-
-echo "Build started at $(date)"
-SECONDS=0
-# Build and install locally
-PATH="$HOME"/.cargo/bin:"$PATH"
-cargo install --force
-
-echo "Build took $SECONDS seconds"
-
-ip_addr_array=()
-# Get IP address array
-# shellcheck source=/dev/null
-source "$ip_addr_file"
-
-# shellcheck disable=SC2089,SC2016
-ssh_command_prefix='export PATH="$HOME/.cargo/bin:$PATH"; cd solana; USE_INSTALL=1'
-
-echo "Deployment started at $(date)"
-SECONDS=0
-count=0
-leader=
-for ip_addr in "${ip_addr_array[@]}"; do
-  echo "$ip_addr"
-
-  ssh-keygen -R "$ip_addr"
-  ssh-keyscan "$ip_addr" >>~/.ssh/known_hosts
-
-  ssh -n -f "$remote_user@$ip_addr" 'mkdir -p ~/.ssh ~/solana ~/.cargo/bin'
-
-  # Killing sshguard for now. TODO: Find a better solution
-  # sshguard is blacklisting IP address after ssh-keyscan and ssh login attempts
-  ssh -n -f "$remote_user@$ip_addr" "sudo service sshguard stop"
-  ssh -n -f "$remote_user@$ip_addr" 'sudo apt-get --assume-yes install rsync libssl-dev'
-
-  # If provided, deploy SSH keys
-  if [[ -z $ssh_keys ]]; then
-    echo "skip copying the ssh keys"
-  else
-    rsync -vPrz "$ssh_keys"/id_rsa "$remote_user@$ip_addr":~/.ssh/
-    rsync -vPrz "$ssh_keys"/id_rsa.pub "$remote_user@$ip_addr":~/.ssh/
-    rsync -vPrz "$ssh_keys"/id_rsa.pub "$remote_user@$ip_addr":~/.ssh/authorized_keys
-    ssh -n -f "$remote_user@$ip_addr" 'chmod 600 ~/.ssh/authorized_keys ~/.ssh/id_rsa'
-  fi
-
-  # Stop current nodes
-  ssh "$remote_user@$ip_addr" 'pkill -9 solana-'
-
-  if [[ -n $leader ]]; then
-    echo "Adding known hosts for $ip_addr"
-    ssh -n -f "$remote_user@$ip_addr" "ssh-keygen -R $leader"
-    ssh -n -f "$remote_user@$ip_addr" "ssh-keyscan $leader >> ~/.ssh/known_hosts"
-
-    ssh -n -f "$remote_user@$ip_addr" "rsync -vPrz ""$remote_user@$leader"":~/.cargo/bin/solana* ~/.cargo/bin/"
-    ssh -n -f "$remote_user@$ip_addr" "rsync -vPrz ""$remote_user@$leader"":~/solana/multinode-demo ~/solana/"
-    ssh -n -f "$remote_user@$ip_addr" "rsync -vPrz ""$remote_user@$leader"":~/solana/fetch-perf-libs.sh ~/solana/"
-  else
-    # Deploy build and scripts to remote node
-    rsync -vPrz ~/.cargo/bin/solana* "$remote_user@$ip_addr":~/.cargo/bin/
-    rsync -vPrz ./multinode-demo "$remote_user@$ip_addr":~/solana/
-    rsync -vPrz ./fetch-perf-libs.sh "$remote_user@$ip_addr":~/solana/
-  fi
-
-  # Run setup
-  ssh "$remote_user@$ip_addr" "$ssh_command_prefix"' ./multinode-demo/setup.sh -p "$ip_addr"'
-
-  if ((!count)); then
-    # Start the leader on the first node
-    echo "Starting leader node $ip_addr"
-    ssh -n -f "$remote_user@$ip_addr" 'cd solana; ./fetch-perf-libs.sh'
-    ssh -n -f "$remote_user@$ip_addr" "$ssh_command_prefix"' SOLANA_CUDA=1 ./multinode-demo/leader.sh > leader.log 2>&1'
-    ssh -n -f "$remote_user@$ip_addr" "$ssh_command_prefix"' ./multinode-demo/drone.sh > drone.log 2>&1'
-    leader=${ip_addr_array[0]}
-  else
-    # Start validator on all other nodes
-    echo "Starting validator node $ip_addr"
-    ssh -n -f "$remote_user@$ip_addr" "$ssh_command_prefix"" ./multinode-demo/validator.sh $remote_user@$leader:~/solana $leader > validator.log 2>&1"
-  fi
-
-  ((count++))
-done
-
-echo "Deployment finished at $(date)"
-echo "Deployment took $SECONDS seconds"

multinode-demo/test/wallet-sanity.sh

@@ -6,7 +6,13 @@
 here=$(dirname "$0")
 cd "$here"
 
-wallet="../wallet.sh $1"
+if [[ -n "$USE_SNAP" ]]; then
+  # TODO: Merge wallet.sh functionality into solana-wallet proper and
+  #       remove this USE_SNAP case
+  wallet="solana.wallet $1"
+else
+  wallet="../wallet.sh $1"
+fi
 
 # Tokens transferred to this address are lost forever...
 garbage_address=vS3ngn1TfQmpsW1Z4NkLuqNAQFF3dYQw8UZ6TCx9bmq

multinode-demo/validator.sh

@@ -65,16 +65,32 @@ fi
 
 rsync_leader_url=$(rsync_url "$leader")
 
-set -ex
-SOLANA_LEADER_CONFIG_DIR="$SOLANA_CONFIG_DIR"/leader-config
-rm -rf "$SOLANA_LEADER_CONFIG_DIR"
-$rsync -vPrz "$rsync_leader_url"/config/ "$SOLANA_LEADER_CONFIG_DIR"
-ls -lh "$SOLANA_LEADER_CONFIG_DIR"
-
 tune_networking
 
-# shellcheck disable=SC2086 # $program should not be quoted
-exec $program \
-  -l "$SOLANA_CONFIG_DIR"/validator.json -t "$leader_address:$leader_port" \
-  < <(shopt -s nullglob && cat "$SOLANA_LEADER_CONFIG_DIR"/genesis.log \
-          "$SOLANA_LEADER_CONFIG_DIR"/tx-*.log)
+SOLANA_LEADER_CONFIG_DIR="$SOLANA_CONFIG_DIR"/leader-config
+rm -rf "$SOLANA_LEADER_CONFIG_DIR"
+set -ex
+$rsync -vPrz "$rsync_leader_url"/config/ "$SOLANA_LEADER_CONFIG_DIR"
+
+# migrate from old ledger format?  why not...
+if [[ ! -f "$SOLANA_LEADER_CONFIG_DIR"/ledger.log &&
+          -f "$SOLANA_LEADER_CONFIG_DIR"/genesis.log ]]; then
+  (shopt -s nullglob &&
+     cat "$SOLANA_LEADER_CONFIG_DIR"/genesis.log \
+         "$SOLANA_LEADER_CONFIG_DIR"/tx-*.log) > "$SOLANA_LEADER_CONFIG_DIR"/ledger.log
+fi
+
+# Ensure the validator has at least 1 token before connecting to the network
+# TODO: Remove this workaround
+while ! $solana_wallet \
+          -l "$SOLANA_LEADER_CONFIG_DIR"/leader.json \
+          -k "$SOLANA_CONFIG_PRIVATE_DIR"/validator-id.json airdrop --tokens 1; do
+  sleep 1
+done
+
+set -o pipefail
+$program \
+  --identity "$SOLANA_CONFIG_DIR"/validator.json \
+  --testnet "$leader_address:$leader_port" \
+  --ledger "$SOLANA_LEADER_CONFIG_DIR"/ledger.log \
+2>&1 | $validator_logger

multinode-demo/wallet.sh

@@ -30,18 +30,16 @@ rsync_leader_url=$(rsync_url "$leader")
 set -e
 mkdir -p "$SOLANA_CONFIG_CLIENT_DIR"
 if [[ ! -r "$SOLANA_CONFIG_CLIENT_DIR"/leader.json ]]; then
-  (
-    set -x
-    $rsync -vPz "$rsync_leader_url"/config/leader.json "$SOLANA_CONFIG_CLIENT_DIR"/
-  )
+  echo "Fetching leader configuration from $rsync_leader_url"
+  $rsync -Pz "$rsync_leader_url"/config/leader.json "$SOLANA_CONFIG_CLIENT_DIR"/
 fi
 
-client_json="$SOLANA_CONFIG_CLIENT_DIR"/client.json
-if [[ ! -r $client_json ]]; then
-  $solana_mint <<<0 > "$client_json"
+client_id_path="$SOLANA_CONFIG_CLIENT_DIR"/id.json
+if [[ ! -r $client_id_path ]]; then
+  echo "Generating client identity: $client_id_path"
+  $solana_keygen -o "$client_id_path"
 fi
 
-set -x
 # shellcheck disable=SC2086 # $solana_wallet should not be quoted
 exec $solana_wallet \
-  -l "$SOLANA_CONFIG_CLIENT_DIR"/leader.json -m "$client_json" "$@"
+  -l "$SOLANA_CONFIG_CLIENT_DIR"/leader.json -k "$client_id_path" "$@"

snap/snapcraft.yaml

@@ -37,40 +37,65 @@ apps:
     plugs:
       - network
       - network-bind
+      - home
   genesis:
     command: solana-genesis
-  mint:
-    command: solana-mint
+  keygen:
+    command: solana-keygen
+    plugs:
+      - home
   client-demo:
-    command: solana-client-demo
+    # TODO: Merge client.sh functionality into solana-client-demo proper
+    command: client.sh
+    #command: solana-client-demo
+    plugs:
+      - network
+      - network-bind
+      - home
   wallet:
     # TODO: Merge wallet.sh functionality into solana-wallet proper
     command: wallet.sh
     #command: solana-wallet
-
+    plugs:
+      - network
+      - home
   daemon-validator:
     daemon: simple
     command: validator.sh
-
+    plugs:
+      - network
+      - network-bind
   daemon-leader:
     daemon: simple
     command: leader.sh
-
+    plugs:
+      - network
+      - network-bind
   daemon-drone:
     daemon: simple
     command: drone.sh
+    plugs:
+      - network
+      - network-bind
 
 parts:
   solana:
     plugin: nil
     prime:
       - bin
-      - usr/lib/libgf_complete.so.1
-      - usr/lib/libJerasure.so.2
+      - usr/lib
     override-build: |
+      # Install CUDA 9.2 runtime
+      mkdir -p $SNAPCRAFT_PART_INSTALL/usr/
+      cp -rav /usr/local/cuda-9.2/targets/x86_64-linux/lib/ $SNAPCRAFT_PART_INSTALL/usr/lib
+      mkdir -p $SNAPCRAFT_PART_INSTALL/usr/lib/x86_64-linux-gnu/
+      cp -rav /usr/lib/x86_64-linux-gnu/libcuda.* $SNAPCRAFT_PART_INSTALL/usr/lib/x86_64-linux-gnu/
+      mkdir -p $SNAPCRAFT_PART_INSTALL/usr/lib/nvidia-396/
+      cp -v /usr/lib/nvidia-396/libnvidia-fatbinaryloader.so* $SNAPCRAFT_PART_INSTALL/usr/lib/nvidia-396/
+
       # Build/install solana-fullnode-cuda
       ./fetch-perf-libs.sh
-      cargo install --features=cuda,erasure --root $SNAPCRAFT_PART_INSTALL --bin solana-fullnode
+      cargo install --features=cuda --root $SNAPCRAFT_PART_INSTALL --bin solana-fullnode
       mv $SNAPCRAFT_PART_INSTALL/bin/solana-fullnode $SNAPCRAFT_PART_INSTALL
       rm -rf $SNAPCRAFT_PART_INSTALL/bin/*
       mv $SNAPCRAFT_PART_INSTALL/solana-fullnode $SNAPCRAFT_PART_INSTALL/bin/solana-fullnode-cuda
@@ -85,8 +110,9 @@ parts:
       mkdir -p $SNAPCRAFT_PART_INSTALL/bin
       cp -av multinode-demo/* $SNAPCRAFT_PART_INSTALL/bin/
 
-      # TODO: build rsync from source instead of sneaking it in from the host
+      # TODO: build rsync/multilog from source instead of sneaking it in from the host
       # system...
       set -x
       mkdir -p $SNAPCRAFT_PART_INSTALL/bin
       cp -av /usr/bin/rsync $SNAPCRAFT_PART_INSTALL/bin/
+      cp -av /usr/bin/multilog $SNAPCRAFT_PART_INSTALL/bin/

src/bank.rs

@@ -6,6 +6,7 @@
 extern crate libc;
 
 use chrono::prelude::*;
+use counter::Counter;
 use entry::Entry;
 use hash::Hash;
 use itertools::Itertools;
@@ -202,6 +203,11 @@ impl Bank {
         {
             let option = bals.get_mut(&tx.from);
             if option.is_none() {
+                if let Instruction::NewVote(_) = &tx.instruction {
+                    inc_new_counter!("bank-appy_debits-vote_account_not_found", 1);
+                } else {
+                    inc_new_counter!("bank-appy_debits-generic_account_not_found", 1);
+                }
                 return Err(BankError::AccountNotFound(tx.from));
             }
             let bal = option.unwrap();

src/banking_stage.rs

@@ -40,7 +40,7 @@ impl BankingStage {
             .name("solana-banking-stage".to_string())
             .spawn(move || loop {
                 if let Err(e) = Self::process_packets(
-                    &bank.clone(),
+                    &bank,
                     &verified_receiver,
                     &signal_sender,
                     &packet_recycler,
@@ -89,7 +89,6 @@ impl BankingStage {
             mms.len(),
         );
         let count = mms.iter().map(|x| x.1.len()).sum();
-        static mut COUNTER: Counter = create_counter!("banking_stage_process_packets", 1);
         let proc_start = Instant::now();
         for (msgs, vers) in mms {
             let transactions = Self::deserialize_transactions(&msgs.read().unwrap());
@@ -125,7 +124,7 @@ impl BankingStage {
             reqs_len,
             (reqs_len as f32) / (total_time_s)
         );
-        inc_counter!(COUNTER, count);
+        inc_new_counter!("banking_stage-process_packets", count);
         Ok(())
     }
 }

src/bin/client-demo.rs

@@ -9,14 +9,13 @@ use bincode::serialize;
 use clap::{App, Arg};
 use rayon::prelude::*;
 use solana::crdt::{Crdt, NodeInfo};
-use solana::drone::DroneRequest;
+use solana::drone::{DroneRequest, DRONE_PORT};
 use solana::fullnode::Config;
 use solana::hash::Hash;
-use solana::mint::Mint;
 use solana::nat::{udp_public_bind, udp_random_bind};
 use solana::ncp::Ncp;
 use solana::service::Service;
-use solana::signature::{GenKeys, KeyPair, KeyPairUtil};
+use solana::signature::{read_keypair, GenKeys, KeyPair, KeyPairUtil};
 use solana::streamer::default_window;
 use solana::thin_client::ThinClient;
 use solana::timing::{duration_as_ms, duration_as_s};
@@ -77,7 +76,7 @@ fn sample_tx_count(
 fn generate_and_send_txs(
     client: &mut ThinClient,
     tx_clients: &[ThinClient],
-    id: &Mint,
+    id: &KeyPair,
     keypairs: &[KeyPair],
     leader: &NodeInfo,
     txs: i64,
@@ -91,7 +90,7 @@ fn generate_and_send_txs(
     let transactions: Vec<_> = if !reclaim {
         keypairs
             .par_iter()
-            .map(|keypair| Transaction::new(&id.keypair(), keypair.pubkey(), 1, *last_id))
+            .map(|keypair| Transaction::new(&id, keypair.pubkey(), 1, *last_id))
             .collect()
     } else {
         keypairs
@@ -164,12 +163,13 @@ fn main() {
                 .help("/path/to/leader.json"),
         )
         .arg(
-            Arg::with_name("mint")
-                .short("m")
-                .long("mint")
+            Arg::with_name("keypair")
+                .short("k")
+                .long("keypair")
                 .value_name("PATH")
                 .takes_value(true)
-                .help("/path/to/mint.json"),
+                .default_value("~/.config/solana/id.json")
+                .help("/path/to/id.json"),
         )
         .arg(
             Arg::with_name("num_nodes")
@@ -205,19 +205,13 @@ fn main() {
         leader = NodeInfo::new_leader(&server_addr);
     };
 
-    let id: Mint;
-    if let Some(m) = matches.value_of("mint") {
-        id = read_mint(m).expect("client mint");
-    } else {
-        eprintln!("No mint found!");
-        exit(1);
-    };
+    let id = read_keypair(matches.value_of("keypair").unwrap()).expect("client keypair");
 
     if let Some(t) = matches.value_of("threads") {
         threads = t.to_string().parse().expect("integer");
     }
 
-    if let Some(n) = matches.value_of("nodes") {
+    if let Some(n) = matches.value_of("num_nodes") {
         num_nodes = n.to_string().parse().expect("integer");
     }
 
@@ -226,12 +220,13 @@ fn main() {
     }
 
     let mut drone_addr = leader.contact_info.tpu;
-    drone_addr.set_port(9900);
+    drone_addr.set_port(DRONE_PORT);
 
     let signal = Arc::new(AtomicBool::new(false));
     let mut c_threads = vec![];
-    let validators = converge(&leader, &signal.clone(), num_nodes, &mut c_threads);
-    assert_eq!(validators.len(), num_nodes);
+    let validators = converge(&leader, &signal, num_nodes, &mut c_threads);
+    println!("Network has {} node(s)", validators.len());
+    assert!(validators.len() >= num_nodes);
 
     let mut client = mk_client(&leader);
 
@@ -259,7 +254,7 @@ fn main() {
     println!("Got last ID {:?}", last_id);
 
     let mut seed = [0u8; 32];
-    seed.copy_from_slice(&id.keypair().public_key_bytes()[..32]);
+    seed.copy_from_slice(&id.public_key_bytes()[..32]);
     let rnd = GenKeys::new(seed);
 
     println!("Creating keypairs...");
@@ -371,6 +366,8 @@ fn spy_node() -> (NodeInfo, UdpSocket) {
     let gossip_socket_pair = udp_public_bind("gossip", 8000, 10000);
     let pubkey = KeyPair::new().pubkey();
     let daddr = "0.0.0.0:0".parse().unwrap();
+    assert!(!gossip_socket_pair.addr.ip().is_unspecified());
+    assert!(!gossip_socket_pair.addr.ip().is_multicast());
     let node = NodeInfo::new(
         pubkey,
         //gossip.local_addr().unwrap(),
@@ -392,14 +389,14 @@ fn converge(
     //lets spy on the network
     let daddr = "0.0.0.0:0".parse().unwrap();
     let (spy, spy_gossip) = spy_node();
-    let mut spy_crdt = Crdt::new(spy);
+    let mut spy_crdt = Crdt::new(spy).expect("Crdt::new");
     spy_crdt.insert(&leader);
     spy_crdt.set_leader(leader.id);
     let spy_ref = Arc::new(RwLock::new(spy_crdt));
     let window = default_window();
     let gossip_send_socket = udp_random_bind(8000, 10000, 5).unwrap();
     let ncp = Ncp::new(
-        &spy_ref.clone(),
+        &spy_ref,
         window.clone(),
         spy_gossip,
         gossip_send_socket,
@@ -421,6 +418,12 @@ fn converge(
             println!("CONVERGED!");
             rv.extend(v.into_iter());
             break;
+        } else {
+            println!(
+                "{} node(s) discovered (looking for {} or more)",
+                v.len(),
+                num_nodes
+            );
         }
         sleep(Duration::new(1, 0));
     }
@@ -433,15 +436,9 @@ fn read_leader(path: &str) -> Config {
     serde_json::from_reader(file).unwrap_or_else(|_| panic!("failed to parse {}", path))
 }
 
-fn read_mint(path: &str) -> Result<Mint, Box<error::Error>> {
-    let file = File::open(path.to_string())?;
-    let mint = serde_json::from_reader(file)?;
-    Ok(mint)
-}
-
 fn request_airdrop(
     drone_addr: &SocketAddr,
-    id: &Mint,
+    id: &KeyPair,
     tokens: u64,
 ) -> Result<(), Box<error::Error>> {
     let mut stream = TcpStream::connect(drone_addr)?;

src/bin/drone.rs

@@ -10,13 +10,12 @@ extern crate tokio_io;
 use bincode::deserialize;
 use clap::{App, Arg};
 use solana::crdt::NodeInfo;
-use solana::drone::{Drone, DroneRequest};
+use solana::drone::{Drone, DroneRequest, DRONE_PORT};
 use solana::fullnode::Config;
-use solana::mint::Mint;
-use std::error;
+use solana::metrics::set_panic_hook;
+use solana::signature::read_keypair;
 use std::fs::File;
 use std::net::{IpAddr, Ipv4Addr, SocketAddr};
-use std::process::exit;
 use std::sync::{Arc, Mutex};
 use std::thread;
 use tokio::net::TcpListener;
@@ -25,7 +24,8 @@ use tokio_codec::{BytesCodec, Decoder};
 
 fn main() {
     env_logger::init();
-    let matches = App::new("solana-client-demo")
+    set_panic_hook("drone");
+    let matches = App::new("drone")
         .arg(
             Arg::with_name("leader")
                 .short("l")
@@ -35,11 +35,12 @@ fn main() {
                 .help("/path/to/leader.json"),
         )
         .arg(
-            Arg::with_name("mint")
-                .short("m")
-                .long("mint")
+            Arg::with_name("keypair")
+                .short("k")
+                .long("keypair")
                 .value_name("PATH")
                 .takes_value(true)
+                .required(true)
                 .help("/path/to/mint.json"),
         )
         .arg(
@@ -68,13 +69,9 @@ fn main() {
         leader = NodeInfo::new_leader(&server_addr);
     };
 
-    let mint: Mint;
-    if let Some(m) = matches.value_of("mint") {
-        mint = read_mint(m).expect("client mint");
-    } else {
-        eprintln!("No mint found!");
-        exit(1);
-    };
+    let mint_keypair =
+        read_keypair(matches.value_of("keypair").expect("keypair")).expect("client keypair");
+
     let time_slice: Option<u64>;
     if let Some(t) = matches.value_of("time") {
         time_slice = Some(t.to_string().parse().expect("integer"));
@@ -88,9 +85,7 @@ fn main() {
         request_cap = None;
     }
 
-    let mint_keypair = mint.keypair();
-
-    let drone_addr: SocketAddr = "0.0.0.0:9900".parse().unwrap();
+    let drone_addr: SocketAddr = format!("0.0.0.0:{}", DRONE_PORT).parse().unwrap();
 
     let drone = Arc::new(Mutex::new(Drone::new(
         mint_keypair,
@@ -152,9 +147,3 @@ fn read_leader(path: &str) -> Config {
     let file = File::open(path).unwrap_or_else(|_| panic!("file not found: {}", path));
     serde_json::from_reader(file).unwrap_or_else(|_| panic!("failed to parse {}", path))
 }
-
-fn read_mint(path: &str) -> Result<Mint, Box<error::Error>> {
-    let file = File::open(path.to_string())?;
-    let mint = serde_json::from_reader(file)?;
-    Ok(mint)
-}

src/bin/fullnode-config.rs

@@ -1,4 +1,5 @@
 extern crate clap;
+extern crate dirs;
 extern crate serde_json;
 extern crate solana;
 
@@ -6,6 +7,7 @@ use clap::{App, Arg};
 use solana::crdt::{get_ip_addr, parse_port_or_addr};
 use solana::fullnode::Config;
 use solana::nat::get_public_ip_addr;
+use solana::signature::read_pkcs8;
 use std::io;
 use std::net::SocketAddr;
 
@@ -18,6 +20,14 @@ fn main() {
                 .takes_value(false)
                 .help("detect network address from local machine configuration"),
         )
+        .arg(
+            Arg::with_name("keypair")
+                .short("k")
+                .long("keypair")
+                .value_name("PATH")
+                .takes_value(true)
+                .help("/path/to/id.json"),
+        )
         .arg(
             Arg::with_name("public")
                 .short("p")
@@ -54,9 +64,18 @@ fn main() {
         bind_addr
     };
 
+    let mut path = dirs::home_dir().expect("home directory");
+    let id_path = if matches.is_present("keypair") {
+        matches.value_of("keypair").unwrap()
+    } else {
+        path.extend(&[".config", "solana", "id.json"]);
+        path.to_str().unwrap()
+    };
+    let pkcs8 = read_pkcs8(id_path).expect("client keypair");
+
     // we need all the receiving sockets to be bound within the expected
     // port range that we open on aws
-    let config = Config::new(&bind_addr);
+    let config = Config::new(&bind_addr, pkcs8);
     let stdout = io::stdout();
     serde_json::to_writer(stdout, &config).expect("serialize");
 }

src/bin/fullnode.rs

@@ -1,4 +1,3 @@
-extern crate atty;
 extern crate clap;
 extern crate env_logger;
 extern crate getopts;
@@ -6,10 +5,10 @@ extern crate log;
 extern crate serde_json;
 extern crate solana;
 
-use atty::{is, Stream};
 use clap::{App, Arg};
 use solana::crdt::{NodeInfo, TestNode};
-use solana::fullnode::{Config, FullNode, InFile, OutFile};
+use solana::fullnode::{Config, FullNode, LedgerFile};
+use solana::metrics::set_panic_hook;
 use solana::service::Service;
 use solana::signature::{KeyPair, KeyPairUtil};
 use std::fs::File;
@@ -19,11 +18,12 @@ use std::process::exit;
 
 fn main() -> () {
     env_logger::init();
+    set_panic_hook("fullnode");
     let matches = App::new("fullnode")
         .arg(
             Arg::with_name("identity")
-                .short("l")
-                .long("local")
+                .short("i")
+                .long("identity")
                 .value_name("FILE")
                 .takes_value(true)
                 .help("run with the identity found in FILE"),
@@ -34,27 +34,23 @@ fn main() -> () {
                 .long("testnet")
                 .value_name("HOST:PORT")
                 .takes_value(true)
-                .help("testnet; connect to the network at this gossip entry point"),
+                .help("connect to the network at this gossip entry point"),
         )
         .arg(
-            Arg::with_name("output")
-                .short("o")
-                .long("output")
+            Arg::with_name("ledger")
+                .short("L")
+                .long("ledger")
                 .value_name("FILE")
                 .takes_value(true)
-                .help("output log to FILE, defaults to stdout (ignored by validators)"),
+                .help("use FILE as persistent ledger (defaults to stdin/stdout)"),
         )
         .get_matches();
-    if is(Stream::Stdin) {
-        eprintln!("nothing found on stdin, expected a log file");
-        exit(1);
-    }
 
     let bind_addr = SocketAddr::new(IpAddr::V4(Ipv4Addr::new(0, 0, 0, 0)), 8000);
     let mut keypair = KeyPair::new();
     let mut repl_data = NodeInfo::new_leader_with_pubkey(keypair.pubkey(), &bind_addr);
-    if let Some(l) = matches.value_of("identity") {
-        let path = l.to_string();
+    if let Some(i) = matches.value_of("identity") {
+        let path = i.to_string();
         if let Ok(file) = File::open(path.clone()) {
             let parse: serde_json::Result<Config> = serde_json::from_reader(file);
             if let Ok(data) = parse {
@@ -69,27 +65,22 @@ fn main() -> () {
             exit(1);
         }
     }
+    let ledger = if let Some(l) = matches.value_of("ledger") {
+        LedgerFile::Path(l.to_string())
+    } else {
+        LedgerFile::StdInOut
+    };
+
     let mut node = TestNode::new_with_bind_addr(repl_data, bind_addr);
     let fullnode = if let Some(t) = matches.value_of("testnet") {
         let testnet_address_string = t.to_string();
         let testnet_addr = testnet_address_string.parse().unwrap();
-        FullNode::new(
-            node,
-            false,
-            InFile::StdIn,
-            Some(keypair),
-            Some(testnet_addr),
-            None,
-        )
+
+        FullNode::new(node, false, ledger, Some(keypair), Some(testnet_addr))
     } else {
         node.data.leader_id = node.data.id;
 
-        let outfile = if let Some(o) = matches.value_of("output") {
-            OutFile::Path(o.to_string())
-        } else {
-            OutFile::StdOut
-        };
-        FullNode::new(node, true, InFile::StdIn, None, None, Some(outfile))
+        FullNode::new(node, true, ledger, None, None)
     };
     fullnode.join().expect("join");
 }

src/bin/genesis.rs

@@ -1,10 +1,13 @@
 //! A command-line executable for generating the chain's genesis block.
 
 extern crate atty;
+#[macro_use]
+extern crate clap;
 extern crate serde_json;
 extern crate solana;
 
 use atty::{is, Stream};
+use clap::{App, Arg};
 use solana::entry_writer::EntryWriter;
 use solana::mint::Mint;
 use std::error;
@@ -12,6 +15,20 @@ use std::io::{stdin, stdout, Read};
 use std::process::exit;
 
 fn main() -> Result<(), Box<error::Error>> {
+    let matches = App::new("solana-genesis")
+        .arg(
+            Arg::with_name("tokens")
+                .short("t")
+                .long("tokens")
+                .value_name("NUMBER")
+                .takes_value(true)
+                .required(true)
+                .help("Number of tokens with which to initialize mint"),
+        )
+        .get_matches();
+
+    let tokens = value_t_or_exit!(matches, "tokens", i64);
+
     if is(Stream::Stdin) {
         eprintln!("nothing found on stdin, expected a json file");
         exit(1);
@@ -24,7 +41,9 @@ fn main() -> Result<(), Box<error::Error>> {
         exit(1);
     }
 
-    let mint: Mint = serde_json::from_str(&buffer)?;
+    let pkcs8: Vec<u8> = serde_json::from_str(&buffer)?;
+    let mint = Mint::new_with_pkcs8(tokens, pkcs8);
+
     let mut writer = stdout();
     EntryWriter::write_entries(&mut writer, mint.create_entries())?;
     Ok(())

src/bin/keygen.rs

@@ -1,14 +1,49 @@
+extern crate clap;
+extern crate dirs;
 extern crate ring;
 extern crate serde_json;
 
+use clap::{App, Arg};
 use ring::rand::SystemRandom;
 use ring::signature::Ed25519KeyPair;
 use std::error;
+use std::fs::{self, File};
+use std::io::Write;
+use std::path::Path;
 
 fn main() -> Result<(), Box<error::Error>> {
+    let matches = App::new("solana-keygen")
+        .arg(
+            Arg::with_name("outfile")
+                .short("o")
+                .long("outfile")
+                .value_name("PATH")
+                .takes_value(true)
+                .help("path to generated file"),
+        )
+        .get_matches();
+
     let rnd = SystemRandom::new();
     let pkcs8_bytes = Ed25519KeyPair::generate_pkcs8(&rnd)?;
     let serialized = serde_json::to_string(&pkcs8_bytes.to_vec())?;
-    println!("{}", serialized);
+
+    let mut path = dirs::home_dir().expect("home directory");
+    let outfile = if matches.is_present("outfile") {
+        matches.value_of("outfile").unwrap()
+    } else {
+        path.extend(&[".config", "solana", "id.json"]);
+        path.to_str().unwrap()
+    };
+
+    if outfile == "-" {
+        println!("{}", serialized);
+    } else {
+        if let Some(outdir) = Path::new(outfile).parent() {
+            fs::create_dir_all(outdir)?;
+        }
+        let mut f = File::create(outfile)?;
+        f.write_all(&serialized.into_bytes())?;
+    }
+
     Ok(())
 }

src/bin/mint.rs

@@ -1,29 +0,0 @@
-extern crate atty;
-extern crate serde_json;
-extern crate solana;
-
-use atty::{is, Stream};
-use solana::mint::Mint;
-use std::io;
-use std::process::exit;
-
-fn main() {
-    let mut input_text = String::new();
-    if is(Stream::Stdin) {
-        eprintln!("nothing found on stdin, expected a token number");
-        exit(1);
-    }
-
-    io::stdin().read_line(&mut input_text).unwrap();
-    let trimmed = input_text.trim();
-    let tokens = trimmed.parse::<i64>().unwrap_or_else(|e| {
-        eprintln!("{}", e);
-        exit(1);
-    });
-    let mint = Mint::new(tokens);
-    let serialized = serde_json::to_string(&mint).unwrap_or_else(|e| {
-        eprintln!("failed to serialize: {}", e);
-        exit(1);
-    });
-    println!("{}", serialized);
-}

src/bin/wallet.rs

@@ -2,6 +2,7 @@ extern crate atty;
 extern crate bincode;
 extern crate bs58;
 extern crate clap;
+extern crate dirs;
 extern crate env_logger;
 extern crate serde_json;
 extern crate solana;
@@ -9,10 +10,9 @@ extern crate solana;
 use bincode::serialize;
 use clap::{App, Arg, SubCommand};
 use solana::crdt::NodeInfo;
-use solana::drone::DroneRequest;
+use solana::drone::{DroneRequest, DRONE_PORT};
 use solana::fullnode::Config;
-use solana::mint::Mint;
-use solana::signature::{PublicKey, Signature};
+use solana::signature::{read_keypair, KeyPair, KeyPairUtil, PublicKey, Signature};
 use solana::thin_client::ThinClient;
 use std::error;
 use std::fmt;
@@ -20,7 +20,6 @@ use std::fs::File;
 use std::io;
 use std::io::prelude::*;
 use std::net::{IpAddr, Ipv4Addr, SocketAddr, TcpStream, UdpSocket};
-use std::process::exit;
 use std::thread::sleep;
 use std::time::Duration;
 
@@ -57,7 +56,7 @@ impl error::Error for WalletError {
 
 struct WalletConfig {
     leader: NodeInfo,
-    id: Mint,
+    id: KeyPair,
     drone_addr: SocketAddr,
     command: WalletCommand,
 }
@@ -67,7 +66,7 @@ impl Default for WalletConfig {
         let default_addr = SocketAddr::new(IpAddr::V4(Ipv4Addr::new(0, 0, 0, 0)), 8000);
         WalletConfig {
             leader: NodeInfo::new_leader(&default_addr),
-            id: Mint::new(0),
+            id: KeyPair::new(),
             drone_addr: default_addr,
             command: WalletCommand::Balance,
         }
@@ -85,12 +84,12 @@ fn parse_args() -> Result<WalletConfig, Box<error::Error>> {
                 .help("/path/to/leader.json"),
         )
         .arg(
-            Arg::with_name("mint")
-                .short("m")
-                .long("mint")
+            Arg::with_name("keypair")
+                .short("k")
+                .long("keypair")
                 .value_name("PATH")
                 .takes_value(true)
-                .help("/path/to/mint.json"),
+                .help("/path/to/id.json"),
         )
         .subcommand(
             SubCommand::with_name("airdrop")
@@ -101,6 +100,7 @@ fn parse_args() -> Result<WalletConfig, Box<error::Error>> {
                         .long("tokens")
                         .value_name("NUMBER")
                         .takes_value(true)
+                        .required(true)
                         .help("The number of tokens to request"),
                 ),
         )
@@ -122,7 +122,6 @@ fn parse_args() -> Result<WalletConfig, Box<error::Error>> {
                         .long("to")
                         .value_name("PUBKEY")
                         .takes_value(true)
-                        .required(true)
                         .help("The pubkey of recipient"),
                 ),
         )
@@ -143,30 +142,33 @@ fn parse_args() -> Result<WalletConfig, Box<error::Error>> {
 
     let leader: NodeInfo;
     if let Some(l) = matches.value_of("leader") {
-        leader = read_leader(l).node_info;
+        leader = read_leader(l)?.node_info;
     } else {
         let server_addr = SocketAddr::new(IpAddr::V4(Ipv4Addr::new(0, 0, 0, 0)), 8000);
         leader = NodeInfo::new_leader(&server_addr);
     };
 
-    let id: Mint;
-    if let Some(m) = matches.value_of("mint") {
-        id = read_mint(m)?;
+    let mut path = dirs::home_dir().expect("home directory");
+    let id_path = if matches.is_present("keypair") {
+        matches.value_of("keypair").unwrap()
     } else {
-        eprintln!("No mint found!");
-        exit(1);
+        path.extend(&[".config", "solana", "id.json"]);
+        path.to_str().unwrap()
     };
+    let id = read_keypair(id_path).or_else(|err| {
+        display_actions();
+        Err(WalletError::BadParameter(format!(
+            "{}: Unable to open keypair file: {}",
+            err, id_path
+        )))
+    })?;
 
     let mut drone_addr = leader.contact_info.tpu;
-    drone_addr.set_port(9900);
+    drone_addr.set_port(DRONE_PORT);
 
     let command = match matches.subcommand() {
         ("airdrop", Some(airdrop_matches)) => {
-            let tokens = if airdrop_matches.is_present("tokens") {
-                airdrop_matches.value_of("tokens").unwrap().parse()?
-            } else {
-                id.tokens
-            };
+            let tokens = airdrop_matches.value_of("tokens").unwrap().parse()?;
             Ok(WalletCommand::AirDrop(tokens))
         }
         ("pay", Some(pay_matches)) => {
@@ -184,11 +186,7 @@ fn parse_args() -> Result<WalletConfig, Box<error::Error>> {
                 id.pubkey()
             };
 
-            let tokens = if pay_matches.is_present("tokens") {
-                pay_matches.value_of("tokens").unwrap().parse()?
-            } else {
-                id.tokens
-            };
+            let tokens = pay_matches.value_of("tokens").unwrap().parse()?;
 
             Ok(WalletCommand::Pay(tokens, to))
         }
@@ -251,20 +249,33 @@ fn process_command(
         // Request an airdrop from Solana Drone;
         // Request amount is set in request_airdrop function
         WalletCommand::AirDrop(tokens) => {
-            println!("Airdrop requested...");
-            println!("Airdropping {:?} tokens", tokens);
-            request_airdrop(&config.drone_addr, &config.id, tokens as u64)?;
-            // TODO: return airdrop Result from Drone
-            sleep(Duration::from_millis(100));
             println!(
-                "Your balance is: {:?}",
-                client.poll_get_balance(&config.id.pubkey()).unwrap()
+                "Requesting airdrop of {:?} tokens from {}",
+                tokens, config.drone_addr
             );
+            let previous_balance = client.poll_get_balance(&config.id.pubkey())?;
+            request_airdrop(&config.drone_addr, &config.id, tokens as u64)?;
+
+            // TODO: return airdrop Result from Drone instead of polling the
+            //       network
+            let mut current_balance = previous_balance;
+            for _ in 0..20 {
+                sleep(Duration::from_millis(500));
+                current_balance = client.poll_get_balance(&config.id.pubkey())?;
+                if previous_balance != current_balance {
+                    break;
+                }
+                println!(".");
+            }
+            println!("Your balance is: {:?}", current_balance);
+            if current_balance - previous_balance != tokens {
+                Err("Airdrop failed!")?;
+            }
         }
         // If client has positive balance, spend tokens in {balance} number of transactions
         WalletCommand::Pay(tokens, to) => {
             let last_id = client.get_last_id();
-            let sig = client.transfer(tokens, &config.id.keypair(), to, &last_id)?;
+            let sig = client.transfer(tokens, &config.id, to, &last_id)?;
             println!("{}", bs58::encode(sig).into_string());
         }
         // Confirm the last client transaction by signature
@@ -290,15 +301,20 @@ fn display_actions() {
     println!();
 }
 
-fn read_leader(path: &str) -> Config {
-    let file = File::open(path.to_string()).unwrap_or_else(|_| panic!("file not found: {}", path));
-    serde_json::from_reader(file).unwrap_or_else(|_| panic!("failed to parse {}", path))
-}
+fn read_leader(path: &str) -> Result<Config, WalletError> {
+    let file = File::open(path.to_string()).or_else(|err| {
+        Err(WalletError::BadParameter(format!(
+            "{}: Unable to open leader file: {}",
+            err, path
+        )))
+    })?;
 
-fn read_mint(path: &str) -> Result<Mint, Box<error::Error>> {
-    let file = File::open(path.to_string())?;
-    let mint = serde_json::from_reader(file)?;
-    Ok(mint)
+    serde_json::from_reader(file).or_else(|err| {
+        Err(WalletError::BadParameter(format!(
+            "{}: Failed to parse leader file: {}",
+            err, path
+        )))
+    })
 }
 
 fn mk_client(r: &NodeInfo) -> io::Result<ThinClient> {
@@ -318,7 +334,7 @@ fn mk_client(r: &NodeInfo) -> io::Result<ThinClient> {
 
 fn request_airdrop(
     drone_addr: &SocketAddr,
-    id: &Mint,
+    id: &KeyPair,
     tokens: u64,
 ) -> Result<(), Box<error::Error>> {
     let mut stream = TcpStream::connect(drone_addr)?;

src/choose_gossip_peer_strategy.rs

@@ -29,7 +29,7 @@ impl<'a, 'b> ChooseRandomPeerStrategy<'a> {
 impl<'a> ChooseGossipPeerStrategy for ChooseRandomPeerStrategy<'a> {
     fn choose_peer<'b>(&self, options: Vec<&'b NodeInfo>) -> Result<&'b NodeInfo> {
         if options.is_empty() {
-            Err(CrdtError::TooSmall)?;
+            Err(CrdtError::NoPeers)?;
         }
 
         let n = ((self.random)() as usize) % options.len();
@@ -174,7 +174,7 @@ impl<'a> ChooseWeightedPeerStrategy<'a> {
 impl<'a> ChooseGossipPeerStrategy for ChooseWeightedPeerStrategy<'a> {
     fn choose_peer<'b>(&self, options: Vec<&'b NodeInfo>) -> Result<&'b NodeInfo> {
         if options.is_empty() {
-            Err(CrdtError::TooSmall)?;
+            Err(CrdtError::NoPeers)?;
         }
 
         let mut weighted_peers = vec![];

src/counter.rs

@@ -1,9 +1,10 @@
 use influx_db_client as influxdb;
 use metrics;
+use std::env;
 use std::sync::atomic::{AtomicUsize, Ordering};
 use timing;
 
-const INFLUX_RATE: usize = 100;
+const DEFAULT_METRICS_RATE: usize = 100;
 
 pub struct Counter {
     pub name: &'static str,
@@ -12,7 +13,7 @@ pub struct Counter {
     pub times: AtomicUsize,
     /// last accumulated value logged
     pub lastlog: AtomicUsize,
-    pub lograte: usize,
+    pub lograte: AtomicUsize,
 }
 
 macro_rules! create_counter {
@@ -22,7 +23,7 @@ macro_rules! create_counter {
             counts: AtomicUsize::new(0),
             times: AtomicUsize::new(0),
             lastlog: AtomicUsize::new(0),
-            lograte: $lograte,
+            lograte: AtomicUsize::new($lograte),
         }
     };
 }
@@ -33,12 +34,38 @@ macro_rules! inc_counter {
     };
 }
 
+macro_rules! inc_new_counter {
+    ($name:expr, $count:expr) => {{
+        static mut INC_NEW_COUNTER: Counter = create_counter!($name, 0);
+        inc_counter!(INC_NEW_COUNTER, $count);
+    }};
+    ($name:expr, $count:expr, $lograte:expr) => {{
+        static mut INC_NEW_COUNTER: Counter = create_counter!($name, $lograte);
+        inc_counter!(INC_NEW_COUNTER, $count);
+    }};
+}
+
 impl Counter {
+    fn default_log_rate() -> usize {
+        let v = env::var("SOLANA_DEFAULT_METRICS_RATE")
+            .map(|x| x.parse().unwrap_or(DEFAULT_METRICS_RATE))
+            .unwrap_or(DEFAULT_METRICS_RATE);
+        if v == 0 {
+            DEFAULT_METRICS_RATE
+        } else {
+            v
+        }
+    }
     pub fn inc(&mut self, events: usize) {
         let counts = self.counts.fetch_add(events, Ordering::Relaxed);
         let times = self.times.fetch_add(1, Ordering::Relaxed);
-        let lastlog = self.lastlog.load(Ordering::Relaxed);
-        if times % self.lograte == 0 && times > 0 {
+        let mut lograte = self.lograte.load(Ordering::Relaxed);
+        if lograte == 0 {
+            lograte = Counter::default_log_rate();
+            self.lograte.store(lograte, Ordering::Relaxed);
+        }
+        if times % lograte == 0 && times > 0 {
+            let lastlog = self.lastlog.load(Ordering::Relaxed);
             info!(
                 "COUNTER:{{\"name\": \"{}\", \"counts\": {}, \"samples\": {},  \"now\": {}}}",
                 self.name,
@@ -46,10 +73,8 @@ impl Counter {
                 times,
                 timing::timestamp(),
             );
-        }
-        if times % INFLUX_RATE == 0 && times > 0 {
             metrics::submit(
-                influxdb::Point::new(&format!("counter_{}", self.name))
+                influxdb::Point::new(&format!("counter-{}", self.name))
                     .add_field(
                         "count",
                         influxdb::Value::Integer(counts as i64 - lastlog as i64),
@@ -63,7 +88,8 @@ impl Counter {
 }
 #[cfg(test)]
 mod tests {
-    use counter::Counter;
+    use counter::{Counter, DEFAULT_METRICS_RATE};
+    use std::env;
     use std::sync::atomic::{AtomicUsize, Ordering};
     #[test]
     fn test_counter() {
@@ -73,7 +99,7 @@ mod tests {
         unsafe {
             assert_eq!(COUNTER.counts.load(Ordering::Relaxed), 1);
             assert_eq!(COUNTER.times.load(Ordering::Relaxed), 1);
-            assert_eq!(COUNTER.lograte, 100);
+            assert_eq!(COUNTER.lograte.load(Ordering::Relaxed), 100);
             assert_eq!(COUNTER.lastlog.load(Ordering::Relaxed), 0);
             assert_eq!(COUNTER.name, "test");
         }
@@ -88,4 +114,42 @@ mod tests {
             assert_eq!(COUNTER.lastlog.load(Ordering::Relaxed), 399);
         }
     }
+    #[test]
+    fn test_inc_new_counter() {
+        //make sure that macros are syntactically correct
+        //the variable is internal to the macro scope so there is no way to introspect it
+        inc_new_counter!("counter-1", 1);
+        inc_new_counter!("counter-2", 1, 2);
+    }
+    #[test]
+    fn test_lograte() {
+        static mut COUNTER: Counter = create_counter!("test_lograte", 0);
+        inc_counter!(COUNTER, 2);
+        unsafe {
+            assert_eq!(
+                COUNTER.lograte.load(Ordering::Relaxed),
+                DEFAULT_METRICS_RATE
+            );
+        }
+    }
+    #[test]
+    fn test_lograte_env() {
+        assert_ne!(DEFAULT_METRICS_RATE, 0);
+        static mut COUNTER: Counter = create_counter!("test_lograte_env", 0);
+        env::set_var("SOLANA_DEFAULT_METRICS_RATE", "50");
+        inc_counter!(COUNTER, 2);
+        unsafe {
+            assert_eq!(COUNTER.lograte.load(Ordering::Relaxed), 50);
+        }
+
+        static mut COUNTER2: Counter = create_counter!("test_lograte_env", 0);
+        env::set_var("SOLANA_DEFAULT_METRICS_RATE", "0");
+        inc_counter!(COUNTER2, 2);
+        unsafe {
+            assert_eq!(
+                COUNTER2.lograte.load(Ordering::Relaxed),
+                DEFAULT_METRICS_RATE
+            );
+        }
+    }
 }

src/crdt.rs

@@ -37,8 +37,6 @@ use streamer::{BlobReceiver, BlobSender, Window};
 use timing::timestamp;
 use transaction::Vote;
 
-const LOG_RATE: usize = 10;
-
 /// milliseconds we sleep for between gossip requests
 const GOSSIP_SLEEP_MILLIS: u64 = 100;
 const GOSSIP_PURGE_MILLIS: u64 = 15000;
@@ -48,8 +46,11 @@ const MIN_TABLE_SIZE: usize = 2;
 
 #[derive(Debug, PartialEq, Eq)]
 pub enum CrdtError {
-    TooSmall,
+    NoPeers,
     NoLeader,
+    BadContactInfo,
+    BadNodeInfo,
+    BadGossipAddress,
 }
 
 pub fn parse_port_or_addr(optstr: Option<String>) -> SocketAddr {
@@ -119,8 +120,6 @@ pub struct ContactInfo {
 pub struct LedgerState {
     /// last verified hash that was submitted to the leader
     pub last_id: Hash,
-    /// last verified entry count, always increasing
-    pub entry_height: u64,
 }
 
 #[derive(Serialize, Deserialize, Clone, Debug, PartialEq)]
@@ -150,7 +149,7 @@ impl NodeInfo {
         rpu: SocketAddr,
         tpu: SocketAddr,
         tvu_window: SocketAddr,
-    ) -> NodeInfo {
+    ) -> Self {
         NodeInfo {
             id,
             version: 0,
@@ -165,10 +164,38 @@ impl NodeInfo {
             leader_id: PublicKey::default(),
             ledger_state: LedgerState {
                 last_id: Hash::default(),
-                entry_height: 0,
             },
         }
     }
+    #[cfg(test)]
+    /// NodeInfo with unspecified addresses for adversarial testing.
+    pub fn new_unspecified() -> Self {
+        let addr: SocketAddr = "0.0.0.0:0".parse().unwrap();
+        assert!(addr.ip().is_unspecified());
+        Self::new(
+            KeyPair::new().pubkey(),
+            addr.clone(),
+            addr.clone(),
+            addr.clone(),
+            addr.clone(),
+            addr.clone(),
+        )
+    }
+    #[cfg(test)]
+    /// NodeInfo with multicast addresses for adversarial testing.
+    pub fn new_multicast() -> Self {
+        let addr: SocketAddr = "224.0.1.255:1000".parse().unwrap();
+        assert!(addr.ip().is_multicast());
+        Self::new(
+            KeyPair::new().pubkey(),
+            addr.clone(),
+            addr.clone(),
+            addr.clone(),
+            addr.clone(),
+            addr.clone(),
+        )
+    }
+
     pub fn debug_id(&self) -> u64 {
         make_debug_id(&self.id)
     }
@@ -256,8 +283,31 @@ enum Protocol {
 }
 
 impl Crdt {
-    pub fn new(me: NodeInfo) -> Crdt {
-        assert_eq!(me.version, 0);
+    pub fn new(me: NodeInfo) -> Result<Crdt> {
+        if me.version != 0 {
+            return Err(Error::CrdtError(CrdtError::BadNodeInfo));
+        }
+        if me.contact_info.ncp.ip().is_unspecified()
+            || me.contact_info.ncp.port() == 0
+            || me.contact_info.ncp.ip().is_multicast()
+        {
+            return Err(Error::CrdtError(CrdtError::BadGossipAddress));
+        }
+        for addr in &[
+            me.contact_info.tvu,
+            me.contact_info.rpu,
+            me.contact_info.tpu,
+            me.contact_info.tvu_window,
+        ] {
+            //dummy address is allowed, services will filter them
+            if addr.ip().is_unspecified() && addr.port() == 0 {
+                continue;
+            }
+            //if addr is not a dummy address, than it must be valid
+            if addr.ip().is_unspecified() || addr.port() == 0 || addr.ip().is_multicast() {
+                return Err(Error::CrdtError(CrdtError::BadContactInfo));
+            }
+        }
         let mut g = Crdt {
             table: HashMap::new(),
             local: HashMap::new(),
@@ -269,7 +319,7 @@ impl Crdt {
         };
         g.local.insert(me.id, g.update_index);
         g.table.insert(me.id, me);
-        g
+        Ok(g)
     }
     pub fn debug_id(&self) -> u64 {
         make_debug_id(&self.me)
@@ -348,8 +398,7 @@ impl Crdt {
         }
     }
     pub fn insert_votes(&mut self, votes: &[(PublicKey, Vote, Hash)]) {
-        static mut COUNTER_VOTE: Counter = create_counter!("crdt-vote-count", LOG_RATE);
-        inc_counter!(COUNTER_VOTE, votes.len());
+        inc_new_counter!("crdt-vote-count", votes.len());
         if !votes.is_empty() {
             info!("{:x}: INSERTING VOTES {}", self.debug_id(), votes.len());
         }
@@ -369,12 +418,15 @@ impl Crdt {
                 v.debug_id(),
                 v.version
             );
+            if self.table.get(&v.id).is_none() {
+                inc_new_counter!("crdt-insert-new_entry", 1, 1);
+            }
 
             self.update_index += 1;
             let _ = self.table.insert(v.id, v.clone());
             let _ = self.local.insert(v.id, self.update_index);
-            static mut COUNTER_UPDATE: Counter = create_counter!("crdt-update-count", LOG_RATE);
-            inc_counter!(COUNTER_UPDATE, 1);
+            inc_new_counter!("crdt-update-count", 1);
+            self.update_liveness(v.id);
         } else {
             trace!(
                 "{:x}: INSERT FAILED data: {:x} new.version: {} me.version: {}",
@@ -384,7 +436,6 @@ impl Crdt {
                 self.table[&v.id].version
             );
         }
-        self.update_liveness(v.id);
     }
 
     fn update_liveness(&mut self, id: PublicKey) {
@@ -404,35 +455,20 @@ impl Crdt {
     /// challenging part is that we are on a permissionless network
     pub fn purge(&mut self, now: u64) {
         if self.table.len() <= MIN_TABLE_SIZE {
+            trace!("purge: skipped: table too small: {}", self.table.len());
             return;
         }
         if self.leader_data().is_none() {
+            trace!("purge: skipped: no leader_data");
             return;
         }
         let leader_id = self.leader_data().unwrap().id;
-
         let limit = GOSSIP_PURGE_MILLIS;
         let dead_ids: Vec<PublicKey> = self.alive
             .iter()
             .filter_map(|(&k, v)| {
                 if k != self.me && (now - v) > limit {
-                    if leader_id == k {
-                        info!(
-                            "{:x}: PURGE LEADER {:x} {}",
-                            self.debug_id(),
-                            make_debug_id(&k),
-                            now - v
-                        );
-                        Some(k)
-                    } else {
-                        info!(
-                            "{:x}: PURGE {:x} {}",
-                            self.debug_id(),
-                            make_debug_id(&k),
-                            now - v
-                        );
-                        Some(k)
-                    }
+                    Some(k)
                 } else {
                     trace!(
                         "{:x} purge skipped {:x} {} {}",
@@ -446,8 +482,7 @@ impl Crdt {
             })
             .collect();
 
-        static mut COUNTER_PURGE: Counter = create_counter!("crdt-purge-count", LOG_RATE);
-        inc_counter!(COUNTER_PURGE, dead_ids.len());
+        inc_new_counter!("crdt-purge-count", dead_ids.len());
 
         for id in &dead_ids {
             self.alive.remove(id);
@@ -455,9 +490,19 @@ impl Crdt {
             self.remote.remove(id);
             self.local.remove(id);
             self.external_liveness.remove(id);
+            info!("{:x}: PURGE {:x}", self.debug_id(), make_debug_id(id));
             for map in self.external_liveness.values_mut() {
                 map.remove(id);
             }
+            if *id == leader_id {
+                info!(
+                    "{:x}: PURGE LEADER {:x}",
+                    self.debug_id(),
+                    make_debug_id(id),
+                );
+                inc_new_counter!("crdt-purge-purged_leader", 1, 1);
+                self.set_leader(PublicKey::default());
+            }
         }
     }
 
@@ -522,7 +567,8 @@ impl Crdt {
     ) -> Result<()> {
         if broadcast_table.is_empty() {
             warn!("{:x}:not enough peers in crdt table", me.debug_id());
-            Err(CrdtError::TooSmall)?;
+            inc_new_counter!("crdt-broadcast-not_enough_peers_error", 1, 1);
+            Err(CrdtError::NoPeers)?;
         }
         trace!("broadcast nodes {}", broadcast_table.len());
 
@@ -628,7 +674,8 @@ impl Crdt {
             .collect();
         for e in errs {
             if let Err(e) = &e {
-                error!("broadcast result {:?}", e);
+                inc_new_counter!("crdt-retransmit-send_to_error", 1, 1);
+                error!("retransmit result {:?}", e);
             }
             e?;
         }
@@ -669,7 +716,7 @@ impl Crdt {
             .filter(|r| r.id != self.me && r.contact_info.tvu_window != daddr)
             .collect();
         if valid.is_empty() {
-            Err(CrdtError::TooSmall)?;
+            Err(CrdtError::NoPeers)?;
         }
         let n = (Self::random() as usize) % valid.len();
         let addr = valid[n].contact_info.ncp;
@@ -684,7 +731,14 @@ impl Crdt {
     /// * A - Address to send to
     /// * B - RequestUpdates protocol message
     fn gossip_request(&self) -> Result<(SocketAddr, Protocol)> {
-        let options: Vec<_> = self.table.values().filter(|v| v.id != self.me).collect();
+        let options: Vec<_> = self.table
+            .values()
+            .filter(|v| {
+                v.id != self.me
+                    && !v.contact_info.ncp.ip().is_unspecified()
+                    && !v.contact_info.ncp.ip().is_multicast()
+            })
+            .collect();
 
         let choose_peer_strategy = ChooseWeightedPeerStrategy::new(
             &self.remote,
@@ -694,7 +748,7 @@ impl Crdt {
 
         let choose_peer_result = choose_peer_strategy.choose_peer(options);
 
-        if let Err(Error::CrdtError(CrdtError::TooSmall)) = &choose_peer_result {
+        if let Err(Error::CrdtError(CrdtError::NoPeers)) = &choose_peer_result {
             trace!(
                 "crdt too small for gossip {:x} {}",
                 self.debug_id(),
@@ -715,12 +769,11 @@ impl Crdt {
         Ok((v.contact_info.ncp, req))
     }
 
-    pub fn new_vote(&mut self, height: u64, last_id: Hash) -> Result<(Vote, SocketAddr)> {
+    pub fn new_vote(&mut self, last_id: Hash) -> Result<(Vote, SocketAddr)> {
         let mut me = self.my_data().clone();
         let leader = self.leader_data().ok_or(CrdtError::NoLeader)?.clone();
         me.version += 1;
         me.ledger_state.last_id = last_id;
-        me.ledger_state.entry_height = height;
         let vote = Vote {
             version: me.version,
             contact_info_version: me.contact_info.version,
@@ -898,24 +951,18 @@ impl Crdt {
                     outblob.meta.set_addr(&from.contact_info.tvu_window);
                     outblob.set_id(sender_id).expect("blob set_id");
                 }
-                static mut COUNTER_REQ_WINDOW_PASS: Counter =
-                    create_counter!("crdt-window-request-pass", LOG_RATE);
-                inc_counter!(COUNTER_REQ_WINDOW_PASS, 1);
+                inc_new_counter!("crdt-window-request-pass", 1);
 
                 return Some(out);
             } else {
-                static mut COUNTER_REQ_WINDOW_OUTSIDE: Counter =
-                    create_counter!("crdt-window-request-outside", LOG_RATE);
-                inc_counter!(COUNTER_REQ_WINDOW_OUTSIDE, 1);
+                inc_new_counter!("crdt-window-request-outside", 1);
                 info!(
                     "requested ix {} != blob_ix {}, outside window!",
                     ix, blob_ix
                 );
             }
         } else {
-            static mut COUNTER_REQ_WINDOW_FAIL: Counter =
-                create_counter!("crdt-window-request-fail", LOG_RATE);
-            inc_counter!(COUNTER_REQ_WINDOW_FAIL, 1);
+            inc_new_counter!("crdt-window-request-fail", 1);
             assert!(window.read().unwrap()[pos].is_none());
             info!(
                 "{:x}: failed RequestWindowIndex {:x} {} {}",
@@ -937,11 +984,35 @@ impl Crdt {
         blob: &Blob,
     ) -> Option<SharedBlob> {
         match deserialize(&blob.data[..blob.meta.size]) {
+            Ok(request) => Crdt::handle_protocol(request, obj, window, blob_recycler),
+            Err(_) => {
+                warn!("deserialize crdt packet failed");
+                None
+            }
+        }
+    }
+
+    fn handle_protocol(
+        request: Protocol,
+        obj: &Arc<RwLock<Self>>,
+        window: &Window,
+        blob_recycler: &BlobRecycler,
+    ) -> Option<SharedBlob> {
+        match request {
             // TODO sigverify these
-            Ok(Protocol::RequestUpdates(v, from_rd)) => {
-                trace!("RequestUpdates {}", v);
+            Protocol::RequestUpdates(v, from_rd) => {
                 let addr = from_rd.contact_info.ncp;
+                trace!("RequestUpdates {} from {}", v, addr);
                 let me = obj.read().unwrap();
+                if addr == me.table[&me.me].contact_info.ncp {
+                    warn!(
+                        "RequestUpdates ignored, I'm talking to myself: me={:x} remoteme={:x}",
+                        me.debug_id(),
+                        make_debug_id(&from_rd.id)
+                    );
+                    inc_new_counter!("crdt-window-request-loopback", 1);
+                    return None;
+                }
                 // only lock for these two calls, dont lock during IO `sock.send_to` or `sock.recv_from`
                 let (from, ups, data) = me.get_updates_since(v);
                 let external_liveness = me.remote.iter().map(|(k, v)| (*k, *v)).collect();
@@ -949,7 +1020,11 @@ impl Crdt {
                 trace!("get updates since response {} {}", v, data.len());
                 let len = data.len();
                 let rsp = Protocol::ReceiveUpdates(from, ups, data, external_liveness);
-                obj.write().unwrap().insert(&from_rd);
+                {
+                    let mut me = obj.write().unwrap();
+                    me.insert(&from_rd);
+                    me.update_liveness(from_rd.id);
+                }
                 if len < 1 {
                     let me = obj.read().unwrap();
                     trace!(
@@ -973,40 +1048,43 @@ impl Crdt {
                     None
                 }
             }
-            Ok(Protocol::ReceiveUpdates(from, ups, data, external_liveness)) => {
+            Protocol::ReceiveUpdates(from, update_index, data, external_liveness) => {
                 trace!(
-                    "ReceivedUpdates {:x} {} {}",
+                    "ReceivedUpdates from={:x} update_index={} len={}",
                     make_debug_id(&from),
-                    ups,
+                    update_index,
                     data.len()
                 );
                 obj.write()
                     .expect("'obj' write lock in ReceiveUpdates")
-                    .apply_updates(from, ups, &data, &external_liveness);
+                    .apply_updates(from, update_index, &data, &external_liveness);
                 None
             }
-            Ok(Protocol::RequestWindowIndex(from, ix)) => {
+            Protocol::RequestWindowIndex(from, ix) => {
                 //TODO this doesn't depend on CRDT module, can be moved
                 //but we are using the listen thread to service these request
                 //TODO verify from is signed
                 obj.write().unwrap().insert(&from);
                 let me = obj.read().unwrap().my_data().clone();
-                static mut COUNTER_REQ_WINDOW: Counter =
-                    create_counter!("crdt-window-request-recv", LOG_RATE);
-                inc_counter!(COUNTER_REQ_WINDOW, 1);
+                inc_new_counter!("crdt-window-request-recv", 1);
                 trace!(
                     "{:x}:received RequestWindowIndex {:x} {} ",
                     me.debug_id(),
                     from.debug_id(),
                     ix,
                 );
-                assert_ne!(from.contact_info.tvu_window, me.contact_info.tvu_window);
+                if from.contact_info.tvu_window == me.contact_info.tvu_window {
+                    warn!(
+                        "Ignored {:x}:received RequestWindowIndex from ME {:x} {} ",
+                        me.debug_id(),
+                        from.debug_id(),
+                        ix,
+                    );
+                    inc_new_counter!("crdt-window-request-address-eq", 1);
+                    return None;
+                }
                 Self::run_window_request(&window, &me, &from, ix, blob_recycler)
             }
-            Err(_) => {
-                warn!("deserialize crdt packet failed");
-                None
-            }
         }
     }
 
@@ -1084,23 +1162,18 @@ pub struct TestNode {
     pub sockets: Sockets,
 }
 
-impl Default for TestNode {
-    fn default() -> Self {
-        Self::new()
-    }
-}
-
 impl TestNode {
-    pub fn new() -> Self {
+    pub fn new_localhost() -> Self {
         let pubkey = KeyPair::new().pubkey();
-        Self::new_with_pubkey(pubkey)
+        Self::new_localhost_with_pubkey(pubkey)
     }
-    pub fn new_with_pubkey(pubkey: PublicKey) -> Self {
-        let transaction = UdpSocket::bind("0.0.0.0:0").unwrap();
-        let gossip = UdpSocket::bind("0.0.0.0:0").unwrap();
-        let replicate = UdpSocket::bind("0.0.0.0:0").unwrap();
-        let requests = UdpSocket::bind("0.0.0.0:0").unwrap();
-        let repair = UdpSocket::bind("0.0.0.0:0").unwrap();
+    pub fn new_localhost_with_pubkey(pubkey: PublicKey) -> Self {
+        let transaction = UdpSocket::bind("127.0.0.1:0").unwrap();
+        let gossip = UdpSocket::bind("127.0.0.1:0").unwrap();
+        let replicate = UdpSocket::bind("127.0.0.1:0").unwrap();
+        let requests = UdpSocket::bind("127.0.0.1:0").unwrap();
+        let repair = UdpSocket::bind("127.0.0.1:0").unwrap();
+
         let gossip_send = UdpSocket::bind("0.0.0.0:0").unwrap();
         let respond = UdpSocket::bind("0.0.0.0:0").unwrap();
         let broadcast = UdpSocket::bind("0.0.0.0:0").unwrap();
@@ -1178,14 +1251,14 @@ impl TestNode {
 #[cfg(test)]
 mod tests {
     use crdt::{
-        parse_port_or_addr, Crdt, CrdtError, NodeInfo, GOSSIP_PURGE_MILLIS, GOSSIP_SLEEP_MILLIS,
-        MIN_TABLE_SIZE,
+        parse_port_or_addr, Crdt, CrdtError, NodeInfo, Protocol, GOSSIP_PURGE_MILLIS,
+        GOSSIP_SLEEP_MILLIS, MIN_TABLE_SIZE,
     };
     use hash::Hash;
     use logger;
     use packet::BlobRecycler;
     use result::Error;
-    use signature::{KeyPair, KeyPairUtil};
+    use signature::{KeyPair, KeyPairUtil, PublicKey};
     use std::sync::atomic::{AtomicBool, Ordering};
     use std::sync::mpsc::channel;
     use std::sync::{Arc, RwLock};
@@ -1203,6 +1276,107 @@ mod tests {
         let p3 = parse_port_or_addr(None);
         assert_eq!(p3.port(), 8000);
     }
+    #[test]
+    fn test_bad_address() {
+        let d1 = NodeInfo::new(
+            KeyPair::new().pubkey(),
+            "0.0.0.0:1234".parse().unwrap(),
+            "0.0.0.0:1235".parse().unwrap(),
+            "0.0.0.0:1236".parse().unwrap(),
+            "0.0.0.0:1237".parse().unwrap(),
+            "0.0.0.0:1238".parse().unwrap(),
+        );
+        assert_matches!(
+            Crdt::new(d1).err(),
+            Some(Error::CrdtError(CrdtError::BadGossipAddress))
+        );
+        let d1_1 = NodeInfo::new(
+            KeyPair::new().pubkey(),
+            "0.0.0.1:1234".parse().unwrap(),
+            "0.0.0.0:1235".parse().unwrap(),
+            "0.0.0.0:1236".parse().unwrap(),
+            "0.0.0.0:1237".parse().unwrap(),
+            "0.0.0.0:1238".parse().unwrap(),
+        );
+        assert_matches!(
+            Crdt::new(d1_1).err(),
+            Some(Error::CrdtError(CrdtError::BadContactInfo))
+        );
+        let d2 = NodeInfo::new(
+            KeyPair::new().pubkey(),
+            "0.0.0.1:0".parse().unwrap(),
+            "0.0.0.1:0".parse().unwrap(),
+            "0.0.0.1:0".parse().unwrap(),
+            "0.0.0.1:0".parse().unwrap(),
+            "0.0.0.1:0".parse().unwrap(),
+        );
+        assert_matches!(
+            Crdt::new(d2).err(),
+            Some(Error::CrdtError(CrdtError::BadGossipAddress))
+        );
+        let d2_1 = NodeInfo::new(
+            KeyPair::new().pubkey(),
+            "0.0.0.1:1234".parse().unwrap(),
+            "0.0.0.1:0".parse().unwrap(),
+            "0.0.0.1:0".parse().unwrap(),
+            "0.0.0.1:0".parse().unwrap(),
+            "0.0.0.1:0".parse().unwrap(),
+        );
+        assert_matches!(
+            Crdt::new(d2_1).err(),
+            Some(Error::CrdtError(CrdtError::BadContactInfo))
+        );
+        let d3 = NodeInfo::new_unspecified();
+        assert_matches!(
+            Crdt::new(d3).err(),
+            Some(Error::CrdtError(CrdtError::BadGossipAddress))
+        );
+        let d4 = NodeInfo::new_multicast();
+        assert_matches!(
+            Crdt::new(d4).err(),
+            Some(Error::CrdtError(CrdtError::BadGossipAddress))
+        );
+        let mut d5 = NodeInfo::new_multicast();
+        d5.version = 1;
+        assert_matches!(
+            Crdt::new(d5).err(),
+            Some(Error::CrdtError(CrdtError::BadNodeInfo))
+        );
+        let d6 = NodeInfo::new(
+            KeyPair::new().pubkey(),
+            "0.0.0.0:1234".parse().unwrap(),
+            "0.0.0.0:0".parse().unwrap(),
+            "0.0.0.0:0".parse().unwrap(),
+            "0.0.0.0:0".parse().unwrap(),
+            "0.0.0.0:0".parse().unwrap(),
+        );
+        assert_matches!(
+            Crdt::new(d6).err(),
+            Some(Error::CrdtError(CrdtError::BadGossipAddress))
+        );
+        let d7 = NodeInfo::new(
+            KeyPair::new().pubkey(),
+            "0.0.0.1:0".parse().unwrap(),
+            "0.0.0.0:0".parse().unwrap(),
+            "0.0.0.0:0".parse().unwrap(),
+            "0.0.0.0:0".parse().unwrap(),
+            "0.0.0.0:0".parse().unwrap(),
+        );
+        assert_matches!(
+            Crdt::new(d7).err(),
+            Some(Error::CrdtError(CrdtError::BadGossipAddress))
+        );
+        let d8 = NodeInfo::new(
+            KeyPair::new().pubkey(),
+            "0.0.0.1:1234".parse().unwrap(),
+            "0.0.0.0:0".parse().unwrap(),
+            "0.0.0.0:0".parse().unwrap(),
+            "0.0.0.0:0".parse().unwrap(),
+            "0.0.0.0:0".parse().unwrap(),
+        );
+        assert_eq!(Crdt::new(d8).is_ok(), true);
+    }
+
     #[test]
     fn insert_test() {
         let mut d = NodeInfo::new(
@@ -1214,30 +1388,43 @@ mod tests {
             "127.0.0.1:1238".parse().unwrap(),
         );
         assert_eq!(d.version, 0);
-        let mut crdt = Crdt::new(d.clone());
+        let mut crdt = Crdt::new(d.clone()).unwrap();
         assert_eq!(crdt.table[&d.id].version, 0);
+        assert!(!crdt.alive.contains_key(&d.id));
+
         d.version = 2;
         crdt.insert(&d);
+        let liveness = crdt.alive[&d.id];
         assert_eq!(crdt.table[&d.id].version, 2);
+
         d.version = 1;
         crdt.insert(&d);
         assert_eq!(crdt.table[&d.id].version, 2);
+        assert_eq!(liveness, crdt.alive[&d.id]);
+
+        // Ensure liveness will be updated for version 3
+        sleep(Duration::from_millis(1));
+
+        d.version = 3;
+        crdt.insert(&d);
+        assert_eq!(crdt.table[&d.id].version, 3);
+        assert!(liveness < crdt.alive[&d.id]);
     }
     #[test]
     fn test_new_vote() {
         let d = NodeInfo::new_leader(&"127.0.0.1:1234".parse().unwrap());
         assert_eq!(d.version, 0);
-        let mut crdt = Crdt::new(d.clone());
+        let mut crdt = Crdt::new(d.clone()).unwrap();
         assert_eq!(crdt.table[&d.id].version, 0);
         let leader = NodeInfo::new_leader(&"127.0.0.2:1235".parse().unwrap());
         assert_ne!(d.id, leader.id);
         assert_matches!(
-            crdt.new_vote(0, Hash::default()).err(),
+            crdt.new_vote(Hash::default()).err(),
             Some(Error::CrdtError(CrdtError::NoLeader))
         );
         crdt.insert(&leader);
         assert_matches!(
-            crdt.new_vote(0, Hash::default()).err(),
+            crdt.new_vote(Hash::default()).err(),
             Some(Error::CrdtError(CrdtError::NoLeader))
         );
         crdt.set_leader(leader.id);
@@ -1247,14 +1434,14 @@ mod tests {
             contact_info_version: 0,
         };
         let expected = (v, crdt.table[&leader.id].contact_info.tpu);
-        assert_eq!(crdt.new_vote(0, Hash::default()).unwrap(), expected);
+        assert_eq!(crdt.new_vote(Hash::default()).unwrap(), expected);
     }
 
     #[test]
     fn test_insert_vote() {
         let d = NodeInfo::new_leader(&"127.0.0.1:1234".parse().unwrap());
         assert_eq!(d.version, 0);
-        let mut crdt = Crdt::new(d.clone());
+        let mut crdt = Crdt::new(d.clone()).unwrap();
         assert_eq!(crdt.table[&d.id].version, 0);
         let vote_same_version = Vote {
             version: d.version,
@@ -1286,7 +1473,7 @@ mod tests {
         // TODO: remove this test once leaders vote
         let d = NodeInfo::new_leader(&"127.0.0.1:1234".parse().unwrap());
         assert_eq!(d.version, 0);
-        let mut crdt = Crdt::new(d.clone());
+        let mut crdt = Crdt::new(d.clone()).unwrap();
         let leader = NodeInfo::new_leader(&"127.0.0.2:1235".parse().unwrap());
         assert_ne!(d.id, leader.id);
         crdt.insert(&leader);
@@ -1355,7 +1542,7 @@ mod tests {
             "127.0.0.1:1237".parse().unwrap(),
             "127.0.0.1:1238".parse().unwrap(),
         );
-        let mut crdt = Crdt::new(d1.clone());
+        let mut crdt = Crdt::new(d1.clone()).expect("Crdt::new");
         let (key, ix, ups) = crdt.get_updates_since(0);
         assert_eq!(key, d1.id);
         assert_eq!(ix, 1);
@@ -1376,7 +1563,7 @@ mod tests {
             sorted(&ups),
             sorted(&vec![d1.clone(), d2.clone(), d3.clone()])
         );
-        let mut crdt2 = Crdt::new(d2.clone());
+        let mut crdt2 = Crdt::new(d2.clone()).expect("Crdt::new");
         crdt2.apply_updates(key, ix, &ups, &vec![]);
         assert_eq!(crdt2.table.values().len(), 3);
         assert_eq!(
@@ -1398,9 +1585,9 @@ mod tests {
             "127.0.0.1:1237".parse().unwrap(),
             "127.0.0.1:1238".parse().unwrap(),
         );
-        let mut crdt = Crdt::new(me.clone());
+        let mut crdt = Crdt::new(me.clone()).expect("Crdt::new");
         let rv = crdt.window_index_request(0);
-        assert_matches!(rv, Err(Error::CrdtError(CrdtError::TooSmall)));
+        assert_matches!(rv, Err(Error::CrdtError(CrdtError::NoPeers)));
         let nxt = NodeInfo::new(
             KeyPair::new().pubkey(),
             "127.0.0.1:1234".parse().unwrap(),
@@ -1411,7 +1598,7 @@ mod tests {
         );
         crdt.insert(&nxt);
         let rv = crdt.window_index_request(0);
-        assert_matches!(rv, Err(Error::CrdtError(CrdtError::TooSmall)));
+        assert_matches!(rv, Err(Error::CrdtError(CrdtError::NoPeers)));
         let nxt = NodeInfo::new(
             KeyPair::new().pubkey(),
             "127.0.0.2:1234".parse().unwrap(),
@@ -1449,6 +1636,30 @@ mod tests {
         assert!(one && two);
     }
 
+    #[test]
+    fn gossip_request_bad_addr() {
+        let me = NodeInfo::new(
+            KeyPair::new().pubkey(),
+            "127.0.0.1:127".parse().unwrap(),
+            "127.0.0.1:127".parse().unwrap(),
+            "127.0.0.1:127".parse().unwrap(),
+            "127.0.0.1:127".parse().unwrap(),
+            "127.0.0.1:127".parse().unwrap(),
+        );
+
+        let mut crdt = Crdt::new(me).expect("Crdt::new");
+        let nxt1 = NodeInfo::new_unspecified();
+        // Filter out unspecified addresses
+        crdt.insert(&nxt1); //<--- attack!
+        let rv = crdt.gossip_request();
+        assert_matches!(rv, Err(Error::CrdtError(CrdtError::NoPeers)));
+        let nxt2 = NodeInfo::new_multicast();
+        // Filter out multicast addresses
+        crdt.insert(&nxt2); //<--- attack!
+        let rv = crdt.gossip_request();
+        assert_matches!(rv, Err(Error::CrdtError(CrdtError::NoPeers)));
+    }
+
     /// test that gossip requests are eventually generated for all nodes
     #[test]
     fn gossip_request() {
@@ -1460,9 +1671,9 @@ mod tests {
             "127.0.0.1:1237".parse().unwrap(),
             "127.0.0.1:1238".parse().unwrap(),
         );
-        let mut crdt = Crdt::new(me.clone());
+        let mut crdt = Crdt::new(me.clone()).expect("Crdt::new");
         let rv = crdt.gossip_request();
-        assert_matches!(rv, Err(Error::CrdtError(CrdtError::TooSmall)));
+        assert_matches!(rv, Err(Error::CrdtError(CrdtError::NoPeers)));
         let nxt1 = NodeInfo::new(
             KeyPair::new().pubkey(),
             "127.0.0.2:1234".parse().unwrap(),
@@ -1519,7 +1730,7 @@ mod tests {
     fn purge_test() {
         logger::setup();
         let me = NodeInfo::new_leader(&"127.0.0.1:1234".parse().unwrap());
-        let mut crdt = Crdt::new(me.clone());
+        let mut crdt = Crdt::new(me.clone()).expect("Crdt::new");
         let nxt = NodeInfo::new_leader(&"127.0.0.2:1234".parse().unwrap());
         assert_ne!(me.id, nxt.id);
         crdt.set_leader(me.id);
@@ -1557,6 +1768,28 @@ mod tests {
         let rv = crdt.gossip_request().unwrap();
         assert_eq!(rv.0, nxt.contact_info.ncp);
     }
+    #[test]
+    fn purge_leader_test() {
+        logger::setup();
+        let me = NodeInfo::new_leader(&"127.0.0.1:1234".parse().unwrap());
+        let mut crdt = Crdt::new(me.clone()).expect("Crdt::new");
+        let nxt = NodeInfo::new_leader(&"127.0.0.2:1234".parse().unwrap());
+        assert_ne!(me.id, nxt.id);
+        crdt.insert(&nxt);
+        crdt.set_leader(nxt.id);
+        let now = crdt.alive[&nxt.id];
+        let mut nxt2 = NodeInfo::new_leader(&"127.0.0.2:1234".parse().unwrap());
+        crdt.insert(&nxt2);
+        while now == crdt.alive[&nxt2.id] {
+            sleep(Duration::from_millis(GOSSIP_SLEEP_MILLIS));
+            nxt2.version = nxt2.version + 1;
+            crdt.insert(&nxt2);
+        }
+        let len = crdt.table.len() as u64;
+        crdt.purge(now + GOSSIP_PURGE_MILLIS + 1);
+        assert_eq!(len as usize - 1, crdt.table.len());
+        assert_eq!(crdt.my_data().leader_id, PublicKey::default());
+    }
 
     /// test window requests respond with the right blob, and do not overrun
     #[test]
@@ -1630,7 +1863,7 @@ mod tests {
         let me = NodeInfo::new_leader(&"127.0.0.1:1234".parse().unwrap());
         let leader0 = NodeInfo::new_leader(&"127.0.0.1:1234".parse().unwrap());
         let leader1 = NodeInfo::new_leader(&"127.0.0.1:1234".parse().unwrap());
-        let mut crdt = Crdt::new(me.clone());
+        let mut crdt = Crdt::new(me.clone()).expect("Crdt::new");
         assert_eq!(crdt.top_leader(), None);
         crdt.set_leader(leader0.id);
         assert_eq!(crdt.top_leader().unwrap(), leader0.id);
@@ -1648,4 +1881,41 @@ mod tests {
         crdt.update_leader();
         assert_eq!(crdt.my_data().leader_id, leader1.id);
     }
+
+    /// Validates the node that sent Protocol::ReceiveUpdates gets its
+    /// liveness updated, but not if the node sends Protocol::ReceiveUpdates
+    /// to itself.
+    #[test]
+    fn protocol_requestupdate_alive() {
+        logger::setup();
+        let window = default_window();
+        let recycler = BlobRecycler::default();
+
+        let node = NodeInfo::new_leader(&"127.0.0.1:1234".parse().unwrap());
+        let node_with_same_addr = NodeInfo::new_leader(&"127.0.0.1:1234".parse().unwrap());
+        assert_ne!(node.id, node_with_same_addr.id);
+        let node_with_diff_addr = NodeInfo::new_leader(&"127.0.0.1:4321".parse().unwrap());
+
+        let crdt = Crdt::new(node.clone()).expect("Crdt::new");
+        assert_eq!(crdt.alive.len(), 0);
+
+        let obj = Arc::new(RwLock::new(crdt));
+
+        let request = Protocol::RequestUpdates(1, node.clone());
+        assert!(Crdt::handle_protocol(request, &obj, &window, &recycler).is_none());
+
+        let request = Protocol::RequestUpdates(1, node_with_same_addr.clone());
+        assert!(Crdt::handle_protocol(request, &obj, &window, &recycler).is_none());
+
+        let request = Protocol::RequestUpdates(1, node_with_diff_addr.clone());
+        Crdt::handle_protocol(request, &obj, &window, &recycler);
+
+        let me = obj.write().unwrap();
+
+        // |node| and |node_with_same_addr| should not be in me.alive, but
+        // |node_with_diff_addr| should now be.
+        assert!(!me.alive.contains_key(&node.id));
+        assert!(!me.alive.contains_key(&node_with_same_addr.id));
+        assert!(me.alive[&node_with_diff_addr.id] > 0);
+    }
 }

src/drone.rs

@@ -16,6 +16,7 @@ use transaction::Transaction;
 
 pub const TIME_SLICE: u64 = 60;
 pub const REQUEST_CAP: u64 = 1_000_000;
+pub const DRONE_PORT: u16 = 9900;
 
 #[derive(Serialize, Deserialize, Debug, Clone, Copy)]
 pub enum DroneRequest {
@@ -112,6 +113,10 @@ impl Drone {
                 airdrop_request_amount,
                 client_public_key,
             } => {
+                info!(
+                    "Requesting airdrop of {} to {:?}",
+                    airdrop_request_amount, client_public_key
+                );
                 request_amount = airdrop_request_amount;
                 tx = Transaction::new(
                     &self.mint_keypair,
@@ -261,7 +266,7 @@ mod tests {
         const TPS_BATCH: i64 = 5_000_000;
 
         logger::setup();
-        let leader = TestNode::new();
+        let leader = TestNode::new_localhost();
 
         let alice = Mint::new(10_000_000);
         let bank = Bank::new(&alice);

src/entry_writer.rs

@@ -20,7 +20,8 @@ impl<'a, W: Write> EntryWriter<'a, W> {
 
     fn write_entry(writer: &mut W, entry: &Entry) -> io::Result<()> {
         let serialized = serde_json::to_string(entry).unwrap();
-        writeln!(writer, "{}", serialized)
+        writeln!(writer, "{}", serialized)?;
+        writer.flush()
     }
 
     pub fn write_entries<I>(writer: &mut W, entries: I) -> io::Result<()>

src/fullnode.rs

@@ -7,13 +7,12 @@ use entry_writer;
 use ledger::Block;
 use ncp::Ncp;
 use packet::BlobRecycler;
-use ring::rand::SystemRandom;
 use rpu::Rpu;
 use service::Service;
 use signature::{KeyPair, KeyPairUtil};
 use std::collections::VecDeque;
 use std::fs::{File, OpenOptions};
-use std::io::{sink, stdin, stdout, BufReader};
+use std::io::{stdin, stdout, BufReader};
 use std::io::{Read, Write};
 use std::net::SocketAddr;
 use std::sync::atomic::{AtomicBool, Ordering};
@@ -31,13 +30,8 @@ pub struct FullNode {
     thread_hdls: Vec<JoinHandle<()>>,
 }
 
-pub enum InFile {
-    StdIn,
-    Path(String),
-}
-
-pub enum OutFile {
-    StdOut,
+pub enum LedgerFile {
+    StdInOut,
     Path(String),
 }
 
@@ -50,11 +44,7 @@ pub struct Config {
 
 /// Structure to be replicated by the network
 impl Config {
-    pub fn new(bind_addr: &SocketAddr) -> Self {
-        let rnd = SystemRandom::new();
-        let pkcs8 = KeyPair::generate_pkcs8(&rnd)
-            .expect("generate_pkcs8 in mint pub fn new")
-            .to_vec();
+    pub fn new(bind_addr: &SocketAddr, pkcs8: Vec<u8>) -> Self {
         let keypair =
             KeyPair::from_pkcs8(Input::from(&pkcs8)).expect("from_pkcs8 in fullnode::Config new");
         let pubkey = keypair.pubkey();
@@ -71,16 +61,24 @@ impl FullNode {
     pub fn new(
         mut node: TestNode,
         leader: bool,
-        infile: InFile,
+        ledger: LedgerFile,
         keypair_for_validator: Option<KeyPair>,
         network_entry_for_validator: Option<SocketAddr>,
-        outfile_for_leader: Option<OutFile>,
     ) -> FullNode {
         info!("creating bank...");
         let bank = Bank::default();
-        let infile: Box<Read> = match infile {
-            InFile::Path(path) => Box::new(File::open(path).unwrap()),
-            InFile::StdIn => Box::new(stdin()),
+        let (infile, outfile): (Box<Read>, Box<Write + Send>) = match ledger {
+            LedgerFile::Path(path) => (
+                Box::new(File::open(path.clone()).expect("opening ledger file")),
+                Box::new(
+                    OpenOptions::new()
+                        .create(true)
+                        .append(true)
+                        .open(path)
+                        .expect("opening ledger file"),
+                ),
+            ),
+            LedgerFile::StdInOut => (Box::new(stdin()), Box::new(stdout())),
         };
         let reader = BufReader::new(infile);
         let entries = entry_writer::read_entries(reader).map(|e| e.expect("failed to parse entry"));
@@ -122,17 +120,6 @@ impl FullNode {
             server
         } else {
             node.data.leader_id = node.data.id;
-            let outfile_for_leader: Box<Write + Send> = match outfile_for_leader {
-                Some(OutFile::Path(file)) => Box::new(
-                    OpenOptions::new()
-                        .create(true)
-                        .append(true)
-                        .open(file)
-                        .expect("opening ledger file"),
-                ),
-                Some(OutFile::StdOut) => Box::new(stdout()),
-                None => Box::new(sink()),
-            };
 
             let server = FullNode::new_leader(
                 bank,
@@ -142,7 +129,7 @@ impl FullNode {
                 None,
                 node,
                 exit.clone(),
-                outfile_for_leader,
+                outfile,
             );
             info!(
                 "leader ready... local request address: {} (advertising {})",
@@ -208,7 +195,7 @@ impl FullNode {
         let bank = Arc::new(bank);
         let mut thread_hdls = vec![];
         let rpu = Rpu::new(
-            &bank.clone(),
+            &bank,
             node.sockets.requests,
             node.sockets.respond,
             exit.clone(),
@@ -216,20 +203,20 @@ impl FullNode {
         thread_hdls.extend(rpu.thread_hdls());
 
         let blob_recycler = BlobRecycler::default();
-        let crdt = Arc::new(RwLock::new(Crdt::new(node.data)));
+        let crdt = Arc::new(RwLock::new(Crdt::new(node.data).expect("Crdt::new")));
         let (tpu, blob_receiver) = Tpu::new(
-            &bank.clone(),
-            &crdt.clone(),
+            &bank,
+            &crdt,
             tick_duration,
             node.sockets.transaction,
-            &blob_recycler.clone(),
+            &blob_recycler,
             exit.clone(),
             writer,
         );
         thread_hdls.extend(tpu.thread_hdls());
         let window = FullNode::new_window(ledger_tail, entry_height, &crdt, &blob_recycler);
         let ncp = Ncp::new(
-            &crdt.clone(),
+            &crdt,
             window.clone(),
             node.sockets.gossip,
             node.sockets.gossip_send,
@@ -291,14 +278,14 @@ impl FullNode {
         let bank = Arc::new(bank);
         let mut thread_hdls = vec![];
         let rpu = Rpu::new(
-            &bank.clone(),
+            &bank,
             node.sockets.requests,
             node.sockets.respond,
             exit.clone(),
         );
         thread_hdls.extend(rpu.thread_hdls());
 
-        let crdt = Arc::new(RwLock::new(Crdt::new(node.data)));
+        let crdt = Arc::new(RwLock::new(Crdt::new(node.data).expect("Crdt::new")));
         crdt.write()
             .expect("'crdt' write lock before insert() in pub fn replicate")
             .insert(&entry_point);
@@ -308,7 +295,7 @@ impl FullNode {
         let window = FullNode::new_window(ledger_tail, entry_height, &crdt, &blob_recycler);
 
         let ncp = Ncp::new(
-            &crdt.clone(),
+            &crdt,
             window.clone(),
             node.sockets.gossip,
             node.sockets.gossip_send,
@@ -317,7 +304,7 @@ impl FullNode {
 
         let tvu = Tvu::new(
             keypair,
-            bank.clone(),
+            &bank,
             entry_height,
             crdt.clone(),
             window.clone(),
@@ -331,8 +318,12 @@ impl FullNode {
         FullNode { exit, thread_hdls }
     }
 
-    pub fn close(self) -> Result<()> {
+    //used for notifying many nodes in parallel to exit
+    pub fn exit(&self) {
         self.exit.store(true, Ordering::Relaxed);
+    }
+    pub fn close(self) -> Result<()> {
+        self.exit();
         self.join()
     }
 }
@@ -356,18 +347,39 @@ mod tests {
     use crdt::TestNode;
     use fullnode::FullNode;
     use mint::Mint;
+    use service::Service;
     use signature::{KeyPair, KeyPairUtil};
     use std::sync::atomic::AtomicBool;
     use std::sync::Arc;
     #[test]
     fn validator_exit() {
         let kp = KeyPair::new();
-        let tn = TestNode::new_with_pubkey(kp.pubkey());
+        let tn = TestNode::new_localhost_with_pubkey(kp.pubkey());
         let alice = Mint::new(10_000);
         let bank = Bank::new(&alice);
         let exit = Arc::new(AtomicBool::new(false));
         let entry = tn.data.clone();
         let v = FullNode::new_validator(kp, bank, 0, None, tn, &entry, exit);
-        v.close().unwrap();
+        v.exit();
+        v.join().unwrap();
+    }
+    #[test]
+    fn validator_parallel_exit() {
+        let vals: Vec<FullNode> = (0..2)
+            .map(|_| {
+                let kp = KeyPair::new();
+                let tn = TestNode::new_localhost_with_pubkey(kp.pubkey());
+                let alice = Mint::new(10_000);
+                let bank = Bank::new(&alice);
+                let exit = Arc::new(AtomicBool::new(false));
+                let entry = tn.data.clone();
+                FullNode::new_validator(kp, bank, 0, None, tn, &entry, exit)
+            })
+            .collect();
+        //each validator can exit in parallel to speed many sequential calls to `join`
+        vals.iter().for_each(|v| v.exit());
+        //while join is called sequentially, the above exit call notified all the
+        //validators to exit from all their threads
+        vals.into_iter().for_each(|v| v.join().unwrap());
     }
 }

src/metrics.rs

@@ -57,7 +57,7 @@ impl InfluxDbMetricsWriter {
 impl MetricsWriter for InfluxDbMetricsWriter {
     fn write(&self, points: Vec<influxdb::Point>) {
         if let Some(ref client) = self.client {
-            info!("submitting {} points", points.len());
+            debug!("submitting {} points", points.len());
             if let Err(err) = client.write_points(
                 influxdb::Points { point: points },
                 Some(influxdb::Precision::Milliseconds),
@@ -184,6 +184,50 @@ pub fn flush() {
     agent.flush();
 }
 
+/// Hook the panic handler to generate a data point on each panic
+pub fn set_panic_hook(program: &'static str) {
+    use std::panic;
+    use std::sync::{Once, ONCE_INIT};
+    static SET_HOOK: Once = ONCE_INIT;
+    SET_HOOK.call_once(|| {
+        let default_hook = panic::take_hook();
+        panic::set_hook(Box::new(move |ono| {
+            default_hook(ono);
+            submit(
+                influxdb::Point::new("panic")
+                    .add_tag("program", influxdb::Value::String(program.to_string()))
+                    .add_tag(
+                        "thread",
+                        influxdb::Value::String(
+                            thread::current().name().unwrap_or("?").to_string(),
+                        ),
+                    )
+                    // The 'one' field exists to give Kapacitor Alerts a numerical value
+                    // to filter on
+                    .add_field("one", influxdb::Value::Integer(1))
+                    .add_field(
+                        "message",
+                        influxdb::Value::String(
+                            // TODO: use ono.message() when it becomes stable
+                            ono.to_string(),
+                        ),
+                    )
+                    .add_field(
+                        "location",
+                        influxdb::Value::String(match ono.location() {
+                            Some(location) => location.to_string(),
+                            None => "?".to_string(),
+                        }),
+                    )
+                    .to_owned(),
+            );
+            // Flush metrics immediately in case the process exits immediately
+            // upon return
+            flush();
+        }));
+    });
+}
+
 #[cfg(test)]
 mod test {
     use super::*;

src/mint.rs

@@ -15,11 +15,7 @@ pub struct Mint {
 }
 
 impl Mint {
-    pub fn new(tokens: i64) -> Self {
-        let rnd = SystemRandom::new();
-        let pkcs8 = KeyPair::generate_pkcs8(&rnd)
-            .expect("generate_pkcs8 in mint pub fn new")
-            .to_vec();
+    pub fn new_with_pkcs8(tokens: i64, pkcs8: Vec<u8>) -> Self {
         let keypair =
             KeyPair::from_pkcs8(Input::from(&pkcs8)).expect("from_pkcs8 in mint pub fn new");
         let pubkey = keypair.pubkey();
@@ -29,6 +25,15 @@ impl Mint {
             tokens,
         }
     }
+
+    pub fn new(tokens: i64) -> Self {
+        let rnd = SystemRandom::new();
+        let pkcs8 = KeyPair::generate_pkcs8(&rnd)
+            .expect("generate_pkcs8 in mint pub fn new")
+            .to_vec();
+        Self::new_with_pkcs8(tokens, pkcs8)
+    }
+
     pub fn seed(&self) -> Hash {
         hash(&self.pkcs8)
     }

src/nat.rs

@@ -106,9 +106,13 @@ pub fn udp_public_bind(label: &str, startport: u16, endport: u16) -> UdpSocketPa
         Err(_) => {
             let sender = udp_random_bind(startport, endport, 5).unwrap();
             let local_addr = sender.local_addr().unwrap();
-            info!("Using local address {} for {}", local_addr, label);
+
+            let pub_ip = get_public_ip_addr().unwrap();
+            let pub_addr = SocketAddr::new(pub_ip, local_addr.port());
+
+            info!("Using source address {} for {}", pub_addr, label);
             UdpSocketPair {
-                addr: private_addr,
+                addr: pub_addr,
                 receiver: sender.try_clone().unwrap(),
                 sender,
             }

src/ncp.rs

@@ -88,12 +88,12 @@ mod tests {
     // test that stage will exit when flag is set
     fn test_exit() {
         let exit = Arc::new(AtomicBool::new(false));
-        let tn = TestNode::new();
-        let crdt = Crdt::new(tn.data.clone());
+        let tn = TestNode::new_localhost();
+        let crdt = Crdt::new(tn.data.clone()).expect("Crdt::new");
         let c = Arc::new(RwLock::new(crdt));
         let w = Arc::new(RwLock::new(vec![]));
         let d = Ncp::new(
-            &c.clone(),
+            &c,
             w,
             tn.sockets.gossip,
             tn.sockets.gossip_send,

src/packet.rs

@@ -19,7 +19,6 @@ pub type SharedBlobs = VecDeque<SharedBlob>;
 pub type PacketRecycler = Recycler<Packets>;
 pub type BlobRecycler = Recycler<Blob>;
 
-const LOG_RATE: usize = 10;
 pub const NUM_PACKETS: usize = 1024 * 8;
 pub const BLOB_SIZE: usize = 64 * 1024;
 pub const BLOB_DATA_SIZE: usize = BLOB_SIZE - BLOB_HEADER_SIZE;
@@ -188,7 +187,6 @@ impl<T: Default> Recycler<T> {
 
 impl Packets {
     fn run_read_from(&mut self, socket: &UdpSocket) -> Result<usize> {
-        static mut COUNTER: Counter = create_counter!("packets", LOG_RATE);
         self.packets.resize(NUM_PACKETS, Packet::default());
         let mut i = 0;
         //DOCUMENTED SIDE-EFFECT
@@ -203,7 +201,7 @@ impl Packets {
             trace!("receiving on {}", socket.local_addr().unwrap());
             match socket.recv_from(&mut p.data) {
                 Err(_) if i > 0 => {
-                    inc_counter!(COUNTER, i);
+                    inc_new_counter!("packets-recv_count", 1);
                     debug!("got {:?} messages on {}", i, socket.local_addr().unwrap());
                     break;
                 }
@@ -214,6 +212,7 @@ impl Packets {
                 Ok((nrecv, from)) => {
                     p.meta.size = nrecv;
                     p.meta.set_addr(&from);
+                    trace!("got {} bytes from {}", nrecv, from);
                     if i == 0 {
                         socket.set_nonblocking(true)?;
                     }
@@ -407,6 +406,7 @@ impl Blob {
                     Ok((nrecv, from)) => {
                         p.meta.size = nrecv;
                         p.meta.set_addr(&from);
+                        trace!("got {} bytes from {}", nrecv, from);
                         if i == 0 {
                             socket.set_nonblocking(true)?;
                         }
@@ -423,7 +423,7 @@ impl Blob {
                 let p = r.read().expect("'r' read lock in pub fn send_to");
                 let a = p.meta.addr();
                 if let Err(e) = socket.send_to(&p.data[..p.meta.size], &a) {
-                    info!(
+                    warn!(
                         "error sending {} byte packet to {:?}: {:?}",
                         p.meta.size, a, e
                     );

src/replicate_stage.rs

@@ -27,7 +27,6 @@ pub struct ReplicateStage {
 }
 
 const VOTE_TIMEOUT_MS: u64 = 1000;
-const LOG_RATE: usize = 10;
 
 impl ReplicateStage {
     /// Process entry blobs, already in order
@@ -48,28 +47,29 @@ impl ReplicateStage {
         }
         let blobs_len = blobs.len();
         let entries = ledger::reconstruct_entries_from_blobs(blobs.clone())?;
-        let votes = entries_to_votes(&entries);
-
-        static mut COUNTER_REPLICATE: Counter = create_counter!("replicate-transactions", LOG_RATE);
-        inc_counter!(
-            COUNTER_REPLICATE,
+        {
+            let votes = entries_to_votes(&entries);
+            let mut wcrdt = crdt.write().unwrap();
+            wcrdt.insert_votes(&votes);
+        };
+        inc_new_counter!(
+            "replicate-transactions",
             entries.iter().map(|x| x.transactions.len()).sum()
         );
         let res = bank.process_entries(entries);
         if res.is_err() {
             error!("process_entries {} {:?}", blobs_len, res);
         }
+        let _ = res?;
         let now = timing::timestamp();
         if now - *last_vote > VOTE_TIMEOUT_MS {
-            let height = res?;
             let last_id = bank.last_id();
             let shared_blob = blob_recycler.allocate();
             let (vote, addr) = {
                 let mut wcrdt = crdt.write().unwrap();
-                wcrdt.insert_votes(&votes);
                 //TODO: doesn't seem like there is a synchronous call to get height and id
-                info!("replicate_stage {} {:?}", height, &last_id[..8]);
-                wcrdt.new_vote(height, last_id)
+                info!("replicate_stage {:?}", &last_id[..8]);
+                wcrdt.new_vote(last_id)
             }?;
             {
                 let mut blob = shared_blob.write().unwrap();
@@ -80,7 +80,9 @@ impl ReplicateStage {
                 blob.meta.set_addr(&addr);
                 blob.meta.size = len;
             }
+            inc_new_counter!("replicate-vote_sent", 1);
             *last_vote = now;
+
             vote_blob_sender.send(VecDeque::from(vec![shared_blob]))?;
         }
         while let Some(blob) = blobs.pop_front() {

src/signature.rs

@@ -8,8 +8,11 @@ use ring::error::Unspecified;
 use ring::rand::SecureRandom;
 use ring::signature::Ed25519KeyPair;
 use ring::{rand, signature};
+use serde_json;
 use std::cell::RefCell;
-use untrusted;
+use std::error;
+use std::fs::File;
+use untrusted::Input;
 
 pub type KeyPair = Ed25519KeyPair;
 pub type PublicKey = GenericArray<u8, U32>;
@@ -24,10 +27,8 @@ impl KeyPairUtil for Ed25519KeyPair {
     /// Return a new ED25519 keypair
     fn new() -> Self {
         let rng = rand::SystemRandom::new();
-        let pkcs8_bytes = signature::Ed25519KeyPair::generate_pkcs8(&rng)
-            .expect("generate_pkcs8 in signature pb fn new");
-        signature::Ed25519KeyPair::from_pkcs8(untrusted::Input::from(&pkcs8_bytes))
-            .expect("from_pcks8 in signature pb fn new")
+        let pkcs8_bytes = Ed25519KeyPair::generate_pkcs8(&rng).expect("generate_pkcs8");
+        Ed25519KeyPair::from_pkcs8(Input::from(&pkcs8_bytes)).expect("from_pcks8")
     }
 
     /// Return the public key for the given keypair
@@ -42,9 +43,9 @@ pub trait SignatureUtil {
 
 impl SignatureUtil for GenericArray<u8, U64> {
     fn verify(&self, peer_public_key_bytes: &[u8], msg_bytes: &[u8]) -> bool {
-        let peer_public_key = untrusted::Input::from(peer_public_key_bytes);
-        let msg = untrusted::Input::from(msg_bytes);
-        let sig = untrusted::Input::from(self);
+        let peer_public_key = Input::from(peer_public_key_bytes);
+        let msg = Input::from(msg_bytes);
+        let sig = Input::from(self);
         signature::verify(&signature::ED25519, peer_public_key, msg, sig).is_ok()
     }
 }
@@ -77,7 +78,7 @@ impl GenKeys {
             .into_par_iter()
             .map(|seed| {
                 let pkcs8 = GenKeys::new(seed).new_key();
-                KeyPair::from_pkcs8(untrusted::Input::from(&pkcs8)).unwrap()
+                KeyPair::from_pkcs8(Input::from(&pkcs8)).unwrap()
             })
             .collect()
     }
@@ -91,6 +92,18 @@ impl SecureRandom for GenKeys {
     }
 }
 
+pub fn read_pkcs8(path: &str) -> Result<Vec<u8>, Box<error::Error>> {
+    let file = File::open(path.to_string())?;
+    let pkcs8: Vec<u8> = serde_json::from_reader(file)?;
+    Ok(pkcs8)
+}
+
+pub fn read_keypair(path: &str) -> Result<KeyPair, Box<error::Error>> {
+    let pkcs8 = read_pkcs8(path)?;
+    let keypair = Ed25519KeyPair::from_pkcs8(Input::from(&pkcs8))?;
+    Ok(keypair)
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;

src/sigverify.rs

@@ -22,6 +22,8 @@ struct Elems {
 #[cfg(feature = "cuda")]
 #[link(name = "cuda_verify_ed25519")]
 extern "C" {
+    fn ed25519_init() -> bool;
+    fn ed25519_set_verbose(val: bool);
     fn ed25519_verify_many(
         vecs: *const Elems,
         num: u32,          //number of vecs
@@ -34,6 +36,11 @@ extern "C" {
     ) -> u32;
 }
 
+#[cfg(not(feature = "cuda"))]
+pub fn init() {
+    // stub
+}
+
 #[cfg(not(feature = "cuda"))]
 fn verify_packet(packet: &Packet) -> u8 {
     use ring::signature;
@@ -70,7 +77,6 @@ fn batch_size(batches: &[SharedPackets]) -> usize {
 #[cfg(not(feature = "cuda"))]
 pub fn ed25519_verify(batches: &Vec<SharedPackets>) -> Vec<Vec<u8>> {
     use rayon::prelude::*;
-    static mut COUNTER: Counter = create_counter!("ed25519_verify", 1);
     let count = batch_size(batches);
     info!("CPU ECDSA for {}", batch_size(batches));
     let rv = batches
@@ -84,14 +90,24 @@ pub fn ed25519_verify(batches: &Vec<SharedPackets>) -> Vec<Vec<u8>> {
                 .collect()
         })
         .collect();
-    inc_counter!(COUNTER, count);
+    inc_new_counter!("ed25519_verify", count);
     rv
 }
 
+#[cfg(feature = "cuda")]
+pub fn init() {
+    unsafe {
+        ed25519_set_verbose(true);
+        if !ed25519_init() {
+            panic!("ed25519_init() failed");
+        }
+        ed25519_set_verbose(false);
+    }
+}
+
 #[cfg(feature = "cuda")]
 pub fn ed25519_verify(batches: &Vec<SharedPackets>) -> Vec<Vec<u8>> {
     use packet::PACKET_DATA_SIZE;
-    static mut COUNTER: Counter = create_counter!("ed25519_verify_cuda", 1);
     let count = batch_size(batches);
     info!("CUDA ECDSA for {}", batch_size(batches));
     let mut out = Vec::new();
@@ -151,7 +167,7 @@ pub fn ed25519_verify(batches: &Vec<SharedPackets>) -> Vec<Vec<u8>> {
             num += 1;
         }
     }
-    inc_counter!(COUNTER, count);
+    inc_new_counter!("ed25519_verify", count);
     rvs
 }
 

src/sigverify_stage.rs

@@ -25,6 +25,7 @@ pub struct SigVerifyStage {
 
 impl SigVerifyStage {
     pub fn new(packet_receiver: Receiver<SharedPackets>) -> (Self, Receiver<VerifiedPackets>) {
+        sigverify::init();
         let (verified_sender, verified_receiver) = channel();
         let thread_hdls = Self::verifier_services(packet_receiver, verified_sender);
         (SigVerifyStage { thread_hdls }, verified_receiver)
@@ -77,7 +78,7 @@ impl SigVerifyStage {
         verified_sender: Arc<Mutex<Sender<VerifiedPackets>>>,
     ) -> JoinHandle<()> {
         spawn(move || loop {
-            if let Err(e) = Self::verifier(&packet_receiver.clone(), &verified_sender.clone()) {
+            if let Err(e) = Self::verifier(&packet_receiver, &verified_sender) {
                 match e {
                     Error::RecvTimeoutError(RecvTimeoutError::Disconnected) => break,
                     Error::RecvTimeoutError(RecvTimeoutError::Timeout) => (),

src/streamer.rs

@@ -18,7 +18,6 @@ use std::sync::{Arc, RwLock};
 use std::thread::{Builder, JoinHandle};
 use std::time::Duration;
 
-const LOG_RATE: usize = 10;
 pub const WINDOW_SIZE: u64 = 2 * 1024;
 pub type PacketReceiver = Receiver<SharedPackets>;
 pub type PacketSender = Sender<SharedPackets>;
@@ -117,7 +116,7 @@ pub fn responder(
                 match e {
                     Error::RecvTimeoutError(RecvTimeoutError::Disconnected) => break,
                     Error::RecvTimeoutError(RecvTimeoutError::Timeout) => (),
-                    _ => error!("{} responder error: {:?}", name, e),
+                    _ => warn!("{} responder error: {:?}", name, e),
                 }
             }
         })
@@ -127,7 +126,7 @@ pub fn responder(
 //TODO, we would need to stick block authentication before we create the
 //window.
 fn recv_blobs(recycler: &BlobRecycler, sock: &UdpSocket, s: &BlobSender) -> Result<()> {
-    trace!("receiving on {}", sock.local_addr().unwrap());
+    trace!("recv_blobs: receiving on {}", sock.local_addr().unwrap());
     let dq = Blob::recv_from(recycler, sock)?;
     if !dq.is_empty() {
         s.send(dq)?;
@@ -224,9 +223,7 @@ fn repair_window(
     let reqs = find_next_missing(locked_window, crdt, consumed, received)?;
     trace!("{:x}: repair_window missing: {}", debug_id, reqs.len());
     if !reqs.is_empty() {
-        static mut COUNTER_REPAIR: Counter =
-            create_counter!("streamer-repair_window-repair", LOG_RATE);
-        inc_counter!(COUNTER_REPAIR, reqs.len());
+        inc_new_counter!("streamer-repair_window-repair", reqs.len());
         debug!(
             "{:x}: repair_window counter times: {} consumed: {} received: {} missing: {}",
             debug_id,
@@ -301,9 +298,7 @@ fn retransmit_all_leader_blocks(
             *received,
             retransmit_queue.len(),
         );
-        static mut COUNTER_RETRANSMIT: Counter =
-            create_counter!("streamer-recv_window-retransmit", LOG_RATE);
-        inc_counter!(COUNTER_RETRANSMIT, retransmit_queue.len());
+        inc_new_counter!("streamer-recv_window-retransmit", retransmit_queue.len());
         retransmit.send(retransmit_queue)?;
     }
     Ok(())
@@ -413,8 +408,7 @@ fn recv_window(
     while let Ok(mut nq) = r.try_recv() {
         dq.append(&mut nq)
     }
-    static mut COUNTER_RECV: Counter = create_counter!("streamer-recv_window-recv", LOG_RATE);
-    inc_counter!(COUNTER_RECV, dq.len());
+    inc_new_counter!("streamer-recv_window-recv", dq.len());
     debug!(
         "{:x}: RECV_WINDOW {} {}: got packets {}",
         debug_id,
@@ -480,9 +474,7 @@ fn recv_window(
             consume_queue.len(),
         );
         trace!("sending consume_queue.len: {}", consume_queue.len());
-        static mut COUNTER_CONSUME: Counter =
-            create_counter!("streamer-recv_window-consume", LOG_RATE);
-        inc_counter!(COUNTER_CONSUME, consume_queue.len());
+        inc_new_counter!("streamer-recv_window-consume", consume_queue.len());
         s.send(consume_queue)?;
     }
     Ok(())
@@ -591,7 +583,10 @@ pub fn window(
                     match e {
                         Error::RecvTimeoutError(RecvTimeoutError::Disconnected) => break,
                         Error::RecvTimeoutError(RecvTimeoutError::Timeout) => (),
-                        _ => error!("window error: {:?}", e),
+                        _ => {
+                            inc_new_counter!("streamer-window-error", 1, 1);
+                            error!("window error: {:?}", e);
+                        }
                     }
                 }
                 let _ = repair_window(
@@ -647,9 +642,7 @@ fn broadcast(
         // Index the blobs
         Crdt::index_blobs(&me, &blobs, receive_index)?;
         // keep the cache of blobs that are broadcast
-        static mut COUNTER_BROADCAST: Counter =
-            create_counter!("streamer-broadcast-sent", LOG_RATE);
-        inc_counter!(COUNTER_BROADCAST, blobs.len());
+        inc_new_counter!("streamer-broadcast-sent", blobs.len());
         {
             let mut win = window.write().unwrap();
             assert!(blobs.len() <= win.len());
@@ -738,8 +731,11 @@ pub fn broadcaster(
                     match e {
                         Error::RecvTimeoutError(RecvTimeoutError::Disconnected) => break,
                         Error::RecvTimeoutError(RecvTimeoutError::Timeout) => (),
-                        Error::CrdtError(CrdtError::TooSmall) => (), // TODO: Why are the unit-tests throwing hundreds of these?
-                        _ => error!("broadcaster error: {:?}", e),
+                        Error::CrdtError(CrdtError::NoPeers) => (), // TODO: Why are the unit-tests throwing hundreds of these?
+                        _ => {
+                            inc_new_counter!("streamer-broadcaster-error", 1, 1);
+                            error!("broadcaster error: {:?}", e);
+                        }
                     }
                 }
             }
@@ -792,7 +788,10 @@ pub fn retransmitter(
                     match e {
                         Error::RecvTimeoutError(RecvTimeoutError::Disconnected) => break,
                         Error::RecvTimeoutError(RecvTimeoutError::Timeout) => (),
-                        _ => error!("retransmitter error: {:?}", e),
+                        _ => {
+                            inc_new_counter!("streamer-retransmit-error", 1, 1);
+                            error!("retransmitter error: {:?}", e);
+                        }
                     }
                 }
             }
@@ -899,9 +898,9 @@ mod test {
     #[test]
     pub fn window_send_test() {
         logger::setup();
-        let tn = TestNode::new();
+        let tn = TestNode::new_localhost();
         let exit = Arc::new(AtomicBool::new(false));
-        let mut crdt_me = Crdt::new(tn.data.clone());
+        let mut crdt_me = Crdt::new(tn.data.clone()).expect("Crdt::new");
         let me_id = crdt_me.my_data().id;
         crdt_me.set_leader(me_id);
         let subs = Arc::new(RwLock::new(crdt_me));

src/thin_client.rs

@@ -274,7 +274,7 @@ mod tests {
     #[test]
     fn test_thin_client() {
         logger::setup();
-        let leader = TestNode::new();
+        let leader = TestNode::new_localhost();
         let leader_data = leader.data.clone();
 
         let alice = Mint::new(10_000);
@@ -317,7 +317,7 @@ mod tests {
     #[ignore]
     fn test_bad_sig() {
         logger::setup();
-        let leader = TestNode::new();
+        let leader = TestNode::new_localhost();
         let alice = Mint::new(10_000);
         let bank = Bank::new(&alice);
         let bob_pubkey = KeyPair::new().pubkey();
@@ -371,7 +371,7 @@ mod tests {
     #[test]
     fn test_client_check_signature() {
         logger::setup();
-        let leader = TestNode::new();
+        let leader = TestNode::new_localhost();
         let alice = Mint::new(10_000);
         let bank = Bank::new(&alice);
         let bob_pubkey = KeyPair::new().pubkey();

src/tpu.rs

@@ -63,7 +63,7 @@ impl Tpu {
         let packet_recycler = PacketRecycler::default();
 
         let (fetch_stage, packet_receiver) =
-            FetchStage::new(transactions_socket, exit, &packet_recycler.clone());
+            FetchStage::new(transactions_socket, exit, &packet_recycler);
 
         let (sigverify_stage, verified_receiver) = SigVerifyStage::new(packet_receiver);
 

src/tvu.rs

@@ -70,7 +70,7 @@ impl Tvu {
     /// * `exit` - The exit signal.
     pub fn new(
         keypair: KeyPair,
-        bank: Arc<Bank>,
+        bank: &Arc<Bank>,
         entry_height: u64,
         crdt: Arc<RwLock<Crdt>>,
         window: Window,
@@ -83,22 +83,27 @@ impl Tvu {
         let (fetch_stage, blob_fetch_receiver) = BlobFetchStage::new_multi_socket(
             vec![replicate_socket, repair_socket],
             exit,
-            &blob_recycler.clone(),
+            &blob_recycler,
         );
         //TODO
         //the packets coming out of blob_receiver need to be sent to the GPU and verified
         //then sent to the window, which does the erasure coding reconstruction
         let (window_stage, blob_window_receiver) = WindowStage::new(
-            &crdt.clone(),
+            &crdt,
             window,
             entry_height,
             retransmit_socket,
-            &blob_recycler.clone(),
+            &blob_recycler,
             blob_fetch_receiver,
         );
 
-        let replicate_stage =
-            ReplicateStage::new(keypair, bank, crdt, blob_recycler, blob_window_receiver);
+        let replicate_stage = ReplicateStage::new(
+            keypair,
+            bank.clone(),
+            crdt,
+            blob_recycler,
+            blob_window_receiver,
+        );
 
         Tvu {
             replicate_stage,
@@ -168,21 +173,21 @@ pub mod tests {
     #[test]
     fn test_replicate() {
         logger::setup();
-        let leader = TestNode::new();
+        let leader = TestNode::new_localhost();
         let target1_kp = KeyPair::new();
-        let target1 = TestNode::new_with_pubkey(target1_kp.pubkey());
-        let target2 = TestNode::new();
+        let target1 = TestNode::new_localhost_with_pubkey(target1_kp.pubkey());
+        let target2 = TestNode::new_localhost();
         let exit = Arc::new(AtomicBool::new(false));
 
         //start crdt_leader
-        let mut crdt_l = Crdt::new(leader.data.clone());
+        let mut crdt_l = Crdt::new(leader.data.clone()).expect("Crdt::new");
         crdt_l.set_leader(leader.data.id);
 
         let cref_l = Arc::new(RwLock::new(crdt_l));
         let dr_l = new_ncp(cref_l, leader.sockets.gossip, exit.clone()).unwrap();
 
         //start crdt2
-        let mut crdt2 = Crdt::new(target2.data.clone());
+        let mut crdt2 = Crdt::new(target2.data.clone()).expect("Crdt::new");
         crdt2.insert(&leader.data);
         crdt2.set_leader(leader.data.id);
         let leader_id = leader.data.id;
@@ -217,7 +222,7 @@ pub mod tests {
         let bank = Arc::new(Bank::new(&mint));
 
         //start crdt1
-        let mut crdt1 = Crdt::new(target1.data.clone());
+        let mut crdt1 = Crdt::new(target1.data.clone()).expect("Crdt::new");
         crdt1.insert(&leader.data);
         crdt1.set_leader(leader.data.id);
         let cref1 = Arc::new(RwLock::new(crdt1));
@@ -225,7 +230,7 @@ pub mod tests {
 
         let tvu = Tvu::new(
             target1_kp,
-            bank.clone(),
+            &bank,
             0,
             cref1,
             dr_1.1,

src/write_stage.rs

@@ -3,6 +3,7 @@
 //! stdout, and then sends the Entry to its output channel.
 
 use bank::Bank;
+use counter::Counter;
 use crdt::Crdt;
 use entry::Entry;
 use entry_writer::EntryWriter;
@@ -12,6 +13,7 @@ use result::{Error, Result};
 use service::Service;
 use std::collections::VecDeque;
 use std::io::Write;
+use std::sync::atomic::AtomicUsize;
 use std::sync::mpsc::{channel, Receiver, RecvTimeoutError};
 use std::sync::{Arc, RwLock};
 use std::thread::{self, Builder, JoinHandle};
@@ -41,6 +43,8 @@ impl WriteStage {
         let mut blobs = VecDeque::new();
         entries.to_blobs(blob_recycler, &mut blobs);
         if !blobs.is_empty() {
+            inc_new_counter!("write_stage-broadcast_vote-count", votes.len());
+            inc_new_counter!("write_stage-broadcast_blobs-count", blobs.len());
             trace!("broadcasting {}", blobs.len());
             blob_sender.send(blobs)?;
         }
@@ -71,7 +75,10 @@ impl WriteStage {
                         match e {
                             Error::RecvTimeoutError(RecvTimeoutError::Disconnected) => break,
                             Error::RecvTimeoutError(RecvTimeoutError::Timeout) => (),
-                            _ => error!("{:?}", e),
+                            _ => {
+                                inc_new_counter!("write_stage-error", 1);
+                                error!("{:?}", e);
+                            }
                         }
                     };
                 }

tests/data_replicator.rs

@@ -16,8 +16,8 @@ use std::thread::sleep;
 use std::time::Duration;
 
 fn test_node(exit: Arc<AtomicBool>) -> (Arc<RwLock<Crdt>>, Ncp, UdpSocket) {
-    let tn = TestNode::new();
-    let crdt = Crdt::new(tn.data.clone());
+    let tn = TestNode::new_localhost();
+    let crdt = Crdt::new(tn.data.clone()).expect("Crdt::new");
     let c = Arc::new(RwLock::new(crdt));
     let w = Arc::new(RwLock::new(vec![]));
     let d = Ncp::new(

tests/multinode.rs

@@ -7,10 +7,11 @@ extern crate solana;
 use solana::crdt::TestNode;
 use solana::crdt::{Crdt, NodeInfo};
 use solana::entry_writer::EntryWriter;
-use solana::fullnode::{FullNode, InFile, OutFile};
+use solana::fullnode::{FullNode, LedgerFile};
 use solana::logger;
 use solana::mint::Mint;
 use solana::ncp::Ncp;
+use solana::service::Service;
 use solana::signature::{KeyPair, KeyPairUtil, PublicKey};
 use solana::streamer::default_window;
 use solana::thin_client::ThinClient;
@@ -24,18 +25,18 @@ use std::time::Duration;
 fn converge(leader: &NodeInfo, num_nodes: usize) -> Vec<NodeInfo> {
     //lets spy on the network
     let exit = Arc::new(AtomicBool::new(false));
-    let mut spy = TestNode::new();
+    let mut spy = TestNode::new_localhost();
     let daddr = "0.0.0.0:0".parse().unwrap();
     let me = spy.data.id.clone();
     spy.data.contact_info.tvu = daddr;
     spy.data.contact_info.rpu = daddr;
-    let mut spy_crdt = Crdt::new(spy.data);
+    let mut spy_crdt = Crdt::new(spy.data).expect("Crdt::new");
     spy_crdt.insert(&leader);
     spy_crdt.set_leader(leader.id);
     let spy_ref = Arc::new(RwLock::new(spy_crdt));
     let spy_window = default_window();
     let ncp = Ncp::new(
-        &spy_ref.clone(),
+        &spy_ref,
         spy_window,
         spy.sockets.gossip,
         spy.sockets.gossip_send,
@@ -86,7 +87,7 @@ fn test_multi_node_validator_catchup_from_zero() {
     logger::setup();
     const N: usize = 5;
     trace!("test_multi_node_validator_catchup_from_zero");
-    let leader = TestNode::new();
+    let leader = TestNode::new_localhost();
     let leader_data = leader.data.clone();
     let bob_pubkey = KeyPair::new().pubkey();
 
@@ -94,22 +95,20 @@ fn test_multi_node_validator_catchup_from_zero() {
     let server = FullNode::new(
         leader,
         true,
-        InFile::Path(ledger_path.clone()),
-        None,
+        LedgerFile::Path(ledger_path.clone()),
         None,
         None,
     );
     let mut nodes = vec![server];
     for _ in 0..N {
         let keypair = KeyPair::new();
-        let validator = TestNode::new_with_pubkey(keypair.pubkey());
+        let validator = TestNode::new_localhost_with_pubkey(keypair.pubkey());
         let mut val = FullNode::new(
             validator,
             false,
-            InFile::Path(ledger_path.clone()),
+            LedgerFile::Path(ledger_path.clone()),
             Some(keypair),
             Some(leader_data.contact_info.ncp),
-            None,
         );
         nodes.push(val);
     }
@@ -137,14 +136,13 @@ fn test_multi_node_validator_catchup_from_zero() {
     success = 0;
     // start up another validator, converge and then check everyone's balances
     let keypair = KeyPair::new();
-    let validator = TestNode::new_with_pubkey(keypair.pubkey());
+    let validator = TestNode::new_localhost_with_pubkey(keypair.pubkey());
     let val = FullNode::new(
         validator,
         false,
-        InFile::Path(ledger_path.clone()),
+        LedgerFile::Path(ledger_path.clone()),
         Some(keypair),
         Some(leader_data.contact_info.ncp),
-        None,
     );
     nodes.push(val);
     //contains the leader and new node
@@ -189,29 +187,27 @@ fn test_multi_node_basic() {
     logger::setup();
     const N: usize = 5;
     trace!("test_multi_node_basic");
-    let leader = TestNode::new();
+    let leader = TestNode::new_localhost();
     let leader_data = leader.data.clone();
     let bob_pubkey = KeyPair::new().pubkey();
     let (alice, ledger_path) = genesis(10_000);
     let server = FullNode::new(
         leader,
         true,
-        InFile::Path(ledger_path.clone()),
-        None,
+        LedgerFile::Path(ledger_path.clone()),
         None,
         None,
     );
     let mut nodes = vec![server];
     for _ in 0..N {
         let keypair = KeyPair::new();
-        let validator = TestNode::new_with_pubkey(keypair.pubkey());
+        let validator = TestNode::new_localhost_with_pubkey(keypair.pubkey());
         let val = FullNode::new(
             validator,
             false,
-            InFile::Path(ledger_path.clone()),
+            LedgerFile::Path(ledger_path.clone()),
             Some(keypair),
             Some(leader_data.contact_info.ncp),
-            None,
         );
         nodes.push(val);
     }
@@ -244,17 +240,16 @@ fn test_multi_node_basic() {
 #[test]
 fn test_boot_validator_from_file() {
     logger::setup();
-    let leader = TestNode::new();
+    let leader = TestNode::new_localhost();
     let bob_pubkey = KeyPair::new().pubkey();
     let (alice, ledger_path) = genesis(100_000);
     let leader_data = leader.data.clone();
     let leader_fullnode = FullNode::new(
         leader,
         true,
-        InFile::Path(ledger_path.clone()),
+        LedgerFile::Path(ledger_path.clone()),
         None,
         None,
-        Some(OutFile::Path(ledger_path.clone())),
     );
     let leader_balance =
         send_tx_and_retry_get_balance(&leader_data, &alice, &bob_pubkey, Some(500)).unwrap();
@@ -264,17 +259,15 @@ fn test_boot_validator_from_file() {
     assert_eq!(leader_balance, 1000);
 
     let keypair = KeyPair::new();
-    let validator = TestNode::new_with_pubkey(keypair.pubkey());
+    let validator = TestNode::new_localhost_with_pubkey(keypair.pubkey());
     let validator_data = validator.data.clone();
     let val_fullnode = FullNode::new(
         validator,
         false,
-        InFile::Path(ledger_path.clone()),
+        LedgerFile::Path(ledger_path.clone()),
         Some(keypair),
         Some(leader_data.contact_info.ncp),
-        None,
     );
-
     let mut client = mk_client(&validator_data);
     let getbal = retry_get_balance(&mut client, &bob_pubkey, Some(leader_balance));
     assert!(getbal == Some(leader_balance));
@@ -285,15 +278,14 @@ fn test_boot_validator_from_file() {
 }
 
 fn create_leader(ledger_path: &str) -> (NodeInfo, FullNode) {
-    let leader = TestNode::new();
+    let leader = TestNode::new_localhost();
     let leader_data = leader.data.clone();
     let leader_fullnode = FullNode::new(
         leader,
         true,
-        InFile::Path(ledger_path.to_string()),
+        LedgerFile::Path(ledger_path.to_string()),
         None,
         None,
-        Some(OutFile::Path(ledger_path.to_string())),
     );
     (leader_data, leader_fullnode)
 }
@@ -337,15 +329,14 @@ fn test_leader_restart_validator_start_from_old_ledger() {
 
     // start validator from old ledger
     let keypair = KeyPair::new();
-    let validator = TestNode::new_with_pubkey(keypair.pubkey());
+    let validator = TestNode::new_localhost_with_pubkey(keypair.pubkey());
     let validator_data = validator.data.clone();
     let val_fullnode = FullNode::new(
         validator,
         false,
-        InFile::Path(stale_ledger_path.clone()),
+        LedgerFile::Path(stale_ledger_path.clone()),
         Some(keypair),
         Some(leader_data.contact_info.ncp),
-        None,
     );
 
     // trigger broadcast, validator should catch up from leader, whose window contains
@@ -379,17 +370,16 @@ fn test_leader_restart_validator_start_from_old_ledger() {
 fn test_multi_node_dynamic_network() {
     logger::setup();
     const N: usize = 60;
-    let leader = TestNode::new();
+    let leader = TestNode::new_localhost();
     let bob_pubkey = KeyPair::new().pubkey();
     let (alice, ledger_path) = genesis(100_000);
     let leader_data = leader.data.clone();
     let server = FullNode::new(
         leader,
         true,
-        InFile::Path(ledger_path.clone()),
+        LedgerFile::Path(ledger_path.clone()),
         None,
         None,
-        Some(OutFile::Path(ledger_path.clone())),
     );
     info!("{:x} LEADER", leader_data.debug_id());
     let leader_balance =
@@ -403,7 +393,7 @@ fn test_multi_node_dynamic_network() {
         .into_iter()
         .map(|n| {
             let keypair = KeyPair::new();
-            let validator = TestNode::new_with_pubkey(keypair.pubkey());
+            let validator = TestNode::new_localhost_with_pubkey(keypair.pubkey());
             let rd = validator.data.clone();
             //send some tokens to the new validator
             let bal =
@@ -412,16 +402,16 @@ fn test_multi_node_dynamic_network() {
             let val = FullNode::new(
                 validator,
                 false,
-                InFile::Path(ledger_path.clone()),
+                LedgerFile::Path(ledger_path.clone()),
                 Some(keypair),
                 Some(leader_data.contact_info.ncp),
-                Some(OutFile::Path(ledger_path.clone())),
             );
             info!("started[{}/{}] {:x}", n, N, rd.debug_id());
             (rd, val)
         })
         .collect();
 
+    let mut consecutive_success = 0;
     for i in 0..N {
         //verify leader can do transfer
         let expected = ((i + 3) * 500) as i64;
@@ -464,13 +454,25 @@ fn test_multi_node_dynamic_network() {
                 validators.len(),
                 distance
             );
-            //assert_eq!(success, validators.len());
+            if success == validators.len() && distance == 0 {
+                consecutive_success += 1;
+            } else {
+                consecutive_success = 0;
+            }
+            if consecutive_success == 10 {
+                break;
+            }
         }
     }
-    for (_, node) in validators {
-        node.close().unwrap();
+    assert_eq!(consecutive_success, 10);
+    for (_, node) in &validators {
+        node.exit();
     }
-    server.close().unwrap();
+    server.exit();
+    for (_, node) in validators {
+        node.join().unwrap();
+    }
+    server.join().unwrap();
 
     std::fs::remove_file(ledger_path).unwrap();
 }