net/ can now deploy Snaps

2018-09-04 09:21:03 -07:00
parent 2cb1375217
commit bcaa0fdcb1
6 changed files with 152 additions and 90 deletions
--- a/net/remote/remote_client.sh
+++ b/net/remote/remote_client.sh
@ -1,14 +1,15 @@
 #!/bin/bash -e

+cd "$(dirname "$0")"/../..
+
 deployMethod="$1"
-netEntrypoint="$2"
+leaderIp="$2"
 numNodes="$3"
 RUST_LOG="$4"
 [[ -n $deployMethod ]] || exit
-[[ -n $netEntrypoint ]] || exit
+[[ -n $leaderIp ]] || exit
 [[ -n $numNodes ]] || exit

-cd "$(dirname "$0")"/../..
 source net/common.sh
 loadConfigFile

@ -17,14 +18,19 @@ if [[ $threadCount -gt 4 ]]; then
  threadCount=4
 fi

-./script/install-earlyoom.sh
+scripts/install-earlyoom.sh

 case $deployMethod in
 snap)
+  rsync -vPr "$leaderIp:~/solana/solana.snap" .
  sudo snap install solana.snap --devmode --dangerous
  rm solana.snap

-  sudo snap set solana metrics-config="$SOLANA_METRICS_CONFIG" rust-log="$RUST_LOG"
+  sudo snap set solana "\
+      leader-ip=$leaderIp \
+      metrics-config=$SOLANA_METRICS_CONFIG \
+      rust-log=$RUST_LOG \
+    "
  solana_bench_tps=/snap/bin/solana.bench-tps
  ;;
 local)
@ -32,20 +38,19 @@ local)
  export USE_INSTALL=1
  export RUST_LOG

-  rsync -vPrz "$netEntrypoint:~/.cargo/bin/solana*" ~/.cargo/bin/
-  solana_bench_tps=multinode-demo/client.sh
-  netEntrypoint="$:~/solana"
+  rsync -vPr "$leaderIp:~/.cargo/bin/solana*" ~/.cargo/bin/
+  solana_bench_tps="multinode-demo/client.sh $leaderIp:~/solana"
  ;;
 *)
  echo "Unknown deployment method: $deployMethod"
  exit 1
 esac

-./scripts/oom-monitor.sh  > oom-monitor.log 2>&1 &
+scripts/oom-monitor.sh  > oom-monitor.log 2>&1 &

 while true; do
  echo "=== Client start: $(date)" >> client.log
-  clientCommand="$solana_bench_tps $netEntrypoint $numNodes --loop -s 600 --sustained -t threadCount"
+  clientCommand="$solana_bench_tps --num-nodes $numNodes --loop -s 600 --sustained -t threadCount"
  echo "$ $clientCommand" >> client.log

  $clientCommand >> client.log 2>&1
--- a/net/remote/remote_node.sh
+++ b/net/remote/remote_node.sh
@ -1,37 +1,51 @@
 #!/bin/bash -e

+cd "$(dirname "$0")"/../..
+
 deployMethod="$1"
 nodeType="$2"
-netEntrypoint="$3"
-setupArgs="$4"
-RUST_LOG="$5"
+leaderIp="$3"
+numNodes="$4"
+setupArgs="$5"
+RUST_LOG="$6"
+
+cat > deployConfig <<EOF
+deployMethod="$deployMethod"
+leaderIp="$leaderIp"
+numNodes="$numNodes"
+EOF

 [[ -n $deployMethod ]] || exit
 [[ -n $nodeType ]] || exit
-[[ -n $netEntrypoint ]] || exit
+[[ -n $leaderIp ]] || exit

-cd "$(dirname "$0")"/../..
 source net/common.sh
 loadConfigFile

-./script/install-earlyoom.sh
+scripts/install-earlyoom.sh
+

 case $deployMethod in
 snap)
  SECONDS=0
+  rsync -vPr "$leaderIp:~/solana/solana.snap" .
  sudo snap install solana.snap --devmode --dangerous
-  rm solana.snap

  commonNodeConfig="\
-    rust-log=$RUST_LOG \
+    leader-ip=$leaderIp \
    metrics-config=$SOLANA_METRICS_CONFIG \
+    rust-log=$RUST_LOG \
    setup-args=$setupArgs \
-    enable-cuda=1 \
  "
+
+  if [[ -e /dev/nvidia0 ]]; then
+    commonNodeConfig="$commonNodeConfig enable-cuda=1"
+  fi
+
  if [[ $nodeType = leader ]]; then
    nodeConfig="mode=leader+drone $commonNodeConfig"
  else
-    nodeConfig="mode=validator leader-address=$netEntrypoint $commonNodeConfig"
+    nodeConfig="mode=validator $commonNodeConfig"
  fi

  logmarker="solana deploy $(date)/$RANDOM"
@ -50,11 +64,13 @@ snap)
 local)
  PATH="$HOME"/.cargo/bin:"$PATH"
  export USE_INSTALL=1
-  export SOLANA_CUDA=1
-  export RUST_LOG=1
+  export RUST_LOG
+  if [[ -e /dev/nvidia0 ]]; then
+    export SOLANA_CUDA=1
+  fi

  ./fetch-perf-libs.sh
-  ./scripts/oom-monitor.sh  > oom-monitor.log 2>&1 &
+  scripts/oom-monitor.sh  > oom-monitor.log 2>&1 &

  case $nodeType in
  leader)
@ -64,11 +80,11 @@ local)
    ./multinode-demo/leader.sh > leader.log 2>&1 &
    ;;
  validator)
-    rsync -vPrz "$netEntrypoint:~/.cargo/bin/solana*" ~/.cargo/bin/
+    rsync -vPr "$leaderIp:~/.cargo/bin/solana*" ~/.cargo/bin/

    # shellcheck disable=SC2086 # Don't want to double quote "$setupArgs"
    ./multinode-demo/setup.sh -t validator -p $setupArgs
-    ./multinode-demo/validator.sh "$netEntrypoint":~/solana "$netEntrypoint" >validator.log 2>&1 &
+    ./multinode-demo/validator.sh "$leaderIp":~/solana "$leaderIp" >validator.log 2>&1 &
    ;;
  *)
    echo "Error: unknown node type: $nodeType"
@ -80,4 +96,3 @@ local)
  echo "Unknown deployment method: $deployMethod"
  exit 1
 esac
-
--- a/net/remote/remote_sanity.sh
+++ b/net/remote/remote_sanity.sh
@ -1,14 +1,25 @@
 #!/bin/bash -e

-deployMethod="$1"
-netEntrypoint="$2"
-numNodes="$3"
+cd "$(dirname "$0")"/../..

-[[ -n $deployMethod ]] || exit
-[[ -n $netEntrypoint ]] || exit
-[[ -n $numNodes ]] || exit
+deployMethod=
+leaderIp=
+numNodes=
+# shellcheck source=/dev/null # deployConfig is written by remote_sanity.sh
+source deployConfig

-shift 3
+[[ -n $deployMethod ]] || {
+  echo "deployMethod empty"
+  exit 1
+}
+[[ -n $leaderIp ]] || {
+  echo "leaderIp empty"
+  exit 1
+}
+[[ -n $numNodes ]] || {
+  echo "numNodes empty"
+  exit 1
+}

 ledgerVerify=true
 validatorSanity=true
@ -29,26 +40,25 @@ while [[ $1 = "-o" ]]; do
  esac
 done

-
-cd "$(dirname "$0")"/../..
 source net/common.sh
 loadConfigFile

 case $deployMethod in
 snap)
+  PATH="/snap/bin:$PATH"
  export USE_SNAP=1
-  solana_bench_tps=/snap/bin/solana.bench-tps
-  solana_ledger_tool=/snap/bin/solana.ledger-tool
+
+  solana_bench_tps=solana.bench-tps
+  solana_ledger_tool=solana.ledger-tool
  ledger=/var/snap/solana/current/config/ledger
  ;;
 local)
  PATH="$HOME"/.cargo/bin:"$PATH"
  export USE_INSTALL=1

-  solana_bench_tps=multinode-demo/client.sh
+  solana_bench_tps="multinode-demo/client.sh $leaderIp:~/solana"
  solana_ledger_tool=solana-ledger-tool
  ledger=config/ledger
-  netEntrypoint="$:~/solana"
  ;;
 *)
  echo "Unknown deployment method: $deployMethod"
@ -56,19 +66,19 @@ local)
 esac


-echo "--- $netEntrypoint: wallet sanity"
+echo "--- $leaderIp: wallet sanity"
 (
  set -x
-  multinode-demo/test/wallet-sanity.sh "$netEntrypoint"
+  multinode-demo/test/wallet-sanity.sh "$leaderIp"
 )

-echo "--- $netEntrypoint: node count"
+echo "--- $leaderIp: node count"
 (
  set -x
-  $solana_bench_tps "$netEntrypoint" "$numNodes" -c
+  $solana_bench_tps --num-nodes "$numNodes" --converge-only
 )

-echo "--- $netEntrypoint: verify ledger"
+echo "--- $leaderIp: verify ledger"
 if $ledgerVerify; then
  if [[ -d $ledger ]]; then
    (
@ -87,12 +97,12 @@ else
 fi


-echo "--- $netEntrypoint: validator sanity"
+echo "--- $leaderIp: validator sanity"
 if $validatorSanity; then
  (
    ./multinode-demo/setup.sh -t validator
    set -e pipefail
-    timeout 10s ./multinode-demo/validator.sh "$netEntrypoint" 2>&1 | tee validator.log
+    timeout 10s ./multinode-demo/validator.sh "$leaderIp" 2>&1 | tee validator.log
  )
  wc -l validator.log
  if grep -C100 panic validator.log; then
--- a/net/remote/remote_startup.sh
+++ b/net/remote/remote_startup.sh
@ -9,3 +9,12 @@ systemctl disable apt-daily.service # disable run when system boot
 systemctl disable apt-daily.timer   # disable timer run
 apt-get --assume-yes install rsync libssl-dev

+cat > /etc/rsyncd.conf <<-EOF
+[config]
+path = /var/snap/solana/current/config
+hosts allow = *
+read only = true
+EOF
+
+systemctl enable rsync
+systemctl start rsync