v0.14: various net/ fixes for large clusters (#4080)

* net.sh: Add -F to discard validator nodes that didn't bootup successfully

* Relax sanity node count when validator bootup failure is permitted

* Less sanity for testnet-demo

* net.sh: Add -F to discard validator nodes that didn't bootup successfully
This commit is contained in:
Michael Vines
2019-04-29 21:38:03 -07:00
parent bd0871cbe7
commit d21fa4a177
7 changed files with 71 additions and 41 deletions

View File

@@ -282,39 +282,34 @@ if ! $skipStart; then
op=start
fi
echo "--- net.sh $op"
args=("$op" -t "$tarChannelOrTag")
maybeRejectExtraNodes=
if ! $publicNetwork; then
maybeRejectExtraNodes="-o rejectExtraNodes"
args+=(-o rejectExtraNodes)
fi
maybeNoValidatorSanity=
if [[ -n $NO_VALIDATOR_SANITY ]]; then
maybeNoValidatorSanity="-o noValidatorSanity"
args+=(-o noValidatorSanity)
fi
maybeNoLedgerVerify=
if [[ -n $NO_LEDGER_VERIFY ]]; then
maybeNoLedgerVerify="-o noLedgerVerify"
args+=(-o noLedgerVerify)
fi
maybeReuseLedger=
if $reuseLedger; then
maybeReuseLedger="-r"
args+=(-r)
fi
if ! $failOnValidatorBootupFailure; then
args+=(-F)
fi
maybeUpdateManifestKeypairFile=
# shellcheck disable=SC2154 # SOLANA_INSTALL_UPDATE_MANIFEST_KEYPAIR_x86_64_unknown_linux_gnu comes from .buildkite/env/
if [[ -n $SOLANA_INSTALL_UPDATE_MANIFEST_KEYPAIR_x86_64_unknown_linux_gnu ]]; then
echo "$SOLANA_INSTALL_UPDATE_MANIFEST_KEYPAIR_x86_64_unknown_linux_gnu" > update_manifest_keypair.json
maybeUpdateManifestKeypairFile="-i update_manifest_keypair.json"
args+=(-i update_manifest_keypair.json)
fi
# shellcheck disable=SC2086 # Don't want to double quote the $maybeXYZ variables
time net/net.sh $op -t "$tarChannelOrTag" \
$maybeUpdateManifestKeypairFile \
$maybeReuseLedger \
$maybeRejectExtraNodes \
$maybeNoValidatorSanity \
$maybeNoLedgerVerify
time net/net.sh "${args[@]}"
) || ok=false
net/net.sh logs

View File

@@ -267,6 +267,7 @@ sanity() {
ok=true
if [[ -n $GCE_NODE_COUNT ]]; then
NO_LEDGER_VERIFY=1 \
NO_VALIDATOR_SANITY=1 \
ci/testnet-sanity.sh demo-testnet-solana-com gce "${GCE_ZONES[0]}" -f || ok=false
else
echo "Error: no GCE nodes"
@@ -413,22 +414,26 @@ deploy() {
fi
# shellcheck disable=SC2068
ci/testnet-deploy.sh -p demo-testnet-solana-com -C gce ${GCE_ZONE_ARGS[@]} \
-t "$CHANNEL_OR_TAG" -n "$GCE_NODE_COUNT" -c 0 -P -u -f \
-a demo-testnet-solana-com \
${skipCreate:+-e} \
${maybeSkipStart:+-s} \
${maybeStop:+-S} \
${maybeDelete:+-D}
NO_LEDGER_VERIFY=1 \
NO_VALIDATOR_SANITY=1 \
ci/testnet-deploy.sh -p demo-testnet-solana-com -C gce ${GCE_ZONE_ARGS[@]} \
-t "$CHANNEL_OR_TAG" -n "$GCE_NODE_COUNT" -c 0 -P -u -f \
-a demo-testnet-solana-com \
${skipCreate:+-e} \
${maybeSkipStart:+-s} \
${maybeStop:+-S} \
${maybeDelete:+-D}
if [[ -n $GCE_LOW_QUOTA_NODE_COUNT ]]; then
# shellcheck disable=SC2068
ci/testnet-deploy.sh -p demo-testnet-solana-com2 -C gce ${GCE_LOW_QUOTA_ZONE_ARGS[@]} \
-t "$CHANNEL_OR_TAG" -n "$GCE_LOW_QUOTA_NODE_COUNT" -c 0 -P -f -x \
${skipCreate:+-e} \
${skipStart:+-s} \
${maybeStop:+-S} \
${maybeDelete:+-D}
NO_LEDGER_VERIFY=1 \
NO_VALIDATOR_SANITY=1 \
ci/testnet-deploy.sh -p demo-testnet-solana-com2 -C gce ${GCE_LOW_QUOTA_ZONE_ARGS[@]} \
-t "$CHANNEL_OR_TAG" -n "$GCE_LOW_QUOTA_NODE_COUNT" -c 0 -P -f -x \
${skipCreate:+-e} \
${skipStart:+-s} \
${maybeStop:+-S} \
${maybeDelete:+-D}
fi
)
;;