solana/ci/testnet-manager.sh

350 lines
8.5 KiB
Bash
Raw Normal View History

#!/usr/bin/env bash
set -e
2018-11-05 13:21:34 -08:00
cd "$(dirname "$0")"/..
if [[ -z $BUILDKITE ]]; then
echo BUILDKITE not defined
exit 1
fi
2018-11-06 07:24:49 -08:00
if [[ -z $SOLANA_METRICS_PARTIAL_CONFIG ]]; then
2018-11-06 19:23:20 -08:00
echo SOLANA_METRICS_PARTIAL_CONFIG not defined
2018-11-05 13:21:34 -08:00
exit 1
fi
if [[ -z $TESTNET ]]; then
TESTNET=$(buildkite-agent meta-data get "testnet" --default "")
fi
if [[ -z $TESTNET_OP ]]; then
TESTNET_OP=$(buildkite-agent meta-data get "testnet-operation" --default "")
fi
if [[ -z $TESTNET || -z $TESTNET_OP ]]; then
(
cat <<EOF
steps:
- block: "Manage Testnet"
fields:
- select: "Network"
key: "testnet"
options:
- label: "testnet"
value: "testnet"
- label: "testnet-perf"
value: "testnet-perf"
2018-11-06 19:23:20 -08:00
- label: "testnet-edge"
value: "testnet-edge"
- label: "testnet-edge-perf"
value: "testnet-edge-perf"
- label: "testnet-beta"
value: "testnet-beta"
- label: "testnet-beta-perf"
value: "testnet-beta-perf"
2018-11-05 13:21:34 -08:00
- select: "Operation"
key: "testnet-operation"
default: "sanity-or-restart"
options:
- label: "Sanity check. Restart network on failure"
value: "sanity-or-restart"
2018-11-05 19:12:43 -08:00
- label: "Start (or restart) the network"
2018-11-05 13:21:34 -08:00
value: "start"
- label: "Update the network software. Restart network on failure"
value: "update-or-restart"
2018-11-05 13:21:34 -08:00
- label: "Stop the network"
value: "stop"
- label: "Sanity check only"
value: "sanity"
- command: "ci/$(basename "$0")"
agents:
- "queue=$BUILDKITE_AGENT_META_DATA_QUEUE"
EOF
) | buildkite-agent pipeline upload
exit 0
fi
if [[ -n $TESTNET_DB_HOST ]]; then
SOLANA_METRICS_PARTIAL_CONFIG="host=$TESTNET_DB_HOST,$SOLANA_METRICS_PARTIAL_CONFIG"
fi
2018-11-06 07:24:49 -08:00
export SOLANA_METRICS_CONFIG="db=$TESTNET,$SOLANA_METRICS_PARTIAL_CONFIG"
echo "SOLANA_METRICS_CONFIG: $SOLANA_METRICS_CONFIG"
2019-01-04 17:36:14 -08:00
source scripts/configure-metrics.sh
2018-11-06 07:24:49 -08:00
2018-11-05 13:21:34 -08:00
ci/channel-info.sh
eval "$(ci/channel-info.sh)"
case $TESTNET in
testnet-edge|testnet-edge-perf)
CHANNEL_OR_TAG=edge
2018-11-05 13:21:34 -08:00
CHANNEL_BRANCH=$EDGE_CHANNEL
;;
2018-11-06 19:23:20 -08:00
testnet-beta|testnet-beta-perf)
CHANNEL_OR_TAG=beta
CHANNEL_BRANCH=$BETA_CHANNEL
;;
2018-11-05 13:21:34 -08:00
testnet|testnet-perf)
CHANNEL_OR_TAG=$STABLE_CHANNEL_LATEST_TAG
CHANNEL_BRANCH=$STABLE_CHANNEL
2018-11-05 13:21:34 -08:00
;;
*)
echo "Error: Invalid TESTNET=$TESTNET"
exit 1
;;
esac
if [[ $BUILDKITE_BRANCH != "$CHANNEL_BRANCH" ]]; then
(
cat <<EOF
steps:
- trigger: "$BUILDKITE_PIPELINE_SLUG"
async: true
build:
message: "$BUILDKITE_MESSAGE"
branch: "$CHANNEL_BRANCH"
env:
TESTNET: "$TESTNET"
TESTNET_OP: "$TESTNET_OP"
TESTNET_DB_HOST: "$TESTNET_DB_HOST"
2018-11-05 13:21:34 -08:00
EOF
) | buildkite-agent pipeline upload
exit 0
fi
sanity() {
echo "--- sanity $TESTNET"
case $TESTNET in
2018-11-06 19:23:20 -08:00
testnet-edge)
2018-11-05 13:21:34 -08:00
(
set -x
ci/testnet-sanity.sh edge-testnet-solana-com ec2 us-west-1a
2018-11-06 19:23:20 -08:00
)
;;
testnet-edge-perf)
(
set -x
REJECT_EXTRA_NODES=1 \
NO_LEDGER_VERIFY=1 \
NO_VALIDATOR_SANITY=1 \
ci/testnet-sanity.sh edge-perf-testnet-solana-com ec2 us-west-2b
2018-11-06 19:23:20 -08:00
)
;;
testnet-beta)
(
set -x
EC2_ZONES=(sa-east-1a us-west-1a ap-northeast-2a eu-central-1a ca-central-1a)
2019-04-05 11:51:05 -07:00
ok=true
for zone in "${EC2_ZONES[@]}"; do
2019-04-05 11:51:05 -07:00
if ! $ok; then
break
fi
ci/testnet-sanity.sh beta-testnet-solana-com ec2 "$zone" || ok=false
done
GCE_ZONES=(us-west1-b asia-east2-a europe-west4-a southamerica-east1-b us-east4-c)
for zone in "${GCE_ZONES[@]}"; do
2019-04-05 11:51:05 -07:00
if ! $ok; then
break
fi
ci/testnet-sanity.sh beta-testnet-solana-com gce "$zone" || ok=false
done
2019-04-05 11:51:05 -07:00
$ok
2018-11-06 19:23:20 -08:00
)
;;
testnet-beta-perf)
(
set -x
REJECT_EXTRA_NODES=1 \
NO_LEDGER_VERIFY=1 \
NO_VALIDATOR_SANITY=1 \
ci/testnet-sanity.sh beta-perf-testnet-solana-com ec2 us-west-2b
2018-11-05 13:21:34 -08:00
)
;;
testnet)
(
set -x
ci/testnet-sanity.sh testnet-solana-com ec2 us-west-1a
2018-11-07 18:30:45 -08:00
#ci/testnet-sanity.sh testnet-solana-com gce us-east1-c
2018-11-05 13:21:34 -08:00
)
;;
testnet-perf)
(
set -x
REJECT_EXTRA_NODES=1 \
NO_LEDGER_VERIFY=1 \
NO_VALIDATOR_SANITY=1 \
ci/testnet-sanity.sh perf-testnet-solana-com gce us-west1-b
2018-11-07 10:48:21 -08:00
#ci/testnet-sanity.sh perf-testnet-solana-com ec2 us-east-1a
2018-11-05 13:21:34 -08:00
)
;;
*)
echo "Error: Invalid TESTNET=$TESTNET"
exit 1
;;
esac
}
start() {
declare maybeDelete=$1
if [[ -z $maybeDelete ]]; then
echo "--- start $TESTNET"
else
echo "--- stop $TESTNET"
fi
declare maybeReuseLedger=$2
2018-11-05 13:21:34 -08:00
case $TESTNET in
2018-11-06 19:23:20 -08:00
testnet-edge)
2018-11-05 13:21:34 -08:00
(
set -x
NO_VALIDATOR_SANITY=1 \
RUST_LOG=solana=info \
2019-04-05 09:41:25 -07:00
ci/testnet-deploy.sh -N edge-testnet-solana-com -C ec2 -z us-west-1a \
2019-02-17 09:49:29 -08:00
-t "$CHANNEL_OR_TAG" -n 3 -c 0 -u -P -a eipalloc-0ccd4f2239886fa94 \
${maybeReuseLedger:+-r} \
${maybeDelete:+-D}
2018-11-06 19:23:20 -08:00
)
;;
testnet-edge-perf)
(
set -x
NO_LEDGER_VERIFY=1 \
NO_VALIDATOR_SANITY=1 \
2019-04-05 09:41:25 -07:00
ci/testnet-deploy.sh -N edge-perf-testnet-solana-com -C ec2 -z us-west-2b \
-g -t "$CHANNEL_OR_TAG" -c 2 \
-b \
${maybeReuseLedger:+-r} \
${maybeDelete:+-D}
2018-11-06 19:23:20 -08:00
)
;;
testnet-beta)
(
set -x
2019-04-05 09:41:25 -07:00
# List of zones to deploy for each cloud provider
GCE_ZONES=(us-west1-b asia-east2-a europe-west4-a southamerica-east1-b us-east4-c)
EC2_ZONES=(sa-east-1a us-west-1a ap-northeast-2a eu-central-1a ca-central-1a)
# Build a string to pass as opts to testnet-deploy.sh: "-z zone1 -z zone2 ..."
for val in "${GCE_ZONES[@]}"; do
GCE_ZONE_ARGS="$GCE_ZONE_ARGS -z $val"
2019-04-05 09:41:25 -07:00
done
for val in "${EC2_ZONES[@]}"; do
EC2_ZONE_ARGS="$EC2_ZONE_ARGS -z $val"
2019-04-05 09:41:25 -07:00
done
2019-04-03 19:30:35 -07:00
NO_VALIDATOR_SANITY=1 \
RUST_LOG=solana=info \
2019-04-05 09:41:25 -07:00
ci/testnet-deploy.sh -N beta-testnet-solana-com -C ec2 "$EC2_ZONE_ARGS"\
-t "$CHANNEL_OR_TAG" -n 60 -c 0 -s -u -P -a eipalloc-0f286cf8a0771ce35 \
2019-04-03 19:30:35 -07:00
${maybeReuseLedger:+-r} \
${maybeDelete:+-D}
NO_VALIDATOR_SANITY=1 \
RUST_LOG=solana=info \
2019-04-05 09:41:25 -07:00
ci/testnet-deploy.sh -N beta-testnet-solana-com -C gce "$GCE_ZONE_ARGS"\
-t "$CHANNEL_OR_TAG" -n 40 -c 0 -x -P \
2019-04-03 19:30:35 -07:00
${maybeReuseLedger:+-r} \
${maybeDelete:+-D}
2018-11-06 19:23:20 -08:00
)
;;
testnet-beta-perf)
(
set -x
NO_LEDGER_VERIFY=1 \
NO_VALIDATOR_SANITY=1 \
2019-04-05 09:41:25 -07:00
ci/testnet-deploy.sh -N beta-perf-testnet-solana-com -C ec2 -z us-west-2b \
-g -t "$CHANNEL_OR_TAG" -c 2 \
-b \
${maybeReuseLedger:+-r} \
${maybeDelete:+-D}
2018-11-05 13:21:34 -08:00
)
;;
testnet)
(
set -x
NO_VALIDATOR_SANITY=1 \
RUST_LOG=solana=info \
2019-04-05 09:41:25 -07:00
ci/testnet-deploy.sh -N testnet-solana-com -C ec2 -z us-west-1a \
2019-02-17 09:49:29 -08:00
-t "$CHANNEL_OR_TAG" -n 3 -c 0 -u -P -a eipalloc-0fa502bf95f6f18b2 \
-b \
${maybeReuseLedger:+-r} \
${maybeDelete:+-D}
2019-04-05 09:41:25 -07:00
#ci/testnet-deploy.sh -N testnet-solana-com -C gce -z us-east1-c \
2019-03-02 17:08:46 -08:00
# -t "$CHANNEL_OR_TAG" -n 3 -c 0 -P -a testnet-solana-com \
# ${maybeReuseLedger:+-r} \
# ${maybeDelete:+-D}
2018-11-05 13:21:34 -08:00
)
;;
testnet-perf)
(
set -x
NO_LEDGER_VERIFY=1 \
NO_VALIDATOR_SANITY=1 \
2019-04-05 09:41:25 -07:00
ci/testnet-deploy.sh -N perf-testnet-solana-com -C gce -z us-west1-b \
-G "n1-standard-16 --accelerator count=2,type=nvidia-tesla-v100" \
-t "$CHANNEL_OR_TAG" -c 2 \
-b \
-d pd-ssd \
${maybeReuseLedger:+-r} \
${maybeDelete:+-D}
2019-04-05 09:41:25 -07:00
#ci/testnet-deploy.sh -N perf-testnet-solana-com -C ec2 -z us-east-1a \
# -g \
# -t "$CHANNEL_OR_TAG" -c 2 \
# ${maybeReuseLedger:+-r} \
# ${maybeDelete:+-D}
2018-11-05 13:21:34 -08:00
)
;;
*)
echo "Error: Invalid TESTNET=$TESTNET"
exit 1
;;
esac
}
stop() {
start delete
}
case $TESTNET_OP in
sanity)
sanity
;;
start)
start
;;
stop)
stop
;;
update-or-restart)
if start "" update; then
echo Update successful
else
echo "+++ Update failed, restarting the network"
$metricsWriteDatapoint "testnet-manager update-failure=1"
start
fi
;;
2018-11-05 13:21:34 -08:00
sanity-or-restart)
if sanity; then
echo Pass
else
echo "+++ Sanity failed, updating the network"
2019-01-04 17:36:14 -08:00
$metricsWriteDatapoint "testnet-manager sanity-failure=1"
if start "" update; then
echo Update successful
else
echo "+++ Update failed, restarting the network"
$metricsWriteDatapoint "testnet-manager update-failure=1"
start
fi
2018-11-05 13:21:34 -08:00
fi
;;
esac
echo --- fin
exit 0