ClickHouse/tests/ci/worker/init_runner.sh

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

131 lines
4.5 KiB
Bash
Raw Normal View History

2021-12-08 17:51:53 +00:00
#!/usr/bin/env bash
set -uo pipefail
####################################
# IMPORTANT! #
# EC2 instance should have #
# `github:runner-type` tag #
# set accordingly to a runner role #
####################################
echo "Running init script"
export DEBIAN_FRONTEND=noninteractive
export RUNNER_HOME=/home/ubuntu/actions-runner
export RUNNER_URL="https://github.com/ClickHouse"
# Funny fact, but metadata service has fixed IP
INSTANCE_ID=$(ec2metadata --instance-id)
2021-12-08 17:51:53 +00:00
export INSTANCE_ID
2022-08-01 16:27:17 +00:00
# Add cloudflare DNS as a fallback
# Get default gateway interface
2022-08-02 00:23:05 +00:00
IFACE=$(ip --json route list | jq '.[]|select(.dst == "default").dev' --raw-output)
2022-08-01 16:27:17 +00:00
# `Link 2 (eth0): 172.31.0.2`
ETH_DNS=$(resolvectl dns "$IFACE") || :
2022-08-02 00:23:53 +00:00
CLOUDFLARE_NS=1.1.1.1
if [[ "$ETH_DNS" ]] && [[ "${ETH_DNS#*: }" != *"$CLOUDFLARE_NS"* ]]; then
2022-08-01 16:27:17 +00:00
# Cut the leading legend
ETH_DNS=${ETH_DNS#*: }
# shellcheck disable=SC2206
2022-08-02 00:23:53 +00:00
new_dns=(${ETH_DNS} "$CLOUDFLARE_NS")
2022-08-01 16:27:17 +00:00
resolvectl dns "$IFACE" "${new_dns[@]}"
fi
2021-12-08 17:51:53 +00:00
# combine labels
RUNNER_TYPE=$(/usr/local/bin/aws ec2 describe-tags --filters "Name=resource-id,Values=$INSTANCE_ID" --query "Tags[?Key=='github:runner-type'].Value" --output text)
2021-12-08 17:51:53 +00:00
LABELS="self-hosted,Linux,$(uname -m),$RUNNER_TYPE"
export LABELS
# Refresh CloudWatch agent config
aws ssm get-parameter --region us-east-1 --name AmazonCloudWatch-github-runners --query 'Parameter.Value' --output text > /opt/aws/amazon-cloudwatch-agent/etc/amazon-cloudwatch-agent.json
systemctl restart amazon-cloudwatch-agent.service
2022-06-27 12:45:50 +00:00
# Refresh teams ssh keys
TEAM_KEYS_URL=$(aws ssm get-parameter --region us-east-1 --name team-keys-url --query 'Parameter.Value' --output=text)
curl "${TEAM_KEYS_URL}" > /home/ubuntu/.ssh/authorized_keys2
chown ubuntu: /home/ubuntu/.ssh -R
2022-11-23 14:29:52 +00:00
# Create a pre-run script that will provide diagnostics info
mkdir -p /tmp/actions-hooks
2022-11-23 14:29:52 +00:00
cat > /tmp/actions-hooks/pre-run.sh << EOF
#!/bin/bash
2022-11-23 14:29:52 +00:00
set -uo pipefail
2022-06-27 12:46:28 +00:00
echo "Runner's public DNS: $(ec2metadata --public-hostname)"
2022-11-23 14:29:52 +00:00
echo "Runner's labels: ${LABELS}"
EOF
2022-11-23 14:29:52 +00:00
# Create a post-run script that will restart docker daemon before the job started
cat > /tmp/actions-hooks/post-run.sh << 'EOF'
#!/bin/bash
set -xuo pipefail
terminate-and-exit() {
echo "Going to terminate the runner"
INSTANCE_ID=$(ec2metadata --instance-id)
2022-06-28 15:51:47 +00:00
# We execute it with at to not have it as an orphan process
# GH Runners kill all remain processes
echo "sleep 10; aws ec2 terminate-instances --instance-ids $INSTANCE_ID" | at now || \
aws ec2 terminate-instances --instance-ids "$INSTANCE_ID" # workaround for complete out of space
exit 0
}
# Free KiB, free percents
ROOT_STAT=($(df / | awk '/\// {print $4 " " int($4/$2 * 100)}'))
if [[ ${ROOT_STAT[0]} -lt 3000000 ]] || [[ ${ROOT_STAT[1]} -lt 5 ]]; then
echo "The runner has ${ROOT_STAT[0]}KiB and ${ROOT_STAT[1]}% of free space on /"
terminate-and-exit
fi
# shellcheck disable=SC2046
docker ps --quiet | xargs --no-run-if-empty docker kill ||:
# shellcheck disable=SC2046
docker ps --all --quiet | xargs --no-run-if-empty docker rm -f ||:
# If we have hanged containers after the previous commands, than we have a hanged one
# and should restart the daemon
if [ "$(docker ps --all --quiet)" ]; then
# Systemd service of docker has StartLimitBurst=3 and StartLimitInterval=60s,
# that's why we try restarting it for long
for i in {1..25};
do
sudo systemctl restart docker && break || sleep 5
done
for i in {1..10}
do
docker info && break || sleep 2
done
# Last chance, otherwise we have to terminate poor instance
docker info 1>/dev/null || { echo Docker unable to start; terminate-and-exit; }
fi
EOF
2021-12-08 17:51:53 +00:00
while true; do
runner_pid=$(pgrep run.sh)
echo "Got runner pid $runner_pid"
cd $RUNNER_HOME || exit 1
if [ -z "$runner_pid" ]; then
echo "Receiving token"
RUNNER_TOKEN=$(/usr/local/bin/aws ssm get-parameter --name github_runner_registration_token --with-decryption --output text --query Parameter.Value)
echo "Will try to remove runner"
sudo -u ubuntu ./config.sh remove --token "$RUNNER_TOKEN" ||:
echo "Going to configure runner"
sudo -u ubuntu ./config.sh --url $RUNNER_URL --token "$RUNNER_TOKEN" --name "$INSTANCE_ID" --runnergroup Default --labels "$LABELS" --work _work
echo "Run"
sudo -u ubuntu \
ACTIONS_RUNNER_HOOK_JOB_STARTED=/tmp/actions-hooks/pre-run.sh \
ACTIONS_RUNNER_HOOK_JOB_COMPLETED=/tmp/actions-hooks/post-run.sh \
./run.sh &
2021-12-08 17:51:53 +00:00
sleep 15
else
echo "Runner is working with pid $runner_pid, nothing to do"
sleep 10
fi
done