Files
docker-autoheal/docker-entrypoint
T
2020-03-11 18:53:40 +01:00

90 lines
2.5 KiB
Bash
Executable File

#!/usr/bin/env sh
set -e
# shellcheck disable=2039
set -o pipefail
DOCKER_SOCK=${DOCKER_SOCK:-/var/run/docker.sock}
CURL_TIMEOUT=${CURL_TIMEOUT:-30}
AUTOHEAL_CONTAINER_LABEL=${AUTOHEAL_CONTAINER_LABEL:-autoheal}
AUTOHEAL_START_PERIOD=${AUTOHEAL_START_PERIOD:-0}
AUTOHEAL_INTERVAL=${AUTOHEAL_INTERVAL:-5}
AUTOHEAL_DEFAULT_STOP_TIMEOUT=${AUTOHEAL_DEFAULT_STOP_TIMEOUT:-10}
# shellcheck disable=2039
docker_curl() {
curl --max-time "$CURL_TIMEOUT" --no-buffer -s --unix-socket "$DOCKER_SOCK" "$@"
}
# shellcheck disable=2039
get_container_info() {
local label_filter
local url
# Set container selector
if [ "$AUTOHEAL_CONTAINER_LABEL" = "all" ]
then
label_filter=""
else
label_filter=",\"label\":\[\"${AUTOHEAL_CONTAINER_LABEL}=true\"\]"
fi
url="http://localhost/containers/json?filters=\{\"health\":\[\"unhealthy\"\]${label_filter}\}"
docker_curl "$url"
}
# shellcheck disable=2039
restart_container() {
local container_id="$1"
local timeout="$2"
docker_curl -f -X POST "http://localhost/containers/${container_id}/restart?t=${timeout}"
}
# SIGTERM-handler
term_handler() {
exit 143 # 128 + 15 -- SIGTERM
}
# shellcheck disable=2039
trap 'kill $$; term_handler' SIGTERM
if [ "$1" = "autoheal" ] && [ -e "$DOCKER_SOCK" ]
then
# https://docs.docker.com/engine/api/v1.25/
# Delayed startup
if [ "$AUTOHEAL_START_PERIOD" -gt 0 ]
then
echo "Monitoring containers for unhealthy status in $AUTOHEAL_START_PERIOD second(s)"
sleep "$AUTOHEAL_START_PERIOD"
fi
while true
do
STOP_TIMEOUT=".Labels[\"autoheal.stop.timeout\"] // $AUTOHEAL_DEFAULT_STOP_TIMEOUT"
get_container_info | \
jq -r "foreach .[] as \$CONTAINER([];[]; \$CONTAINER | .Id, .Names[0], ${STOP_TIMEOUT})" | \
while read -r CONTAINER_ID && read -r CONTAINER_NAME && read -r TIMEOUT
do
# shellcheck disable=2039
CONTAINER_SHORT_ID=${CONTAINER_ID:0:12}
DATE=$(date +%d-%m-%Y" "%H:%M:%S)
if [ "$CONTAINER_NAME" = "null" ]
then
echo "$DATE Container name of (${CONTAINER_SHORT_ID}) is null, which implies container does not exist - don't restart" >&2
else
echo "$DATE Container $CONTAINER_NAME (${CONTAINER_SHORT_ID}) found to be unhealthy - Restarting container now with ${TIMEOUT}s timeout"
if ! restart_container "$CONTAINER_ID" "$TIMEOUT"
then
echo "$DATE Restarting container $CONTAINER_SHORT_ID failed" >&2
fi
fi
done
sleep "$AUTOHEAL_INTERVAL"
done
else
exec "$@"
fi