Improve rebuild feature: handle check upgrade status changes, run rebuild via cron and add a deploy cron job

This commit is contained in:
Benjamin Renard 2024-07-08 02:15:21 +02:00
parent 2b07eab7e2
commit 525473a7fc
2 changed files with 479 additions and 269 deletions

View file

@ -25,14 +25,23 @@ service nagios-nrpe-server reload
```
Usage : check_container_upgrade [-d] [-E /path/to/engine] [container1,...]
-E [path] Force a specific engine (possible values: auto docker podman, default: auto)
-E [path] Force a specific engine (possible values: auto docker podman,
default: auto)
-x [container] Exclude specified container (could be repeat)
-M [integer] Max number of container checks to run in parallel (default: 4, 0=no limit)
-f [docker-compose.yml] To check upgrade on docker compose project, specified the path of the docker-compose.yml file
-b|--build|--rebuild Trigger container build if upgrade detected (only possible if a docker compose file if provided)
--rebuild-lock Specify rebuild lock file path (default: /var/tmp/check_container_upgrade.lock)
--rebuild-log Specify rebuild log file path (default: /var/tmp/check_container_upgrade.log)
-M [integer] Max number of container checks to run in parallel
(default: 4, 0=no limit)
-f [docker-compose.yml] To check upgrade on docker compose project, specified the path of the
docker-compose.yml file
-b|--build|--rebuild Trigger container build if upgrade detected (only possible if a docker
compose file if provided)
--rebuild-path Specify rebuild data directory path (default: /var/log/check_container_upgrade)
--rebuild-cron Start in rebuild cron mode: rebuild containers detected and mark to be
rebuilt on status file.
--deploy-cron Start in deploy cron mode: deploy containers known as rebuilt in status
file.
-d Debug mode
-l Log file
-C Console logging (even if log file is specify)
-X Enable bash tracing (=set -x)
-h Show this message
```

View file

@ -10,24 +10,42 @@ ENGINE="auto"
POSSIBLE_ENGINES=( "auto" "docker" "podman" )
DOCKERCOMPOSE_FILE=""
DEBUG=0
CONSOLE=0
LOG_FILE=0
MAX_PARALLEL_CHECKS=4
ONLY_CONTAINERS=()
EXCLUDED_CONTAINERS=( buildx_buildkit_default )
REBUILD=0
REBUILD_LOCK_FILE="/var/tmp/$( basename $0 ).lock"
REBUILD_LOG_FILE="/var/tmp/$( basename $0 ).log"
REBUILD_DATA_DIR="/var/log/$(basename $0)"
REBUILD_CRON=0
DEPLOY_CRON=0
declare -rA CHECK_PLUGINS=(
["/usr/lib/nagios/plugins/check_apt"]="/usr/lib/nagios/plugins/check_apt -u -U -t 60 -l"
["/usr/lib/nagios/plugins/check_apk"]="/usr/lib/nagios/plugins/check_apk"
)
function now() { [[ -z "$1" ]] && date "+%F %H:%M:%S" || date -d "@$1" "+%F %H:%M:%S" ; }
function current_time() { date "+%s"; }
function debug() {
if [ $DEBUG -eq 1 ]
then
[[ $DEBUG -eq 0 ]] && return
if [[ -n "$LOG_FILE" ]] && [[ $CONSOLE -eq 1 ]]; then
echo -e "$(now) - [DEBUG] $@" | tee -a "$LOG_FILE" 2>&1
elif [[ -n "$LOG_FILE" ]]; then
echo -e "$(now) - [DEBUG] $@" >> "$LOG_FILE"
else
>&2 echo -e "[DEBUG] $@"
fi
}
function log() {
[[ -n "$LOG_FILE" ]] && echo -e "$(now) - [$1] ${@:2}" >> "$LOG_FILE"
[[ "$1" == "ERROR" ]] && >&2 echo -e "ERROR - ${@:2}" || echo -e "${@:2}"
}
function message() { log INFO $@ ; }
function error() { log ERROR $@ ; }
function is_empty() {
[ $# -gt 0 ] && return 1
return 0
@ -50,19 +68,91 @@ function implode() {
fi
}
function format_duration {
local T=$1
local D=$((T/60/60/24))
local H=$((T/60/60%24))
local M=$((T/60%60))
local S=$((T%60))
(( $D > 0 )) && printf '%d days and ' $D
printf '%02d:%02d:%02d' $H $M $S
}
REBUILD_STATUS_FILE=""
function rebuild_status_file() {
[[ ! -e "$REBUILD_DATA_DIR" ]] && mkdir -p "$REBUILD_DATA_DIR"
[[ -z "$REBUILD_STATUS_FILE" ]] && \
REBUILD_STATUS_FILE="$REBUILD_DATA_DIR/status.json"
[[ -e "$REBUILD_STATUS_FILE" ]] || echo '{}' > "$REBUILD_STATUS_FILE"
echo "$REBUILD_STATUS_FILE"
}
function rebuild_status() {
if [[ -z "$1" ]]; then
cat "$(rebuild_status_file)"
else
rebuild_status | jq -r --arg container "$1" '.[$container]'
fi
}
function update_rebuild_status() {
if [[ "$1" == "-d" ]]; then
local data=$(
rebuild_status | jq \
--arg container "$2" \
'del(.[$container])'
)
else
local args=( --arg container "$1" )
local expr=( )
for arg in "${@:2}"; do
local name=$( echo "$arg" | cut -d'=' -f1 )
local value=$( echo "$arg" | sed 's/[^=]\+=//' )
args+=( --arg "$name" "$value" )
expr+=( ".[\$container].$name=\$$name" )
done
local data=$(
rebuild_status | jq \
"${args[@]}" "$( implode ' | ' "${expr[@]}" )"
)
fi
cat <<< $data > $(rebuild_status_file)
}
function remove_rebuild_status() {
rebuild_status_file > /dev/null
[[ -e "$REBUILD_STATUS_FILE" ]] || return 0
debug "Remove previous rebuild status file ($REBUILD_STATUS_FILE) and log files it contains"
for log in $( jq -r '.[] | .log' $REBUILD_STATUS_FILE ); do
debug " remove old container log file $log"
rm -f $log
done
debug " remove status file"
rm -f "$REBUILD_STATUS_FILE"
}
function usage() {
error="$1"
[ -n "$error" ] && echo "$error"
[ -n "$error" ] && message "$error"
cat << EOF
Usage : $(basename $0) [-d] [-E /path/to/engine] [container1,...]
-E [path] Force a specific engine (possible values: ${POSSIBLE_ENGINES[@]}, default: $ENGINE)
-E [path] Force a specific engine (possible values: ${POSSIBLE_ENGINES[@]},
default: $ENGINE)
-x [container] Exclude specified container (could be repeat)
-M [integer] Max number of container checks to run in parallel (default: $MAX_PARALLEL_CHECKS, 0=no limit)
-f [docker-compose.yml] To check upgrade on docker compose project, specified the path of the docker-compose.yml file
-b|--build|--rebuild Trigger container build if upgrade detected (only possible if a docker compose file if provided)
--rebuild-lock Specify rebuild lock file path (default: ${REBUILD_LOCK_FILE})
--rebuild-log Specify rebuild log file path (default: ${REBUILD_LOG_FILE})
-M [integer] Max number of container checks to run in parallel
(default: $MAX_PARALLEL_CHECKS, 0=no limit)
-f [docker-compose.yml] To check upgrade on docker compose project, specified the path of the
docker-compose.yml file
-b|--build|--rebuild Trigger container build if upgrade detected (only possible if a docker
compose file if provided)
--rebuild-path Specify rebuild data directory path (default: ${REBUILD_DATA_DIR})
--rebuild-cron Start in rebuild cron mode: rebuild containers detected and mark to be
rebuilt on status file.
--deploy-cron Start in deploy cron mode: deploy containers known as rebuilt in status
file.
-d Debug mode
-l Log file
-C Console logging (even if log file is specify)
-X Enable bash tracing (=set -x)
-h Show this message
EOF
@ -78,6 +168,13 @@ do
-d)
DEBUG=1
;;
-C)
CONSOLE=1
;;
-l)
((idx++))
LOG_FILE=${!idx}
;;
-h)
usage
;;
@ -96,13 +193,9 @@ do
-b|--build|--rebuild)
REBUILD=1
;;
--rebuild-lock)
--rebuild-path)
((idx++))
REBUILD_LOCK_FILE="${!idx}"
;;
--rebuild-log)
((idx++))
REBUILD_LOG_FILE="${!idx}"
REBUILD_DATA_DIR="${!idx}"
;;
-x)
((idx++))
@ -115,6 +208,12 @@ do
-X)
set -x
;;
--rebuild-cron)
REBUILD_CRON=1
;;
--deploy-cron)
DEPLOY_CRON=1
;;
*)
ONLY_CONTAINERS+=( $OPT )
;;
@ -122,6 +221,8 @@ do
((idx++))
done
debug "Start with parameters: $@"
! is_empty $ONLY_CONTAINERS && debug "Only containers: ${ONLY_CONTAINERS[@]}"
if [ "$ENGINE" == "auto" ]
@ -141,17 +242,89 @@ then
done
if [ -z "$ENGINE" ]
then
echo "UNKNOWN - Fail to auto-detect engine"
message "UNKNOWN - Fail to auto-detect engine"
exit 3
fi
debug "Auto-detected engine: $ENGINE"
fi
if [ -n "$DOCKERCOMPOSE_FILE" -a ! -e "$DOCKERCOMPOSE_FILE" ]
then
echo "UNKNOWN - Docker compose file not found ($DOCKERCOMPOSE_FILE)"
if [[ -n "$DOCKERCOMPOSE_FILE" ]]; then
if [[ ! -e "$DOCKERCOMPOSE_FILE" ]]; then
message "UNKNOWN - Docker compose file not found ($DOCKERCOMPOSE_FILE)"
exit 3
fi
which "${ENGINE}-compose" > /dev/null 2>&1 && \
COMPOSE_BIN="${ENGINE}-compose" || \
COMPOSE_BIN="$ENGINE compose"
debug "Docker compose bin: $COMPOSE_BIN"
fi
if [[ $REBUILD_CRON -eq 1 ]]; then
to_rebuild=(
$(
rebuild_status | jq -r -c \
'to_entries[] | select((.value.start_date|not) and (.value.error|not)) | .key'
)
)
if [[ ${#to_rebuild[@]} -eq 0 ]]; then
debug "No container need to be rebuild"
exit 0
fi
message "${#to_rebuild[@]} container(s) to rebuild: ${to_rebuild[@]}"
error=0
for container in ${to_rebuild[@]}; do
log="$REBUILD_DATA_DIR/$container.log"
message " $container: start building image (log=$log)"
start_time=$(current_time)
update_rebuild_status "$container" "start_date=$(now $start_time)" "log=$log"
$COMPOSE_BIN -f $DOCKERCOMPOSE_FILE build --no-cache $container >> $log 2>&1
result=$?
end_time=$(current_time)
(( duration=end_time-start_time ))
duration=$(format_duration $duration)
container_info=( "end_date=$(now $end_time)" "duration=$duration" )
if [[ $result -eq 0 ]]; then
message " $container: rebuilt in $duration"
else
error " $container: fail to rebuild image"
container_info+=( "error=fail to rebuild image" )
fi
update_rebuild_status "$container" "${container_info[@]}"
done
message "No more container to rebuild, stop"
exit 0
fi
if [[ $DEPLOY_CRON -eq 1 ]]; then
to_deploy=(
$(
rebuild_status | jq -r -c \
'to_entries[] | select((.value.end_date) and (.value.error|not)) | .key'
)
)
if [[ ${#to_deploy[@]} -eq 0 ]]; then
debug "No container need to be deploy"
exit 0
fi
message "${#to_deploy[@]} container(s) to deploy: ${to_deploy[@]}"
error=0
for container in ${to_deploy[@]}; do
message " $container: deploying..."
log="$REBUILD_DATA_DIR/$container.log"
$COMPOSE_BIN -f $DOCKERCOMPOSE_FILE up -d --no-deps $container > $log 2>&1
if [[ $? -eq 0 ]]; then
message " $container: done"
update_rebuild_status -d "$container"
rm -f $log
else
error " $container: fail to deploy new container image"
update_rebuild_status "$container" "error=fail to deploy new container image"
error=1
fi
done
message "done"
exit $error
fi
EXIT_CODE=0
declare -A CONTAINER_STATUS_FILE
@ -159,19 +332,13 @@ declare -A CONTAINER_PID
declare -A UP_TO_DATE
declare -A ERRORS
declare -A UNKNOWNS
UPGRADABLE_CONTAINERS=( )
declare -A UPGRADABLE_CONTAINERS
CHECKED_CONTAINERS=( )
debug "List running containers..."
if [ -n "$DOCKERCOMPOSE_FILE" ]
then
which "${ENGINE}-compose" > /dev/null 2>&1 \
&& COMPOSE_BIN="${ENGINE}-compose" \
|| COMPOSE_BIN="$ENGINE compose"
RUNNING_CONTAINERS=$($COMPOSE_BIN -f $DOCKERCOMPOSE_FILE ps --format '{{.Service}}' | tr '\n' ' ')
else
[[ -n "$DOCKERCOMPOSE_FILE" ]] && \
RUNNING_CONTAINERS=$($COMPOSE_BIN -f $DOCKERCOMPOSE_FILE ps --format '{{.Service}}' | tr '\n' ' ') ||
RUNNING_CONTAINERS=$($ENGINE ps --format '{{.Names}}' | tr '\n' ' ')
fi
debug "Running containers: $RUNNING_CONTAINERS"
function exec_in_container() {
@ -256,7 +423,8 @@ do
UP_TO_DATE+=( ["$container"]=$STATUS )
else
ERRORS+=( ["$container"]=$STATUS )
[ $ex -ge 3 ] && UNKNOWNS+=( "$container" ) || UPGRADABLE_CONTAINERS+=( "$container" )
[ $ex -ge 3 ] && UNKNOWNS+=( "$container" ) || \
UPGRADABLE_CONTAINERS+=( ["$container"]="$STATUS" )
fi
[ $EXIT_CODE -ge $ex ] && continue
[ $ex -gt 3 ] && ex=3
@ -282,7 +450,7 @@ debug "Final exit code: $EXIT_CODE"
debug "Check containers (${#CHECKED_CONTAINERS[@]}): $( implode ", " "${CHECKED_CONTAINERS[@]}" )"
debug "Up-to-date containers (${#UP_TO_DATE[@]}): $( implode ", " "${!UP_TO_DATE[@]}" )"
debug "Upgradable containers (${#UPGRADABLE_CONTAINERS[@]}): $( implode ", " "${UPGRADABLE_CONTAINERS[@]}" )"
debug "Upgradable containers (${#UPGRADABLE_CONTAINERS[@]}): $( implode ", " "${!UPGRADABLE_CONTAINERS[@]}" )"
debug "Containers with errors (${#ERRORS[@]}): $( implode ", " "${!ERRORS[@]}" )"
debug "Not found containers (${#NOTFOUNDS[@]}): $( implode ", " "${NOTFOUNDS[@]}" )"
@ -295,83 +463,116 @@ PERF_DATA=(
"unknown_state_containers=${#UNKNOWNS[@]};;;0;$CONTAINER_COUNTS"
)
# Compute performance data as string
PERF_DATA_TXT="$( implode " " "${PERF_DATA[@]}" )"
# Display check result message
case $EXIT_CODE in
0)
echo -n "OK - All ${#UP_TO_DATE[@]} container(s) are up-to-date"
message "OK - All ${#UP_TO_DATE[@]} container(s) are up-to-date |$PERF_DATA_TXT"
;;
1)
echo -n "WARNING - ${#ERRORS[@]} container(s) need to be updated"
message "WARNING - ${#ERRORS[@]} container(s) need to be updated |$PERF_DATA_TXT"
;;
2)
echo -n "CRITICAL - ${#ERRORS[@]} container(s) need to be updated"
message "CRITICAL - ${#ERRORS[@]} container(s) need to be updated |$PERF_DATA_TXT"
;;
*)
echo -n "UNKNOWN - fail to retrieve status of ${#UNKNOWNS[@]} container(s)"
message "UNKNOWN - fail to retrieve status of ${#UNKNOWNS[@]} container(s) |$PERF_DATA_TXT"
;;
esac
# Add performance data
echo " |$( implode " " "${PERF_DATA[@]}" )"
# Trigger container build (if need, enabled and docker compose file is provided)
if [ $REBUILD -eq 1 ]
then
debug "Check if we have to trigger some rebuild (status file: $(rebuild_status_file))"
if [ ${#UPGRADABLE_CONTAINERS[@]} -eq 0 ]
then
debug "No upgradable container to rebuild"
if [ -e "$REBUILD_LOCK_FILE" ]
then
debug "Remove previous rebuild lock file ($REBUILD_LOCK_FILE)"
rm -f "$REBUILD_LOCK_FILE"
fi
remove_rebuild_status
elif [ -z "$DOCKERCOMPOSE_FILE" ]
then
echo
echo "WARNING: No docker compose file provided, can't trigger rebuild of following container(s):"
echo "- $( implode "\n- " ${UPGRADABLE_CONTAINERS[@]} )"
elif [ -e "$REBUILD_LOCK_FILE" ]
then
REBUILD_PID=$(cat $REBUILD_LOCK_FILE)
echo
if [ -d "/proc/$REBUILD_PID" ]
then
echo "Rebuild already triggered on $( stat --format %y "$REBUILD_LOCK_FILE" ) (PID: $REBUILD_PID)"
echo "You could follow it in its log file: ${REBUILD_LOG_FILE}"
message
message "WARNING: No docker compose file provided, can't trigger rebuild of following" \
"container(s):"
message "- $( implode "\n- " ${UPGRADABLE_CONTAINERS[@]} )"
else
echo "Container images already rebuilt (see ${REBUILD_LOG_FILE} for details)."
echo "You could recreate and restart them using the following command:"
echo
echo " $COMPOSE_BIN -f $DOCKERCOMPOSE_FILE up -d --no-deps ${UPGRADABLE_CONTAINERS[@]}"
echo
echo "Note:"
echo " If new updates was published since rebuild was triggered, remove the"
echo " '$REBUILD_LOCK_FILE' file and recheck it to trigger a new build."
message "Rebuilding containers:"
REBUILT_CONTAINERS=()
for container in ${!UPGRADABLE_CONTAINERS[@]}; do
need_rebuild=0
container_data=$( rebuild_status "$container" )
if [[ "$container_data" != "null" ]]; then
debug "$container: data=$container_data"
trigger_date=$( jq -r .trigger_date <<< $container_data )
start_date=$( jq -r .start_date <<< $container_data )
end_date=$( jq -r .end_date <<< $container_data )
log=$( jq -r .log <<< $container_data )
if [[ "$start_date" == "null" ]]; then
debug "$container: build triggered but not yet started"
message "- $container: rebuild triggered on $trigger_date and not yet started"
elif [[ "$end_date" == "null" ]]; then
debug "$container: rebuild triggered on $trigger_date and started on" \
"$start_date, but not yet finish"
message "- $container: rebuild triggered on $trigger_date and started on" \
"$start_date, but not yet finish (log: $log)"
else
duration=$( jq -r .duration <<< $container_data )
debug "$container: rebuilt in $duration on $start_date (finish on $end_date)"
prev_status=$( jq -r .status <<< $container_data )
if [[ "$prev_status" == "${UPGRADABLE_CONTAINERS[$container]}" ]]; then
error=$( jq -r .error <<< $container_data )
if [[ "$error" != "null" ]]; then
message "- $container: $error (log: $log)"
else
message "- $container: already rebuilt in $duration (rebuild" \
"triggered on $trigger_date, started on $start_date and finish" \
"at $end_date,log: $log)"
REBUILT_CONTAINERS+=( $container )
fi
echo
else
echo
echo "Trigger rebuild of following container(s): ${UPGRADABLE_CONTAINERS[@]}"
echo "You could follow the rebuild process in its log file: ${REBUILD_LOG_FILE}"
echo "You will able to recreate and restart them using the following command:"
echo
echo " $COMPOSE_BIN -f $DOCKERCOMPOSE_FILE up -d --no-deps ${UPGRADABLE_CONTAINERS[@]}"
echo
update_rebuild_status -d "$container"
message "- $container: upgrade status change since last rebuild, rebuild" \
"it again"
need_rebuild=1
fi
fi
else
debug "$container: not found in status file"
need_rebuild=1
fi
if [[ $need_rebuild -eq 1 ]]; then
update_rebuild_status "$container" \
"trigger_date=$(now)" "status=${UPGRADABLE_CONTAINERS[$container]}"
message "- $container: rebuild triggered"
else
debug "$container: rebuild not need"
fi
done
$COMPOSE_BIN -f $DOCKERCOMPOSE_FILE build --no-cache ${UPGRADABLE_CONTAINERS[@]} > $REBUILD_LOG_FILE 2>&1 &
echo $! > "$REBUILD_LOCK_FILE"
# Handle rebuilt containers
if [[ ${#REBUILT_CONTAINERS[@]} -gt 0 ]]; then
message
message "Some containers are ready to be recreated and restarted."
message "Run the following command to do it:"
message
message " $COMPOSE_BIN -f $DOCKERCOMPOSE_FILE up -d --no-deps ${REBUILT_CONTAINERS[@]}"
fi
message
fi
fi
# Display details, starting by errors
for container in ${!ERRORS[@]}
do
echo ${container} - ${ERRORS[${container}]}
message ${container} - ${ERRORS[${container}]}
done
for container in ${!UP_TO_DATE[@]}
do
echo ${container} - ${UP_TO_DATE[${container}]}
message ${container} - ${UP_TO_DATE[${container}]}
done
exit $EXIT_CODE
# vim: shiftwidth=4 tabstop=4 expandtab