From 8c78dc71ed5650638e1fff2d6e67adf22e2ccb72 Mon Sep 17 00:00:00 2001
From: Valentin Lab <valentin.lab@kalysto.org>
Date: Tue, 29 Nov 2022 18:33:10 +0100
Subject: [PATCH] new: [check-fix] new action to launch all fixes

A new fix was also added to find and fix `no-matching-entries` issue.
By default, all containers are checked against all checks.
---
 README.org | 29 ++++++++++++++++++++
 bin/vps    | 79 +++++++++++++++++++++++++++++++++++++++++++++++-------
 2 files changed, 99 insertions(+), 9 deletions(-)

diff --git a/README.org b/README.org
index add7650..3591e79 100644
--- a/README.org
+++ b/README.org
@@ -311,6 +311,35 @@ compose --debug logs odoo
 docker-ip
 #+END_SRC
 
+*** Vérification de santé générale
+
+La commande =vps check-fix= contrôler les containers en cours de
+fonctionnement pour vérifier s'ils ne sont pas touché par quelques
+problème identifiés et connus. Et elle va réparer ce qu'elle peut.
+
+#+begin_src sh
+vps check-fix
+#+end_src
+
+Il est possible de lancer ces vérifications sur une liste de service
+spécifique ou de sélectionner le test voulu. Voir =vps check-fix
+--help= pour plus d'info.
+
+Il peut être opportun d'ajouter dans sur l'hôte, par exemple, une
+vérification périodique et automatique:
+
+#+begin_src sh
+cat <<EOF > /etc/cron.d/vps-check
+SHELL=/bin/bash
+PATH=/usr/local/sbin:/usr/local/bin:/sbin:/bin:/usr/sbin:/usr/bin
+
+*/15 * * * *   root   vps check-fix -s -c container-aliveness 2>&1 | logger -t check-fix
+
+*/5 * * * *    root   vps check-fix -s rocketchat 2>&1 | logger -t check-fix
+EOF
+#+end_src
+
+
 ** Par services
 
 *** odoo
diff --git a/bin/vps b/bin/vps
index 21cdc42..71555e7 100755
--- a/bin/vps
+++ b/bin/vps
@@ -936,7 +936,7 @@ nextcloud:src:version() {
 }
 
 
-container:health:check-fix() {
+container:health:check-fix:container-aliveness() {
     local container_id="$1"
 
     timeout 5s docker inspect "$container_id" >/dev/null 2>&1
@@ -963,6 +963,32 @@ EOF
 }
 
 
+container:health:check-fix:no-matching-entries() {
+    local container_id="$1"
+
+    out=$(docker exec "$container_id" echo 2>&1)
+    errlvl=$?
+    [ "$errlvl" == 0 ] && return 0
+    service_name=$(docker ps --filter id="$container_id" --format '{{.Label "com.docker.compose.service"}}')
+    container_name=($(docker ps --filter id="$container_id" --format '{{.Names}}'))
+    if [ "$errlvl" == 126 ] && [[ "$out" == *"no matching entries in passwd file"* ]]; then
+        echo "container ${DARKCYAN}${container_name[0]}${NORMAL} for ${DARKYELLOW}$service_name${NORMAL} has ${DARKRED}no-matching-entries${NORMAL} bug." >&2
+        Wrap -d "restarting container of ${DARKYELLOW}$service_name${NORMAL} twice" <<EOF
+docker restart "$container_id"
+sleep 2
+docker restart "$container_id"
+EOF
+        return $errlvl
+    fi
+    warn "Unknown issue with ${DARKYELLOW}$service_name${NORMAL}'s container:"
+    echo "  ${WHITE}cmd:${NORMAL} docker exec -ti $container_id echo" >&2
+    echo "$out" | prefix "  ${DARKGRAY}|${NORMAL} " >&2
+    echo "  ${DARKGRAY}..${NORMAL} leaving this as-is."
+    return $errlvl
+}
+
+
+
 
 [ "$SOURCED" ] && return 0
 
@@ -1611,22 +1637,57 @@ cmdline.spec:nextcloud:cmd:upgrade:run() {
 
 
 cmdline.spec.gnu check-fix
-
 cmdline.spec::cmd:check-fix:run() {
-    :
-}
 
-cmdline.spec.gnu upgrade
-cmdline.spec:check-fix:cmd:container-aliveness:run() {
+    : :posarg: [SERVICES...]  "Optional service to check"
+    : :optval: --check,-c     "Specify a check or a list of checks separated by commas"
+    : :optfla: --silent,-s    "Don't ouput anything if everything goes well"
 
-    local project_name containers container
+    local project_name service_name containers container check
 
+    all_checks=$(declare -F |
+                     egrep '^declare -fx? container:health:check-fix:[^ ]+$' |
+                     cut -f 4 -d ":")
+    checks=(${opt_check//,/ })
+    for check in "${checks[@]}"; do
+        fn.exists container:health:check-fix:$check || {
+            err "check '$check' not found."
+            return 1
+        }
+    done
+    if [ "${#checks[*]}" == 0 ]; then
+        checks=($all_checks)
+    fi
+    ## XXXvlab: could make it parallel
     project_name=$(compose:project_name) || exit 1
     containers=($(compose:project:containers "${project_name}")) || exit 1
-    ## XXXvlab: could make it parallel
+
+    found=
     for container in "${containers[@]}"; do
-        container:health:check-fix "$container"
+        service_name=$(docker ps --filter id="$container" --format '{{.Label "com.docker.compose.service"}}')
+        if [ "${#SERVICES[@]}" -gt 0 ]; then
+            [[ " ${SERVICES[*]} " == *" $service_name "* ]] || continue
+        fi
+        found=1
+        one_bad=
+        for check in "${checks[@]}"; do
+            if ! container:health:check-fix:"$check" "$container"; then
+                one_bad=1
+            fi
+        done
+        if [ -z "$opt_silent" ] && [ -z "$one_bad" ]; then
+            Elt "containers have been checked for ${DARKYELLOW}$service_name${NORMAL}"
+            Feedback
+        fi
     done
+    if [ -z "$found" ]; then
+        if [ "${#SERVICES[@]}" -gt 0 ]; then
+            warn "No container for given services found in current project '$project_name'."
+        else
+            warn "No container found for current project '$project_name'."
+        fi
+        return 1
+    fi
 
 }