Merge pull request #259 from City-of-Helsinki/UHF-10354-cron-failures

UHF-10354: Check cron pod failure codes

docker/openshift/Dockerfile

@@ -14,7 +14,8 @@ COPY docker/openshift/init.sh /
 # Copy cron scripts
 RUN mkdir /crons
 COPY docker/openshift/crons/ /crons
-RUN chmod +x /crons/*
+COPY docker/openshift/cron-entrypoint.sh /usr/local/bin/cron-entrypoint
+RUN chmod +x /crons/* /usr/local/bin/cron-entrypoint
 
 # Copy nginx overrides.
 COPY docker/openshift/custom.locations /etc/nginx/conf.d/custom.locations

docker/openshift/cron-entrypoint.sh

@@ -0,0 +1,42 @@
+#!/bin/bash
+
+source /init.sh
+
+echo "Starting cron: $(date)"
+
+# You can add any additional cron "daemons" to docker/openshift/crons/ folder.
+#
+# Example:
+# @code
+# #!/bin/bash
+# while true
+# do
+#   drush some-command
+#   sleep 600
+# done
+# @endcode
+
+for cron in /crons/*.sh; do
+  # Skip legacy base.sh script if it exists.
+  # Skip Kubernetes hooks that are stored in crons directory.
+  if [[ "${cron##*/}" == "base.sh" ]] || [[ "${cron##*/}" == *-hook.sh ]]; then
+    continue
+  elif [[ -r "$cron" ]]; then
+    echo "Starting $cron"
+    exec "$cron" &
+  fi
+done
+
+while true
+do
+  # Rudimentary process supervisor:
+  # Waits for the next process to terminate. The parent
+  # process is killed if any subprocess exists with failure.
+  # OpenShift should then restart the cron pod.
+  wait -n
+  exit_code=$?
+  if [[ "$exit_code" -ne 0 ]]; then
+    output_error_message "Cron subprocess failed with exit code $exit_code"
+    exit 1
+  fi
+done

docker/openshift/crons/cron.sh

@@ -0,0 +1,9 @@
+#!/bin/bash
+
+while true
+do
+  echo "Running cron: $(date +'%Y-%m-%dT%H:%M:%S%:z')\n"
+  drush cron
+  # Sleep for 10 minutes.
+  sleep 600
+done

docker/openshift/crons/linked-events.sh

@@ -1,5 +1,11 @@
 #!/bin/bash
 
+source /init.sh
+
+if ! is_drupal_module_enabled "helfi_react_search"; then
+  exit 0
+fi
+
 while true
 do
   # Allow migrations to be run every 3 hours and reset stuck migrations every 12 hours.

docker/openshift/crons/migrate-tpr.sh

@@ -1,5 +1,11 @@
 #!/bin/bash
 
+source /init.sh
+
+if ! is_drupal_module_enabled "helfi_tpr"; then
+  exit 0
+fi
+
 while true
 do
   # Allow migrations to be run every 3 hours and reset stuck migrations every 12 hours.

docker/openshift/crons/pubsub.sh

@@ -2,14 +2,9 @@
 
 echo "Running PubSub daemon: $(date +'%Y-%m-%dT%H:%M:%S%:z')"
 
-i=0
-# Attempt to start this service five times.
-until [ $i -gt 5 ]
+while true
 do
-  drush helfi:azure:pubsub-listen
-
-  if [[ "$?" -ne 0 ]]; then
-    ((i=i+1))
-    sleep 10
-  fi
+  # PubSub process exists with success return code after
+  # certain number of messages and should then be restarted.
+  drush helfi:azure:pubsub-listen || exit 1
 done

docker/openshift/init.sh

@@ -30,6 +30,14 @@ function deployment_in_progress {
   return 1
 }
 
+function is_drupal_module_enabled {
+  if drush pm-list --status=Enabled --filter=${1} --format=json | jq --exit-status '. == []' > /dev/null; then
+    return 1
+  fi
+
+  return 0
+}
+
 if [ ! -d "sites/default/files" ]; then
   output_error_message "Container start error: Public file folder does not exist. Exiting early."
   exit 1