Merge pull request #11955 from shanemcd/fail-better

Increase resiliency when application crashes

awx/main/dispatch/reaper.py

@@ -35,9 +35,11 @@ def reap(instance=None, status='failed', excluded_uuids=[]):
     """
     me = instance
     if me is None:
-        (changed, me) = Instance.objects.get_or_register()
+        try:
-        if changed:
+            me = Instance.objects.me()
-            logger.info("Registered node '{}'".format(me.hostname))
+        except RuntimeError as e:
+            logger.warning(f'Local instance is not registered, not running reaper: {e}')
+            return
     now = tz_now()
     workflow_ctype_id = ContentType.objects.get_for_model(WorkflowJob).id
     jobs = UnifiedJob.objects.filter(

awx/main/management/commands/provision_instance.py

@@ -3,6 +3,7 @@
 
 from django.core.management.base import BaseCommand, CommandError
 from django.db import transaction
+from django.conf import settings
 
 from awx.main.models import Instance
 
@@ -13,7 +14,7 @@ class Command(BaseCommand):
     Register this instance with the database for HA tracking.
     """
 
-    help = "Add instance to the database. Specify `--hostname` to use this command."
+    help = "Add instance to the database. When no options are provided, the hostname of the current system will be used. Override with `--hostname`."
 
     def add_arguments(self, parser):
         parser.add_argument('--hostname', dest='hostname', type=str, help="Hostname used during provisioning")
@@ -22,8 +23,11 @@ class Command(BaseCommand):
 
     def _register_hostname(self, hostname, node_type, uuid):
         if not hostname:
-            return
+            if not settings.AWX_AUTO_DEPROVISION_INSTANCES:
-        (changed, instance) = Instance.objects.register(hostname=hostname, node_type=node_type, uuid=uuid)
+                raise CommandError('Registering with values from settings only intended for use in K8s installs')
+            (changed, instance) = Instance.objects.get_or_register()
+        else:
+            (changed, instance) = Instance.objects.register(hostname=hostname, node_type=node_type, uuid=uuid)
         if changed:
             print("Successfully registered instance {}".format(hostname))
         else:
@@ -32,8 +36,6 @@ class Command(BaseCommand):
 
     @transaction.atomic
     def handle(self, **options):
-        if not options.get('hostname'):
-            raise CommandError("Specify `--hostname` to use this command.")
         self.changed = False
         self._register_hostname(options.get('hostname'), options.get('node_type'), options.get('uuid'))
         if self.changed:

awx/main/tasks/system.py

@@ -490,10 +490,6 @@ def cluster_node_heartbeat():
         if inst.hostname == settings.CLUSTER_HOST_ID:
             this_inst = inst
             break
-    else:
-        (changed, this_inst) = Instance.objects.get_or_register()
-        if changed:
-            logger.info("Registered tower control node '{}'".format(this_inst.hostname))
 
     inspect_execution_nodes(instance_list)
 

tools/ansible/roles/dockerfile/files/launch_awx.sh

@@ -17,4 +17,11 @@ set -e
 
 wait-for-migrations
 
-supervisord -c /etc/supervisord.conf
+# This file will be re-written when the dispatcher calls reconfigure_rsyslog(),
+# but it needs to exist when supervisor initially starts rsyslog to prevent the
+# container from crashing. This was the most minimal config I could get working.
+cat << EOF > /var/lib/awx/rsyslog/rsyslog.conf
+action(type="omfile" file="/dev/null")
+EOF
+
+exec supervisord -c /etc/supervisord.conf

tools/ansible/roles/dockerfile/files/launch_awx_task.sh

@@ -17,4 +17,6 @@ set -e
 
 wait-for-migrations
 
-supervisord -c /etc/supervisord_task.conf
+awx-manage provision_instance
+
+exec supervisord -c /etc/supervisord_task.conf

tools/ansible/roles/dockerfile/files/stop-supervisor

@@ -0,0 +1,8 @@
+#!/bin/bash
+
+printf "READY\n";
+
+while read line; do
+  echo "Processing Event: $line" >&2;
+  kill -SIGQUIT $PPID
+done < /dev/stdin

tools/ansible/roles/dockerfile/templates/Dockerfile.j2

@@ -120,11 +120,7 @@ RUN curl -fsSL -o get_helm.sh https://raw.githubusercontent.com/helm/helm/master
     chmod 700 get_helm.sh && \
     ./get_helm.sh
 
-# Install tini
+RUN pip3 install virtualenv supervisor dumb-init
-RUN curl -L -o /usr/bin/tini https://github.com/krallin/tini/releases/download/v0.19.0/tini-{{ tini_architecture | default('amd64') }} && \
-    chmod +x /usr/bin/tini
-
-RUN pip3 install virtualenv supervisor
 
 RUN rm -rf /root/.cache && rm -rf /tmp/*
 
@@ -194,6 +190,7 @@ RUN mkdir -p /etc/containers/registries.conf.d/ && echo "unqualified-search-regi
 # Create default awx rsyslog config
 ADD tools/ansible/roles/dockerfile/files/rsyslog.conf /var/lib/awx/rsyslog/rsyslog.conf
 ADD tools/ansible/roles/dockerfile/files/wait-for-migrations /usr/local/bin/wait-for-migrations
+ADD tools/ansible/roles/dockerfile/files/stop-supervisor /usr/local/bin/stop-supervisor
 
 ## File mappings
 {% if build_dev|bool %}
@@ -292,7 +289,7 @@ CMD ["/bin/bash"]
 USER 1000
 EXPOSE 8052
 
-ENTRYPOINT ["/usr/bin/tini", "--"]
+ENTRYPOINT ["dumb-init", "--"]
 CMD /usr/bin/launch_awx.sh
 VOLUME /var/lib/nginx
 VOLUME /var/lib/awx/.local/share/containers

tools/ansible/roles/dockerfile/templates/supervisor.conf.j2

@@ -12,9 +12,10 @@ directory = /awx_devel
 {% else %}
 command = nginx -g "daemon off;"
 {% endif %}
-autostart = true
 autorestart = true
-stopwaitsecs = 5
+startsecs = 30
+stopasgroup=true
+killasgroup=true
 stdout_logfile=/dev/stdout
 stdout_logfile_maxbytes=0
 stderr_logfile=/dev/stderr
@@ -31,12 +32,10 @@ environment =
 command = /var/lib/awx/venv/awx/bin/uwsgi /etc/tower/uwsgi.ini
 directory = /var/lib/awx
 {% endif %}
-autostart = true
 autorestart = true
-stopwaitsecs = 15
+startsecs = 30
 stopasgroup=true
 killasgroup=true
-stopsignal=KILL
 stdout_logfile=/dev/stdout
 stdout_logfile_maxbytes=0
 stderr_logfile=/dev/stderr
@@ -50,10 +49,8 @@ directory = /awx_devel
 command = /var/lib/awx/venv/awx/bin/daphne -b 127.0.0.1 -p 8051 --websocket_timeout -1 awx.asgi:channel_layer
 directory = /var/lib/awx
 {% endif %}
-autostart = true
-stopsignal=KILL
 autorestart = true
-stopwaitsecs = 5
+startsecs = 30
 stopasgroup=true
 killasgroup=true
 stdout_logfile=/dev/stdout
@@ -69,9 +66,8 @@ directory = /awx_devel
 command = awx-manage run_wsbroadcast
 directory = /var/lib/awx
 {% endif %}
-autostart = true
 autorestart = true
-stopwaitsecs = 5
+startsecs = 30
 stopasgroup=true
 killasgroup=true
 stdout_logfile=/dev/stdout
@@ -81,33 +77,26 @@ stderr_logfile_maxbytes=0
 
 [program:awx-rsyslogd]
 command = rsyslogd -n -i /var/run/awx-rsyslog/rsyslog.pid -f /var/lib/awx/rsyslog/rsyslog.conf
-autostart = true
 autorestart = true
-startretries = 10
+startsecs = 30
-stopwaitsecs = 5
-stopsignal=TERM
 stopasgroup=true
 killasgroup=true
-redirect_stderr=true
+stdout_logfile=/dev/stdout
-stdout_logfile=/dev/stderr
 stdout_logfile_maxbytes=0
+stderr_logfile=/dev/stderr
+stderr_logfile_maxbytes=0
 
 [group:tower-processes]
 programs=nginx,uwsgi,daphne,wsbroadcast,awx-rsyslogd
 priority=5
 
-# TODO: Exit Handler
+[eventlistener:superwatcher]
-
+command=stop-supervisor
-{% if kube_dev | bool %}
+events=PROCESS_STATE_FATAL
-[eventlistener:awx-config-watcher]
-command=/usr/bin/config-watcher
-stderr_logfile=/dev/stdout
-stderr_logfile_maxbytes=0
 stdout_logfile=/dev/stdout
 stdout_logfile_maxbytes=0
-events=TICK_60
+stderr_logfile=/dev/stderr
-priority=0
+stderr_logfile_maxbytes=0
-{% endif %}
 
 [unix_http_server]
 file=/var/run/supervisor/supervisor.web.sock

tools/ansible/roles/dockerfile/templates/supervisor_task.conf.j2

@@ -13,9 +13,8 @@ directory = /awx_devel
 command = awx-manage run_dispatcher
 directory = /var/lib/awx
 {% endif %}
-autostart = true
 autorestart = true
-stopwaitsecs = 5
+startsecs = 30
 stopasgroup=true
 killasgroup=true
 stdout_logfile=/dev/stdout
@@ -31,9 +30,8 @@ directory = /awx_devel
 command = awx-manage run_callback_receiver
 directory = /var/lib/awx
 {% endif %}
-autostart = true
 autorestart = true
-stopwaitsecs = 5
+startsecs = 30
 stopasgroup=true
 killasgroup=true
 stdout_logfile=/dev/stdout
@@ -45,18 +43,14 @@ stderr_logfile_maxbytes=0
 programs=dispatcher,callback-receiver
 priority=5
 
-# TODO: Exit Handler
+[eventlistener:superwatcher]
-
+command=stop-supervisor
-{% if kube_dev | bool %}
+events=PROCESS_STATE_FATAL
-[eventlistener:awx-config-watcher]
+autorestart = true
-command=/usr/bin/config-watcher
-stderr_logfile=/dev/stdout
-stderr_logfile_maxbytes=0
 stdout_logfile=/dev/stdout
 stdout_logfile_maxbytes=0
-events=TICK_60
+stderr_logfile=/dev/stderr
-priority=0
+stderr_logfile_maxbytes=0
-{% endif %}
 
 [unix_http_server]
 file=/var/run/supervisor/supervisor.sock

tools/docker-compose/launch_awx.sh

@@ -5,4 +5,4 @@ bootstrap_development.sh
 
 cd /awx_devel
 # Start the services
-exec tini -- make supervisor
+exec make supervisor

tools/docker-compose/supervisor.conf

@@ -5,79 +5,75 @@ nodaemon=true
 
 [program:awx-dispatcher]
 command = make dispatcher
-autostart = true
 autorestart = true
-stopwaitsecs = 1
+startsecs = 30
-stopsignal=KILL
 stopasgroup=true
 killasgroup=true
-redirect_stderr=true
+stdout_logfile=/dev/stdout
-stdout_events_enabled = true
+stdout_logfile_maxbytes=0
-stderr_events_enabled = true
+stderr_logfile=/dev/stderr
-
+stderr_logfile_maxbytes=0
 
 [program:awx-receiver]
 command = make receiver
-autostart = true
 autorestart = true
-stopwaitsecs = 1
+startsecs = 30
-stopsignal=KILL
 stopasgroup=true
 killasgroup=true
-redirect_stderr=true
+stdout_logfile=/dev/stdout
-stdout_events_enabled = true
+stdout_logfile_maxbytes=0
-stderr_events_enabled = true
+stderr_logfile=/dev/stderr
+stderr_logfile_maxbytes=0
 
 [program:awx-wsbroadcast]
 command = make wsbroadcast
-autostart = true
 autorestart = true
-stopwaitsecs = 1
+startsecs = 30
-stopsignal=KILL
+autorestart = true
 stopasgroup=true
 killasgroup=true
-redirect_stderr=true
+stdout_logfile=/dev/stdout
-stdout_events_enabled = true
+stdout_logfile_maxbytes=0
-stderr_events_enabled = true
+stderr_logfile=/dev/stderr
+stderr_logfile_maxbytes=0
 
 [program:awx-uwsgi]
 command = make uwsgi
-autostart = true
 autorestart = true
-redirect_stderr=true
+startsecs = 30
-stopwaitsecs = 1
-stopsignal=KILL
 stopasgroup=true
 killasgroup=true
-stdout_events_enabled = true
+stdout_logfile=/dev/stdout
-stderr_events_enabled = true
+stdout_logfile_maxbytes=0
+stderr_logfile=/dev/stderr
+stderr_logfile_maxbytes=0
 
 [program:awx-daphne]
 command = make daphne
-autostart = true
 autorestart = true
-redirect_stderr=true
+startsecs = 30
-stopwaitsecs = 1
-stopsignal=KILL
 stopasgroup=true
 killasgroup=true
-stdout_events_enabled = true
+stdout_logfile=/dev/stdout
-stderr_events_enabled = true
+stdout_logfile_maxbytes=0
+stderr_logfile=/dev/stderr
+stderr_logfile_maxbytes=0
 
 [program:awx-nginx]
 command = make nginx
-autostart = true
 autorestart = true
-redirect_stderr=true
+startsecs = 30
-stdout_events_enabled = true
+stopasgroup=true
-stderr_events_enabled = true
+killasgroup=true
+stdout_logfile=/dev/stdout
+stdout_logfile_maxbytes=0
+stderr_logfile=/dev/stderr
+stderr_logfile_maxbytes=0
 
 [program:awx-rsyslogd]
 command = rsyslogd -n -i /var/run/awx-rsyslog/rsyslog.pid -f /var/lib/awx/rsyslog/rsyslog.conf
-autostart = true
 autorestart = true
-stopwaitsecs = 5
+startsecs = 30
-stopsignal=TERM
 stopasgroup=true
 killasgroup=true
 redirect_stderr=true
@@ -86,19 +82,28 @@ stderr_events_enabled = true
 
 [program:awx-receptor]
 command = receptor --config /etc/receptor/receptor.conf
-autostart = true
 autorestart = true
-stopsignal = KILL
+startsecs = 30
-stopasgroup = true
+stopasgroup=true
-killasgroup = true
+killasgroup=true
-redirect_stderr=true
+stdout_logfile=/dev/stdout
-stdout_events_enabled = true
+stdout_logfile_maxbytes=0
-stderr_events_enabled = true
+stderr_logfile=/dev/stderr
+stderr_logfile_maxbytes=0
 
 [group:tower-processes]
 programs=awx-dispatcher,awx-receiver,awx-uwsgi,awx-daphne,awx-nginx,awx-wsbroadcast,awx-rsyslogd
 priority=5
 
+[eventlistener:superwatcher]
+command=stop-supervisor
+events=PROCESS_STATE_FATAL
+autorestart = true
+stderr_logfile=/dev/stdout
+stderr_logfile_maxbytes=0
+stdout_logfile=/dev/stdout
+stdout_logfile_maxbytes=0
+
 [unix_http_server]
 file=/var/run/supervisor/supervisor.sock