Merge lp:~allenap/maas/packaging--upstart-workers--bug-1452276 into lp:~maas-maintainers/maas/packaging

[Gavin Panella]

Launch 4 maas-regiond workers using Upstart instance jobs.

[Andres Rodriguez (andreserl)]

Hi Gavin,

My understanding is that we don't have to create a worker upstart job, but rather, do it within the same upstart job. Can you look itno that please? Otherwise reach out to infinity or someone in the foundations team to check with them.

review: Needs Information

[Gavin Panella (allenap)]

I can't get restarts to work when putting all the stuff in a single service definition. I've asked Adam Conrad for help. In the meantime please consider landing this. It's only a cosmetic difference, but it's costing a fortune in lost time for me. In addition, using separate service definitions is easier to reason about and understand; smooshing everything into one is somewhat hacky.

[lp:~allenap/maas/packaging--upstart-workers--bug-1452276]

392. By Gavin Panella on 2015-05-14

Remove the 'sleep hack' on the advice of jodh.

[Gavin Panella (allenap)]

James Hunt has said that "2 jobs is tbh recommended - it's much clearer: a manager and a bunch of workers."

He had some other useful advice about the 'sleep hack', i.e. that it's not needed. I've removed it, and I'll give this another run locally.

[Gavin Panella (allenap)]

> I'll give this another run locally.

It all seems to work nicely. Let's land this!

[Andres Rodriguez (andreserl)]

+1

review: Approve

=== modified file 'debian/control'
--- debian/control	2015-04-15 16:48:02 +0000
+++ debian/control	2015-05-14 13:16:20 +0000
@@ -49,7 +49,6 @@
 Architecture: all
 Depends: apache2,
          bind9utils,
-         daemontools,
          dbconfig-common,
          iproute,
          libjs-angularjs,
 
=== added file 'debian/maas-region-controller-min.maas-regiond-worker.upstart'
--- debian/maas-region-controller-min.maas-regiond-worker.upstart	1970-01-01 00:00:00 +0000
+++ debian/maas-region-controller-min.maas-regiond-worker.upstart	2015-05-14 13:16:20 +0000
@@ -0,0 +1,25 @@
+description "MAAS Region Controller (Worker)"
+author "Gavin Panella <[e-mail]>"
+
+# This instance job is started by the parent `maas-regiond` job, and
+# stops when that job is stopped. This is differentiated from its
+# siblings by the $core instance argument.
+
+instance $core
+
+stop on stopping maas-regiond
+
+setuid maas
+setgid maas
+
+respawn
+
+kill timeout 15
+
+env DJANGO_SETTINGS_MODULE=maas.settings
+env PYTHONPATH=/usr/share/maas
+env LOGFILE=/var/log/maas/regiond.log
+
+script
+    exec /usr/bin/twistd --nodaemon --pidfile= maas-regiond >> ${LOGFILE} 2>&1
+end script
 
=== modified file 'debian/maas-region-controller-min.maas-regiond.upstart'
--- debian/maas-region-controller-min.maas-regiond.upstart	2015-04-15 21:47:48 +0000
+++ debian/maas-region-controller-min.maas-regiond.upstart	2015-05-14 13:16:20 +0000
@@ -1,87 +1,17 @@
 description "MAAS Region Controller"
 author "Gavin Panella <[e-mail]>"
 
+# This starts multiple instances of `maas-regiond-worker` and does
+# nothing else itself; there is no main script so this runs "forever".
+# By default, it launches four workers.
+
 start on filesystem and net-device-up
 stop on runlevel [016]
 
-env svcdir=/var/lib/maas/regiond
-
-setuid maas
-setgid maas
-
 respawn
 
-# Give `supervise` time to exit in post-stop. It's not clear that this
-# time applies to pre/post scripts, so this may be for naught.
-kill timeout 15
-
 pre-start script
-    set -eu
-    # Create the service directory.
-    mkdir -p "${svcdir}"
-    # Create a `run` script in the service directory.
-    svcrun="${svcdir}/run"
-    echo "#!/bin/sh" > "${svcrun}"
-    echo "exec >> /var/log/maas/regiond.log 2>&1" >> "${svcrun}"
-    echo "export PYTHONPATH=/usr/share/maas" >> "${svcrun}"
-    echo "export DJANGO_SETTINGS_MODULE=maas.settings" >> "${svcrun}"
-    echo "exec /usr/bin/twistd --nodaemon --pidfile= maas-regiond" >> "${svcrun}"
-    chmod a+x "${svcrun}"
-    # We're going to create between 2 and 5 daemons, depending on core count.
-    daemons="$(nproc)"
-    if [ "${daemons}" -lt 2 ]
-    then
-        daemons=2
-    elif [ "${daemons}" -gt 5 ]
-    then
-        daemons=5
-    fi
-    # Create the run directories and link to the shared `run` script.
-    for daemon in $(seq "${daemons}")
-    do
-        rundir="${svcdir}/${daemon}"
-        mkdir -p "${rundir}"
-        runfile="${rundir}/run"
-        ln -snf "${svcrun}" "${runfile}"
-    done
-end script
-
-script
-    # In the case where this service has been restarted, we may have
-    # running `supervise` daemons. Bring each service up.
-    find "${svcdir}" -name run -executable \
-        -execdir svok . \; -execdir svc -u . \;
-    # In the case where this service has been restarted, we may also
-    # have leftover `down` markers. To remove these, we remove the
-    # `supervise` directories.
-    find "${svcdir}" -name supervise -type d \
-        -not -execdir svok . \; -execdir rm -rf {} \;
-    # Exec the `svscan` daemon, as the meta-supervisor.
-    exec svscan "${svcdir}"
-end script
-
-pre-stop script
-    # Bring each service down. We must *not* exit the `supervise` daemon
-    # because `svscan` may then have time in which to respawn it before
-    # it's stopped.
-    find "${svcdir}" -name run -executable -execdir svok . \; \
-        -printf "Bringing %h down.\n" -execdir svc -d . \;
-end script
-
-post-stop script
-    set -eu
-    # Bring each service down (belt-n-braces) and quit supervise; it's
-    # safe to do so now because `svscan` is guaranteed to have stopped.
-    find "${svcdir}" -name run -executable -execdir svok . \; \
-        -printf "Bringing %h down.\n" -execdir svc -dx . \;
-    # For up to 10 seconds -- less than the `kill timeout` -- check if
-    # any supervise daemons are still running every 0.1 seconds.
-    for attempt in $(seq 100)
-    do
-        # If one or more `supervise` daemons are still running, sleep.
-        find "${svcdir}" -name run -execdir svok . \; -exec sleep 0.1 \; -quit
-    done
-    # Kill with fire any remaining supervise daemons.
-    find "${svcdir}" -name run -executable -execdir svok . \; \
-        -printf "Killing %h; timed-out.\n" -execdir svc -k . \;
+    for core in $(seq 4); do
+        start maas-regiond-worker core=${core}
+    done
 end script
 
=== modified file 'debian/rules'
--- debian/rules	2015-04-15 16:48:02 +0000
+++ debian/rules	2015-05-14 13:16:20 +0000
@@ -17,6 +17,8 @@
 	dh_systemd_enable --name maas-regiond
 	dh_installinit --name maas-regiond
 	dh_systemd_start --name maas-regiond
+	# maas-regiond-worker
+	dh_installinit --name maas-regiond-worker
 	# maas-dhcpd
 	dh_systemd_enable --name maas-dhcpd
 	dh_installinit --name maas-dhcpd