Split bodhi-backend stuff out amongst the two nodes.

handlers/restart_services.yml

@@ -22,6 +22,9 @@
 
 - name: restart fedmsg-hub
   command: /usr/local/bin/conditional-restart.sh fedmsg-hub fedmsg-hub
+  # Note that, we're cool with arbitrary restarts on bodhi-backend02, just
+  # not bodhi-backend01.  01 is where the releng/mash stuff happens and we
+  # don't want to interrupt that.
   when: inventory_hostname != 'bodhi-backend01.phx2.fedoraproject.org'
 
 - name: restart fedmsg-irc

inventory/group_vars/bodhi-backend

@@ -72,26 +72,4 @@ nfs_mount_opts: "rw,hard,bg,intr,noatime,nodev,nosuid,nfsvers=3"
 fas_client_groups: sysadmin-releng
 sudoers: "{{ private }}/files/sudo/arm-releng-sudoers"
 
-
-# For the MOTD
-csi_security_category: Medium
-csi_primary_contact: Releng Admins sysadmin-releng-members@fedoraproject.org
-csi_purpose: Run the Bodhi masher.
-csi_relationship: |
-    The mashing of repos here happens as part of the 'fedmsg-hub' daemon.  Check
-    logs with 'journalctl -u fedmsg-hub'.  Check the bodhi masher docs/code for
-    more detail on what it does:
-    https://github.com/fedora-infra/bodhi/blob/develop/bodhi/consumers/masher.py
-
-    * This host relies on:
-      * db01 for its database, which is shares with the bodhi2 frontend nodes.
-      * An NFS mount of koji data in /mnt/koji/
-      * The fedmsg bus for triggering mashes.
-      * XMLRPC calls to koji for tagging and untagging updates.
-      * bugzilla for posting comments about status changes
-      * the wiki for getting information about QA "Test Cases"
-      * taksotron (resultsdb) for getting status-check results (gating updates).
-
-    * No other systems rely directly on this host.  Everything depends on it
-      indirectly for the creation of new updates repos (which get synced out to
-      the master mirror for distribution.
+## XXX - note that the csi_ stuff is kept at the host_vars/ level.

inventory/group_vars/bodhi2

@@ -79,7 +79,7 @@ csi_relationship: |
       * taksotron (resultsdb) for getting status-check results (gating updates).
 
     * It also depends on these things, but we're trying to move them exclusively
-      to the backend.
+      to bodhi-backend02.
       * koji for tagging and untagging updates and listing candidate builds
       * bugzilla, for getting bug title information and for posting comments
         about status changes

inventory/host_vars/bodhi-backend01.phx2.fedoraproject.org

@@ -8,3 +8,27 @@ volgroup: /dev/vg_Server
 eth0_ip: 10.5.125.135
 eth1_ip: 10.5.127.61
 vmhost: bvirthost10.phx2.fedoraproject.org
+
+
+# For the MOTD
+csi_security_category: Medium
+csi_primary_contact: Releng Admins sysadmin-releng-members@fedoraproject.org
+csi_purpose: Run the Bodhi masher.
+csi_relationship: |
+    The mashing of repos here happens as part of the 'fedmsg-hub' daemon.  Check
+    logs with 'journalctl -u fedmsg-hub'.  Check the bodhi masher docs/code for
+    more detail on what it does:
+    https://github.com/fedora-infra/bodhi/blob/develop/bodhi/consumers/masher.py
+
+    * This host relies on:
+      * db01 for its database, which is shares with the bodhi2 frontend nodes.
+      * An NFS mount of koji data in /mnt/koji/
+      * The fedmsg bus for triggering mashes.
+      * XMLRPC calls to koji for tagging and untagging updates.
+      * bugzilla for posting comments about status changes
+      * the wiki for getting information about QA "Test Cases"
+      * taksotron (resultsdb) for getting status-check results (gating updates).
+
+    * No other systems rely directly on this host.  Everything depends on it
+      indirectly for the creation of new updates repos (which get synced out to
+      the master mirror for distribution.

inventory/host_vars/bodhi-backend02.phx2.fedoraproject.org

@@ -8,3 +8,23 @@ volgroup: /dev/vg_bvirthost06
 eth0_ip: 10.5.125.136
 eth1_ip: 10.5.127.62
 vmhost: bvirthost06.phx2.fedoraproject.org
+
+
+# For the MOTD
+csi_security_category: Low
+csi_primary_contact: Bodhi Admins bodhiadmin-members@fedoraproject.org
+csi_purpose: Run the Bodhi UpdatesHandler and other backend processes
+csi_relationship: |
+    This host handles all of the bodhi backend stuff **except** for mashing.
+    https://github.com/fedora-infra/bodhi/blob/develop/bodhi/consumers/
+
+    * This host relies on:
+      * db01 for its database, which is shares with the bodhi2 frontend nodes.
+      * The fedmsg bus for triggering work.
+      * XMLRPC calls to koji for tagging and untagging updates.
+      * bugzilla for posting comments about status changes
+      * the wiki for getting information about QA "Test Cases"
+
+    * No other systems rely directly on this host.  Everything depends on it
+      indirectly for the creation of new updates repos (which get synced out to
+      the master mirror for distribution.

roles/bodhi2/base/templates/updates_handler.py

@@ -1,6 +1,7 @@
 config = dict(
-    # TODO -- we're thinking about moving this to bodhi-backend02
-    {% if inventory_hostname.startswith('bodhi-backend01') %}
+    # Note, the masher runs on bodhi-backend01, but this
+    # runs on bodhi-backend02 (separation of concerns).
+    {% if inventory_hostname.startswith('bodhi-backend02') %}
     updates_handler=True,
     {% else %}
     updates_handler=False,