fixed bug in unified_model regarding the new email routines.
authorStephen Soltesz <soltesz@cs.princeton.edu>
Tue, 4 Nov 2008 16:19:45 +0000 (16:19 +0000)
committerStephen Soltesz <soltesz@cs.princeton.edu>
Tue, 4 Nov 2008 16:19:45 +0000 (16:19 +0000)
two other tweaks in monitor-server.cron and added a new condition for rins in
bootman.py

bootman.py
monitor-server.cron
unified_model.py

index 82ee201..8a1baf5 100755 (executable)
@@ -587,6 +587,9 @@ def reboot(hostname, config=None, forced_action=None):
                        "bminit-cfg-auth-getplc-update-hardware-installinit-installdisk-installbootfs-installcfg-installstop-update-installinit-validate-rebuildinitrd-netcfg-disk-update4-update3-update3-kernelcopyfail-exception-update-debug-done",
                        "bminit-cfg-auth-getplc-hardware-installinit-installdisk-installbootfs-installcfg-installstop-update-installinit-validate-rebuildinitrd-netcfg-disk-update4-update3-update3-kernelcopyfail-exception-update-debug-done",
                        "bminit-cfg-auth-getplc-installinit-validate-exception-noinstall-update-debug-done",
+                       # actual solution appears to involve removing the bad files, and
+                       # continually trying to boot the node.
+                       "bminit-cfg-auth-getplc-update-installinit-validate-rebuildinitrd-netcfg-disk-update4-update3-update3-implementerror-update-debug-done",
                        ]:
                sequences.update({n : "restart_bootmanager_rins"})
 
index dc5302e..1433b79 100644 (file)
@@ -1,5 +1,6 @@
 # Runs every three hours to poll all nodes and PCUs, as well as take some
 # automated actions for debug nodes.
 
-01 6,9,12,15,18,21 * * * root /usr/share/monitor-server/automate.sh
+01 * * * * root /usr/share/monitor-server/automate.sh 2>&1 > /usr/share/monitor-server/monitor.log
+
 
index 76fb705..86c415b 100755 (executable)
@@ -564,12 +564,12 @@ class Record(object):
                        contacts += [config.email]
                if TECH & roles:
                        #contacts += [TECHEMAIL % self.loginbase]
-                       contacts += plc.getTechEmails(loginbase)
+                       contacts += plc.getTechEmails(self.loginbase)
                if PI & roles:
                        #contacts += [PIEMAIL % self.loginbase]
-                       contacts += plc.getSliceUserEmails(loginbase)
+                       contacts += plc.getSliceUserEmails(self.loginbase)
                if USER & roles:
-                       contacts += plc.getSliceUserEmails(loginbase)
+                       contacts += plc.getSliceUserEmails(self.loginbase)
                        slices = plc.slices(self.loginbase)
                        if len(slices) >= 1:
                                #for slice in slices: