[Xen-devel] elapse time computing when restarting VM?

[Yufang Zhang <yuzhang@xxxxxxxxxx>]

Hi all,
Currently, xend would compute elapse time since vm starts before restarting a 
vm. If the elapse time is larger than MINIMUM_RESTART_TIME (which is 60s), xend 
would refuse to restart the vm but destroy it to avoid loops. However, when a 
guest crashes at boot time and enable-dump is enabled, core dump is done before 
restarting the guest which may take quite a while (depends on memory size of 
the guest). At this situation, elapse time computed is expanded thus xend 
wouldn't destory the guest. Then the guest drops into a restart-crash-dumpcore 
loop, which is either a waist of cpu time or *disk space* of Domain0.  
Actually, I have hit this problem when I upgraded a 2048M guest to a 
problematic kernel. The guest crashed at boot time and core dump was done for 
it, after which the guest rebooted and go-through the previous steps. My 
domain0 was full of core dump files of that guest. So does it make sense to 
figure out a way to solve the problem but not just enlarging 
MINIMUM_RESTART_TIME? Is the following patch reasonable? 

diff -r 774dfc178c39 tools/python/xen/xend/XendDomainInfo.py
--- a/tools/python/xen/xend/XendDomainInfo.py   Thu Aug 12 17:06:21 2010 +0100
+++ b/tools/python/xen/xend/XendDomainInfo.py   Mon Aug 16 12:16:45 2010 +0800
@@ -2060,7 +2060,7 @@
                 log.warn('Domain has crashed: name=%s id=%d.',
                          self.info['name_label'], self.domid)
                 self._writeVm(LAST_SHUTDOWN_REASON, 'crash')
-
+                self.info['crash_time'] = time.time()
                 restart_reason = 'crash'
                 self._stateSet(DOM_STATE_HALTED)

@@ -2188,7 +2188,12 @@
         old_domid = self.domid
         self._writeVm(RESTART_IN_PROGRESS, 'True')

-        elapse = time.time() - self.info['start_time']
+        if xoptions.get_enable_dump() or self.get_on_crash() \
+               in ['coredump_and_destroy', 'coredump_and_restart']:
+            elapse = self.info['crash_time'] - self.info['start_time']
+        else:
+            elapse = time.time() - self.info['start_time']
+
         if elapse < MINIMUM_RESTART_TIME:
             log.error('VM %s restarting too fast (Elapsed time: %f seconds). '
                       'Refusing to restart to avoid loops.',

I have test the situation with the patch, and it works well when the guest 
crashes at boot time.

Best Regards.

Yufang


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel