OpenVZ Forum


Home » General » Support » undump fails because of .nfs file in Container (/var/run/apache2/.nfs000000... prevents migration)
undump fails because of .nfs file in Container [message #42451] Tue, 19 April 2011 09:01 Go to next message
astnwt is currently offline  astnwt
Messages: 5
Registered: April 2011
Junior Member
Hey there,

we are not able to zero-downtime-migrate a CT,
the target HN brings the following error description:

root@ovz-b:~# vzctl restore 2000
Restoring container ...
Starting container ...
Container is mounted
	undump...
Setting CPU units: 1000
Configure veth devices: veth2000.0 
Error: undump failed: No such file or directory
Restoring failed:
Error: can't open file /var/run/apache2/.nfs00000000005430960000004e
Error: rst_file: -2 46240
Error: rst_files: -2
Error: make_baby: -2
Error: rst_clone_children
Container start failed
Stopping container ...
Container was stopped
Container is unmounted


The file is created everytime Apache2 starts on the CT and is removed as soon as Apache2 stops.

Also, we can't just "delete" the file because Apache2 keeps filesystem locks on it:

root@ct2000:/# lsof | grep .nfs
apache2   456     root    9w      REG               0,27        0    5517462 /var/run/apache2/.nfs000000000054309600000073
apache2   472 www-data    9w      REG               0,27        0    5517462 /var/run/apache2/.nfs000000000054309600000073
apache2   473 www-data    9w      REG               0,27        0    5517462 /var/run/apache2/.nfs000000000054309600000073
apache2   474 www-data    9w      REG               0,27        0    5517462 /var/run/apache2/.nfs000000000054309600000073
apache2   475 www-data    9w      REG               0,27        0    5517462 /var/run/apache2/.nfs000000000054309600000073
apache2   476 www-data    9w      REG               0,27        0    5517462 /var/run/apache2/.nfs000000000054309600000073



It seems that this file prevents us from using zero-downtime migration completely, is there any way to debug this?
Re: undump fails because of .nfs file in Container [message #42491 is a reply to message #42451] Thu, 21 April 2011 20:29 Go to previous messageGo to next message
astnwt is currently offline  astnwt
Messages: 5
Registered: April 2011
Junior Member
The problem still persists, but now we have a little more information
about this file:

It seems to be a kind of lock file since it is empty, in this case created by SQLITE. As soon as we power down the process that's using the database, the file vanished.

Such files also show up on /tmp on containers that are hosting MySQL servers. Obviously, these services recognize that the containers on our host nodes are actually a NFS share mounted to /var/lib/vz.

We use NFS shares for achieve fast migrations of the containers between different host nodes. When powering those containers from our local HDDs, these .nfs files are not created and the container migration succeeds.

We believe this might be a bug on openVZs restore mechanism since it seems to handle such lock files wrong - maybe because they are locked by an active process. Therefore, I will create a new bug report to Bugzilla and post the link to the report afterwards.


Thank you in advance for your help Smile

Daniel
Re: undump fails because of .nfs file in Container [message #42492 is a reply to message #42451] Thu, 21 April 2011 21:02 Go to previous message
astnwt is currently offline  astnwt
Messages: 5
Registered: April 2011
Junior Member
can not post a Link yet (10 posts limitation) -
so the BugID in BugZilla we reported is:

1857
Previous Topic: checkpoint-restore fails on undump
Next Topic: vzlist sort
Goto Forum:
  


Current Time: Mon Aug 18 17:37:54 GMT 2025

Total time taken to generate the page: 0.34687 seconds