Plesk and vhosts on NFS

[ErlandM]

Hi all,

At our hostingcompany we work with a Netapp filer from which we mount the /srv/www/vhosts dir via NFS.
When deleting a domain with plesk we always get the following error:

ERROR: PleskFatalException

Unable to remove hosting: Unable to delete system user: SysUser->systemRemove() failed: usermng failed: usermng: Unable to unlink directory entry .nfs0072ed3b0000b345: Device or resource busy

System error 16: Device or resource busy

0: /usr/local/psa/admin/plib/class.PhDomain.php:290
PhDomain->reset(integer '0', boolean true, boolean false)
1: /usr/local/psa/admin/plib/class.BsDomain.php:307
BsDomain->reset(integer '0')
2: /usr/local/psa/admin/plib/class.BsDomain.php:301
BsDomain->delete(integer '0')
3: /usr/local/psa/admin/plib/class.BsDomain.php:541
mdeleteDomains(array)
4: /usr/local/psa/admin/plib/class.Manager.php:352
Manager->removeDomains(array)
5: /usr/local/psa/admin/htdocs/domains/removeDomains.php3:44

I have tracked the problem down to the fact that NFS makes "silly names" copies .nfsXXXXXXXXXXXXX of files still in use. Since the access log and error log seem to be in use durig the delete of the vhost tree this creates the problem. The .nfs files are owned root.root.

Since i am currently setting up a server for another customer under plesk 9.2 i'm going to try reproducing the error in that version.

Does anybody have a clue how to resolve this problem?

Tia,
Erland.

 

[atomicturtle]

Yeah thats a locking issue with NFS, Ive run into it before as well. Some services are more NFS friendly than others, qmail and postfix for example you can use NFS volumes with safely. Apache, and Mysql not so much, let me guess that was in /var/www/vhosts right?

 

[ErlandM]

It's indeed a locking problem with NFS, but should be, imho, possible to work around if plesk would delete the domain in an other order.

The files that give problems are /srv/www/vhosts/<domainname>/statistics/logs/access_log and error_log. I think plesk tries to delete the directory at the moment apache and the vhost are still active.

By disabling the vhost first and restarting apache, and then deleting the directory-tree the problem should be resolved.

I would have liked to enter a problem ticket with plesk but unfortunately we bought plesk from a reseller who's standard reply is "we can't do anything about it".

As i have written i'm at the moment setting up another server with plesk 9.2.1. I'll try to reproduce the problem there. Maybe plesk has solved the problem in a new version.

Greets,
Erland.

 

[atomicturtle]

Its not pretty, but what you can do is either build a wrapper around the various utilities like usermng to shut down services or force a sync via nfs. Another option I hadnt tried yet was to try and do the same thing using the event manager.

 

[ErlandM]

I have just tested it in 9.2.1, same setup as with the 8.3 version and the bug doesn't seem to exist there.

Erland.

 

[internezzo]

We have this issue with Plesk 9.5.x and 10.2.0. As a workaround we remove the domain from the zz010_psa_httpd.conf file, reload the apache daemon and delete the domain using the Plesk CLI. This works fine.
We just opened a case with Parallels and hope they fix it soon (really annoying for reseller customers to ask us for domain deletions...).

 

[internezzo]

Parallels found a workaround: Suspend the domain before deletion. For us it works with Plesk 9.5.x and 10.2.
Can anyone else confirm it?