Fred Kilbourn
2012-Dec-05 11:08 UTC
[Dovecot] Error: unlink_directory(.../lucene-indexes) with doveadm index
Hi, Per discussion on this previous thread: http://dovecot.org/pipermail/dovecot/2012-October/068849.html I am updating search indexes on all my mailboxes via a nightly cronjob that runs: doveadm index -A '*' The problem we're running into is that random mailboxes will throw the following error during the update (sanitized to conceal actual domain name): doveadm(user at domain.com): Error: unlink_directory(/var/spool/mail/domain.com/user/sdbox/lucene-indexes) failed: Permission denied We have around 250 domains on the server, with a handful of mailboxes on average in each domain. We get this for a varying number of total mailboxes (1 to 6 every night). It is common for the same mailbox to show up a few days in a row and then stop throwing the error. The script is running as root, so there shouldn't be any file level permissions. Any thoughts on the root cause of these errors? What can I provide to help further diagnose? Thanks, Fred Kilbourn
Timo Sirainen
2012-Dec-18 18:52 UTC
[Dovecot] Error: unlink_directory(.../lucene-indexes) with doveadm index
On Wed, 2012-12-05 at 05:08 -0600, Fred Kilbourn wrote:> I am updating search indexes on all my mailboxes via a nightly cronjob that > runs: doveadm index -A '*' > > The problem we're running into is that random mailboxes will throw the > following error during the update (sanitized to conceal actual domain name): > > doveadm(user at domain.com): Error: > unlink_directory(/var/spool/mail/domain.com/user/sdbox/lucene-indexes) > failed: Permission denied > > We have around 250 domains on the server, with a handful of mailboxes on > average in each domain. We get this for a varying number of total mailboxes > (1 to 6 every night). It is common for the same mailbox to show up a few > days in a row and then stop throwing the error. The script is running as > root, so there shouldn't be any file level permissions.Try if this works: doveadm index -A -q '*' This way the indexing is done by indexer processes with proper permissions. Although I wonder if it floods the indexer processes too quickly with the commands if there are a lot of users..
Maybe Matching Threads
- unlink_directory failed with 'Device or resource busy' on NFS
- ForceCommand internal-sftp causes sftp logging to fail (openssh-5.0p1)
- unlink_directory(....Trash.NewFolder) failed: Directory not empty
- Difference between Indexing and Rescan in FTS
- Dovecot cannot unlink non-empty directory