huge increase in storage activity afther dovecot upgrade

[Adrian M]

Yes, more than one server.

On Thu, Nov 15, 2018 at 11:42 PM Adi Pircalabu <adi at ddns.com.au> wrote:

> On 2018-11-16 07:24, Adrian Minta wrote:
> > Yes, multiple imap servers using one shared nfs storage. With the same
> > config on 2.2.13 the public interface traffic was similar to the
> > storage interface, around 100 mbps.
> >
> > After we switch to 2.2.27 the storage interface traffic jumped 10
> > times while the public interface stayed the same. This make us
> > thinking that something is wrong and each time a user logs in the
> > whole Inbox content is read by dovecot.
> >
> > What you are suggesting goes against the documentations and it may not
> > be save, but I will give a  thought.
>
> I was expecting you have multiple IMAP servers using the same shared NFS
> storage, however my question was: are the *individual mailboxes* on that
> share accessed *at the same time* from more than one IMAP server?
>
> --
> Adi Pircalabu
>
> > On 11/15/18 6:23 AM, Adi Pircalabu wrote:
> >> Are you connecting to the same mailbox over NFS from multiple IMAP
> >> servers? If not and, at any given time, any mailbox will be accessed
> >> from a single NFS client, try to "dupe" Dovecot into thinking it's not
> >> using NFS. We're running quite successfully such setup with NFSv3 over
> >> TCP, which turned out to be the fastest and most reliable throughout
> >> the years. Here are the mount options:
> >>
> rw,noatime,nodiratime,vers=3,rsize=1048576,wsize=1048576,namlen=255,hard,nordirplus,proto=tcp,timeo=600,retrans=2,sec=sys,mountvers=3,mountport=1892,mountproto=tcp,local_lock=none
>
> >> On the Dovecot side we're running with:
> >> lock_method = dotlock
> >> mail_fsync = never
> >> mail_nfs_index = no
> >> mail_nfs_storage = no
> >> maildir_very_dirty_syncs = yes
> >> mmap_disable = yes
> >> protocol lda {
> >>   mail_fsync = optimized
> >> }
> >> protocol lmtp {
> >>   mail_fsync = optimized
> >> }
> >>
> >> Note: we're using Maildir and the usual "works for me(c), may not work
> >> for everyone" applies.
> >>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://dovecot.org/pipermail/dovecot/attachments/20181116/8722f197/attachment.html>