Thank you very much for your reply, Noel,

Firstly, posting the same message with a slightly modified subject wont get you any answers any sooner, please refrain from doing this, it only gives you a bad name and may have the opposite affect to one you desire.

Yeah, typical rookie move by posting twice...didn't see my post, so went into the dovecot list setting for my account to adjust it to get a confirmation, and then sent the message again. Really stupid, and I apologize.

Generally when lodging requests for help, it pays to use the latest version of software, 1.1.2 is old.

Definitely will be doing this, but just wanted to get some information on the locking issue first.

mbox over NFS ? That is asking for trouble, mbox is *not* NFS safe, you are best changing to maildir which is.

Yeah, this is definitely wrong. Everything is in maildir format on the mailstore. Question is, why would this even be working at all if it's wrong in the config? The sql connection string has maildir defined in it, so, I'm wondering if that might explain it. Also, the indexes are not local which I believe has been the recommendation. Also Lockd appears to be enabled on both sides.

-Rod

Firstly, posting the same message with a slightly modified subject wont get you any answers any sooner, please refrain from doing this, it only gives you a bad name and may have the opposite affect to one you desire.

On Tue, Nov 11, 2008 at 4:13 AM, Noel Butler noel.butler@ausics.net wrote:

Firstly, posting the same message with a slightly modified subject wont get you any answers any sooner, please refrain from doing this, it only gives you a bad name and may have the opposite affect to one you desire.

On Tue, 2008-11-11 at 10:06, Rod Treweek wrote:

...

Hello All,

We are struggling to find a solution to a problem we are encountering

with a load-balanced email setup. Currently, we have a Coyote loadbalancer, and 3 Postfix/Dovecot nodes that then get their information from a mysql database. The problem is that after a couple weeks, we start seeing NFS locking issues occur, which then takes email completely down, requiring a site visit to basically do an interactive mode reboot to shut off NFS so that we can remount the filesystem again. We have our mailstore on an Aberdeen NAS which is accessed over NFS v. 4.

Not seen this before, but we use Foundry LB's and NetApp filers. Coyote Is BSD based isn't it... How is your setup, you are not running the NFS mounts through the LB as well are you? I would hope they are on a separate private LAN

...

Dovecot version 1.1.2

Generally when lodging requests for help, it pays to use the latest version of software, 1.1.2 is old.

...

Here's the relevant mysql string:

If the queries work at all, there's nothing wrong with them.

...

mbox:/NFS1MAILDIR/mailSysV2/%d/%u:INDEX=/usr/local/mail/indexes/%d/%1n/%n

mbox over NFS ? That is asking for trouble, mbox is *not* NFS safe, you are best changing to maildir which is.

Cheers

Thank you very much for your reply, Noel,

Yeah, typical rookie move by posting twice...didn't see my post, so went into the dovecot list setting for my account to adjust it to get a confirmation, and then sent the message again. Really stupid, and I apologize.

Yeah, Sorry, still kind of figuring out list protocol. Thanks for the

responses.

I've not built this system, but I've inherited the responsibility of maintaining it, so I'm still kind of trying to gradually learn about how everything works together.

I think that the main things are:

fcntl; pretty sure that dotlock is what we should be using, however, I was under the impression that lockd would/could deal with the file locking issues, if it were present on both client and server.

It also appears that the sql string is overwriting, or rather replacing information gathered from the config...Am I correct in this observation? Even though mbox is there in the config, it looks like the sql string is overwriting, or replacing elements of the config, which leads me to my next question:

If the mbox directive in the config is basically not being used because of the connnection/query string, then I'm thinking that maybe the INDEX directive is also probably not being used as well...Which would probably screw things up pretty badly...Any thoughts on this?

I'm going to be setting up a test environment here quite soon, so hopefully I'll have some of my own answers to post back to the list, however, if anyone has anything to add in the meantime, I'd very much appreciate it.

-Rod

On Tue, Nov 11, 2008 at 1:38 PM, Eduardo M KALINOWSKI < eduardo@kalinowski.com.br> wrote:

Well, you just posted your message three times...

Rod Treweek wrote:

...

Yeah, typical rookie move by posting twice...didn't see my post, so went into the dovecot list setting for my account to adjust it to get a confirmation, and then sent the message again. Really stupid, and I apologize.

It's a gmail "feature": you do not see your own posts to a list, at least not until you get a reply to them.

-- Bell Labs Unix -- Reach out and grep someone.

Eduardo M KALINOWSKI eduardo@kalinowski.com.br http://move.to/hpkb

Thanks for the response. I also forgot to mention that this is on FreeBSD 6.3. Not sure if that makes a difference.

On Wed, Nov 12, 2008 at 4:07 PM, Seth Mattinen sethm@rollernet.us wrote:

Rod Treweek wrote:

...

Yeah, Sorry, still kind of figuring out list protocol. Thanks for the responses.

I've not built this system, but I've inherited the responsibility of maintaining it, so I'm still kind of trying to gradually learn about how everything works together.

I think that the main things are:

fcntl; pretty sure that dotlock is what we should be using, however, I was under the impression that lockd would/could deal with the file locking issues, if it were present on both client and server.

Maybe, maybe not. I'd try dotlock and see if it helps.

~Seth

Like you said before, the SQL user_query is overriding mail_location
setting completely.

On Nov 13, 2008, at 8:22 PM, Rod Treweek wrote:

I'm also noticing that the INDEX=/usr/local/mail/indexes/ path is
completely empty....shouldn't this get populated if the config is getting read properly?

On Thu, Nov 13, 2008 at 9:24 AM, Rod Treweek rtreweek@gmail.com
wrote:

...

Thanks for the response. I also forgot to mention that this is on
FreeBSD 6.3. Not sure if that makes a difference.

On Wed, Nov 12, 2008 at 4:07 PM, Seth Mattinen sethm@rollernet.us
wrote:

...

Rod Treweek wrote:

...

Yeah, Sorry, still kind of figuring out list protocol. Thanks
for the responses.

I've not built this system, but I've inherited the responsibility
of maintaining it, so I'm still kind of trying to gradually learn
about how everything works together.

I think that the main things are:

fcntl; pretty sure that dotlock is what we should be using,
however, I was under the impression that lockd would/could deal with the file
locking issues, if it were present on both client and server.

Maybe, maybe not. I'd try dotlock and see if it helps.

~Seth

On Nov 19, 2008, at 9:01 PM, Rod Treweek wrote:

password_query = SELECT username as user, password, concat('/NFS1MAILDIR/mailSysV2/', maildir) as userdb_home, concat('maildir:/NFS1MAILDIR/ mailSysV2/',maildir, ':INDEX=/usr/local/dovecot/indexes/',maildir) as userdb_mail, 143 AS userdb_uid, 143 AS userdb_gid, concat('*:storage=',
floor(quota*1024)) AS quota_rule FROM mailbox WHERE username = '%u' AND active = '1'

The userdb_* fields here are used only if you're using prefetch userdb.

user_query = SELECT concat('/NFS1MAILDIR/mailSysV2/', maildir) as
home, concat('maildir:/NFS1MAILDIR/mailSysV2/', maildir) as mail, concat('maildir:/NFS1MAILDIR/mailSysV2/',maildir, ':INDEX=/usr/local/dovecot/indexes/',maildir) as mail_location,

"mail_location" field is ignored. It's called mail, which you're also
returning without the :INDEX.

So I guess you're 1) not using prefetch userdb so the userdb_mail in
password_query doesn't get used and 2) the "mail" field without :INDEX
is used from user_query.

I've got mail working more or less, however the indexes simply will
not appear in the path referenced above. My suspicion has been that this
is where my nfs locking problems are occurring, as the indexes are
just not getting populated. Any ideas? I gather that the cacheing done
through the mysql database is desirable as it would tend to speed things up
quite a bit....any ideas on how to pass the index location through the
string as well? Or is this generally regarded as a negligible gain?

What caching? I'm not really sure what you're asking here.

On Nov 20, 2008, at 4:15 AM, Rod Treweek wrote:

Hi Timo,

Thanks Timo. This is great. I followed your instructions, and sure
enough, everything is working as expected now. Thank you very very much. One question I have, What's the advantage of using prefetch? Is this a considerable performance benefit, or is it just more or less another
way to do things? Thanks again.

With prefetch you're doing a single SQL query. Without prefetch you're
doing two. If your SQL server isn't loaded, I doubt it helps much.