--On Monday, March 18, 2013 4:26 PM -0400 Wietse Venema
<wie...@porcupine.org> wrote:
Quanah Gibson-Mount:
--On Sunday, March 17, 2013 8:15 PM -0400 Wietse Venema
<wie...@porcupine.org> wrote:
> Snaphot 20130317 addresses sub-optimal behavior in the LMDB client
> code that affected tlsmgr and "postmap -i", and it makes the code
> more resilient.
>
> In particular, the Postfix LMDB client will no longer keep crashing
> on a "database full" error. Instead, Postfix can now recover without
> immediately requiring human intervention. This is important because
> many Postfix databases contain data that is maintained by a Postfix
> daemon process, and the size of the data is not known in advance.
>
> With this, LMDB no longer requires constant watching for "database
> full" errors. As the system recovers from an error, it logs a warning
> that humans can take care of the next day.
Excellent news, thank you Wietse! Does this mean it may be included in
a future release?
Yes. And with these failure modes eliminated, it becomes worthwhile
to explore new opportunities, such as updating a shared table.
For example, multiple postscreen(8) or tlsmgr(8) or verify(8) daemons
should be able to update a shared LMDB database, something that
cannot safely be done with Berkeley DB. Occasionally some daemon
may restart to automatically recover from an LMDB "table full"
condition, but the system won't come to a halt. You can postpone
these restarts by increasing lmdb_map_size in main.cf before the
table reaches that limit.
These sound like really excellent improvements. I've been using MDB with
OpenLDAP for a while, and we're working hard on removing our reliance on
BDB everywhere, so now that postfix has support for it, it is a major win
in my book. Thank you!
--Quanah
--
Quanah Gibson-Mount
Sr. Member of Technical Staff
Zimbra, Inc
A Division of VMware, Inc.
--------------------
Zimbra :: the leader in open source messaging and collaboration
