On 07/03/14 23:33, Noel Jones wrote:
On 3/7/2014 11:18 PM, Rodolfo González González wrote:
On 07/03/14 22:53, Noel Jones wrote:
On 3/7/2014 8:22 PM, Rodolfo González González wrote:
Hello,
a customer is reporting that some of their users tried to send
e-mail messages using my postfix 2.9.6 server (will be
updating soon) but messages were not sent. I don't have
snapshots of their screens nor error messages displayed by
their clients yet, but checking the logs I find several
messages like this one:
Mar 4 12:40:55 email postfix/smtpd[30013]: ***********:
client=unknown[a.b.c.d], sasl_method=LOGIN,
sasl_username=john.smith Mar 4 12:40:56 email
postfix/cleanup[30115]: ***********:
message-id=<*********.***%john.sm...@example.com> Mar 4
12:40:56 email postfix/qmgr[21597]: ***********:
from=<john.sm...@example.com>, size=2509, nrcpt=1 (queue
active) Mar 4 12:59:57 email postfix/smtp[17624]: warning:
open active ***********: No such file or directory
It looks like the queued message is being lost, or it's never
written to disk, but I have no clue. Any hint is appreciated.
Thank you.
Your filesystem is broken. Shut down the computer and run fsck
repeatedly until it stops finding errors.
Thanks. The filesystem where the queue is, right? One detail I
forgot to mention, this is an ext4 filesystem on a RAID-10 sitting
on EBS volumes at Amazon Web Services. The instance is
EBS-optimized, and I have IOPS-enabled volumes. There are no other
messages in dmesg nor syslog regarding this problem. Could it be
related to EBS?
I have no idea if that's related or not.
I do know that the filesystem reports to postfix that a queue file
has been written and then the file isn't there. If it were a local
disk, that would be a clear sign of filesystem corruption.
I'm not familiar enough with Amazon's services to tell if that's the
problem. Maybe the options you've chosen cause the filesystem to
report operations are completed before they really are. Postfix
cannot use a filesystem that lies.
Thanks Noel, it makes sense. It could be possible that the file is not
being written because of some bottle neck at EBS level, which is not an
unusual sitiation. I'll contact Amazon's techs. Now, this is a general
question: which would be the best FS and settings to try to avoid this
happening? I mean, to make sure that the file really is written to "disk".
