[Mailman-Users] Mailman / Sendmail horribly slowperformance..suggestions please?!?
Richard Barrett
r.barrett at openinfo.co.uk
Mon Aug 25 23:05:22 CEST 2003
for closure of this topic:
Begin forwarded message:
> From: "Andrew Falgout" <Andrew.Falgout at usm.edu>
> Date: Mon Aug 25, 2003 8:22:05 pm Europe/London
> To: "Richard Barrett" <r.barrett at openinfo.co.uk>
> Subject: Re: [Mailman-Users] Mailman / Sendmail horribly
> slowperformance..suggestions please?!?
> Reply-To: "Andrew Falgout" <Andrew.Falgout at usm.edu>
>
> Problem solved. It was archiving. That's all I needed was to figure
> out
> what it was pausing on. I turned off archival and informed my manager
> that
> the overhead associated with it would be too much for the system to
> bear.
> It is FLYING through the queued mails now and delivering them now.
>
> Been a lifesaver, and THANKS!!!
>
> ----- Original Message -----
> From: "Richard Barrett" <r.barrett at openinfo.co.uk>
> To: "Andrew Falgout" <Andrew.Falgout at usm.edu>
> Sent: Monday, August 25, 2003 12:11 PM
> Subject: Re: [Mailman-Users] Mailman / Sendmail horribly
> slowperformance..suggestions please?!?
>
>
>>
>> On Monday, August 25, 2003, at 04:22 pm, Andrew Falgout wrote:
>>
>>> I have not modified the default values in the SMTPDirect module or
>>> Defaults.py. If I could get qrunner to tell me anything about what
>>> message(s) it is working I could tell you how many recipients were on
>>> that
>>> list. Each list in the current system should be no more than 2-5
>>> people.
>>> We moved over a small subset of mailing lists from the older
>>> majordomo
>>> server. However, there are some lists with several hundres
>>> recipients
>>> on it
>>> that will be moving in the future when these issues get worked out.
>>>
>>> Orphaned locks?
>>> (ocean) root $ ls -lt
>>> (~mailman/locks)
>>> total 16
>>> -rw-rw-r-- 2 mailman mailman 55 Aug 25 2003 qrunner.lock
>>> -rw-rw-r-- 2 mailman mailman 55 Aug 25 2003
>>> qrunner.lock.ocean.otr.usm.edu.16302
>>> -rw-rw-r-- 2 mailman mailman 62 Aug 25 2003
>>> gifted.studies.lock
>>> -rw-rw-r-- 2 mailman mailman 62 Aug 25 2003
>>> gifted.studies.lock.ocean.otr.usm.edu.16302
>>>
>>> smtp log
>>> Aug 25 10:16:47 2003 (16302) smtp for 1 recips, completed in 1.400
>>> seconds
>>> Aug 25 10:16:47 2003 (16302) smtp for 1 recips, completed in 3.799
>>> seconds
>>> Aug 25 10:16:49 2003 (16302) smtp for 1 recips, completed in 1.968
>>> seconds
>>> Aug 25 10:16:49 2003 (16302) smtp for 1 recips, completed in 1.990
>>> seconds
>>>
>>> I'm guessing this would mean that there are not any orphaned locks.
>>> When I
>>> restart qrunner by hand, I kill the proces and remove all the locks
>>> from the
>>> locks directory.
>>>
>>> I thought the 3.799 seconds in the smtp log was qrunner telling you
>>> how long
>>> it took to send the mail through the MTA.
>>>
>> Mailman's qrunner process each message through a pipeline.
>>
>> Attached is a modified version of the file
>> $prefix/Mailman/Handlers/HandlerAPI.py
>>
>> This logs extra information about the progress of each message through
>> the pipeline into the smtp log. This may give some insight as to where
>> the qrunner is spending its time.
>>
>> The changes in the revised file are as follows:
>>
>> /Users/richard/development/mailman-htdig/mailman-2.0.13/Mailman/
>> Handlers
>> bash-2.05a$ diff -u HandlerAPI.py.orig HandlerAPI.py
>> --- HandlerAPI.py.orig Mon Aug 25 17:58:17 2003
>> +++ HandlerAPI.py Mon Aug 25 18:06:01 2003
>> @@ -73,10 +73,12 @@
>>
>>
>>
>> def do_pipeline(mlist, msg, msgdata, pipeline):
>> + t0x = t1x = t2x = time.time()
>> while pipeline:
>> modname = pipeline.pop(0)
>> mod = __import__('Mailman.Handlers.' + modname)
>> func = getattr(getattr(getattr(mod, 'Handlers'), modname),
>> 'process')
>> + t1x = time.time()
>> try:
>> pid = os.getpid()
>> func(mlist, msg, msgdata)
>> @@ -138,6 +140,10 @@
>> traceback.print_exc(file=s)
>> syslog('error', s.getvalue())
>> break
>> + t2x = time.time()
>> + syslog('smtp', 'do_pipeline: list %s, modname %s, startat
>> %.3f, durn %.3f' %
>> + (mlist.internal_name(), modname, (t1x-t0x),
>> (t2x-t1x)))
>> + t1x = t2x
>> return pipeline
>>
>>
>> The revised files is:
>>
>>
>
>
> -----------------------------------------------------------------------
> -----
> ----
>
>
>>
>>
>> About the best I can do to help.
>>> ----- Original Message -----
>>> From: "Richard Barrett" <r.barrett at openinfo.co.uk>
>>> To: "Andrew Falgout" <Andrew.Falgout at usm.edu>
>>> Cc: <mailman-users at python.org>
>>> Sent: Monday, August 25, 2003 9:52 AM
>>> Subject: Re: [Mailman-Users] Mailman / Sendmail horribly
>>> slowperformance..suggestions please?!?
>>>
>>>
>>>>
>>>> On Monday, August 25, 2003, at 02:28 pm, Andrew Falgout wrote:
>>>>
>>>>> The error log is at 0 bytes. I have not overridden the Defaults.py
>>>>> which
>>>>> shows Defaults.py:MTA_ALIASES_STYLE = 'sendmail'. So I assume that
>>>>> means
>>>>> that sendmail is the MTA.
>>>> You will be using SMTP to the local MTA (which you say is Sendmail)
>>>> if
>>>> the following MM config variables are set as follows in either
>>>> mm_cfg.py or the Defaults.py file. The following are th standard MM
>>>> 2.0.13 default values:
>>>>
>>>> DELIVERY_MODULE = 'SMTPDirect'
>>>> SMTPHOST = 'localhost'
>>>>
>>>> Also are the following default values still in place:
>>>>
>>>> SMTP_MAX_RCPTS = 500
>>>> MAX_DELIVERY_THREADS = 0
>>>>
>>>>>
>>>>> Aug 25 08:08:31 2003 (27798) smtp for 2 recips, completed in 1.045
>>>>> seconds
>>>>> Aug 25 08:08:31 2003 (27798) smtp for 2 recips, completed in 1.149
>>>>> seconds
>>>>> Aug 25 08:12:47 2003 (27798) smtp for 1 recips, completed in 0.208
>>>>> seconds
>>>>> Aug 25 08:12:47 2003 (27798) smtp for 1 recips, completed in 0.241
>>>>> seconds
>>>>> Aug 25 08:14:25 2003 (31184) smtp for 1 recips, completed in 0.701
>>>>> seconds
>>>>> Aug 25 08:14:25 2003 (31184) smtp for 1 recips, completed in 0.716
>>>>> seconds
>>>>> Aug 25 08:17:21 2003 (31184) smtp for 1 recips, completed in 0.167
>>>>> seconds
>>>>> Aug 25 08:17:21 2003 (31184) smtp for 1 recips, completed in 0.179
>>>>> seconds
>>>>> Aug 25 08:19:34 2003 (31184) smtp for 1 recips, completed in 3.962
>>>>> seconds
>>>>> Aug 25 08:19:34 2003 (31184) smtp for 1 recips, completed in 4.234
>>>>> seconds
>>>>>
>>>>> This is just a few of the lines in my /var/log/mailman/smtp file.
>>>>> There are
>>>>> a few bounces from time to time, but not on great frequency. I did
>>>>> send an
>>>>> email to a test list, and attempted to time when I would see my
>>>>> email
>>>>> address in the post log, but after 10 minutes I have not seen it.
>>>>> This
>>>>> really confuses me, because the mail system on this server delivers
>>>>> mail
>>>>> directly to a mailbox in seconds. Why the delay?
>>>>>
>>>> I am going to assume that you had not made the edit to
>>>> $prefix/Mailman/Handlers/SMTPDirect.py I suggested in one of my
>>>> response and that these log entries are the regular ones?
>>>>
>>>> Bear in mind that the number in brackets in these log entries -
>>>> (27798)
>>>> for instance - is the process id of the qurnner handling the message
>>>> and the same qrunner is handling multiple messages.
>>>>
>>>> The smtp log entries are interesting if you have a large backlog of
>>>> messages in the qfiles directory because:
>>>>
>>>> 1. only a few recipients are being delivered for each log entry. Are
>>>> your mailing list that sparely populated for subscribers? Do these
>>>> smtp
>>>> log entries tie in with post entries for lists with few members and
>>>> /var/log/mail entries for outgoing mail through the MTA?
>>>>
>>>> 2. quite long periods - 2, 3, 4 minutes in the sample log entries
>>>> you
>>>> give - elapse between log entries. But the log entries are for the
>>>> same
>>>> process so it has been continuing to run during these pregnant
>>>> pauses.
>>>> But what is it doing?
>>>>
>>>> Some thoughts. Mailman's qrunner can get slow if it is:
>>>>
>>>> a. archiving and
>>>>
>>>> b. the messages for a list are large messages and/or
>>>>
>>>> c. the number of messages in the current archive of a list is
>>>> large.
>>>>
>>>> On the other hand, the pregnant pauses could be because MM is
>>>> encountering long delays when opening SMTP connections to the MTA.
>>>>
>>>> What sort of shape are the traffic/lists that MM is handling.
>>>>
>>>> If you can catch the qrunner at it then maybe strace will give you
>>>> some
>>>> idea of what is doing: waiting on a socket open to the MTA maybe.
>>>>
>>>> Another possibility is if by some means a bunch of lists have got
>>>> orphaned locks over them so that the qrunner is spinning its wheels
>>>> trying to find a list it can operate on. Does anything indicative of
>>>> this show up in the $prefix/locks directory.
>>>>
>>>>> ----- Original Message -----
>>>>> From: "Richard Barrett" <r.barrett at ftel.co.uk>
>>>>> To: "Andrew Falgout" <Andrew.Falgout at usm.edu>
>>>>> Cc: <mailman-users at python.org>
>>>>> Sent: Sunday, August 24, 2003 3:36 PM
>>>>> Subject: Re: [Mailman-Users] Mailman / Sendmail horribly slow
>>>>> performance..suggestions please?!?
>>>>>
>>>>>
>>>>>>
>>>>>> On Sunday, August 24, 2003, at 12:51 am, Andrew Falgout wrote:
>>>>>>
>>>>>>> I'm currently running mailman-2.0.13-1 eith
>>>>>>> sendmail-8.11.6-25.72.
>>>>>>> Posts
>>>>>>> are processed very fast, but delivery is more than slow. It
>>>>>>> takes
>>>>>>> 1-2
>>>>>>> minutes per post to delivery mail. Evaluating what is going on
>>>>>>> in
>>>>>>> the
>>>>>>> queue is a little more than cumbersome. Over the course of 3-4
>>>>>>> days
>>>>>>> mails
>>>>>>> have been piling up in the qfiles directory to the tune of 4000
>>>>>>> mails
>>>>>>> and
>>>>>>> waiting. There are mails 3 days old waiting to go out. I'm
>>>>>>> looking
>>>>>>> for
>>>>>>> ways to evaluate what is in the queue, why it's in the queue, and
>>>>>>> how
>>>>>>> to
>>>>>>> tell qrunner to report more information on what it's thinking.
>>>>>>> Like
>>>>>>> perhaps the msgid of the email it's trying to process at this
>>>>>>> time.
>>>>>>>
>>>>>> Are you using the local sendmail instance as the outgoing SMTP
>>>>>> server?
>>>>>>
>>>>>> Regardless, what do the Mailman post, smtp and smtp-failure logs
>>>>>> show
>>>>>> about mail being handled by the system. How does this tie in with
>>>>>> the
>>>>>> MTA's mail log.
>>>>>>
>>>>>> If your outgoing mail is being handled by the local sendmail MTA
>>>>>> then
>>>>>> I
>>>>>> would normally expect Mailman to handoff to the MTA without much
>>>>>> delay.
>>>>>> If MM is having problems handing off to sendmail then you might
>>>>>> find
>>>>>> some evidence of this in the smtp-failure log.
>>>>>>> -- Thank You,
>>>>>>> Andrew
>>>>>>>
>>>>>>> -+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-
>>>>>>> +--
>>>>>>> +-
>>>>>>> +-
>>>>>>> +-+-+-
>>>>>>> - Andrew Wilson Falgout (RHCE) Linux, E
>>>>>>> Pluribus
>>>>>>> UNIX! -
>>>>>>> - Systems Admin II, TIU Office: (601)
>>>>>>> 266-5603
>>>>>>> -
>>>>>>> - University of Southern Mississippi
>>>>>>> -
>>>>>>> - Fax: (601)
>>>>>>> 266-6788
>>>>>>> -
>>>>>>> -+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-
>>>>>>> +-+-
>>>>>>> +-+-+-+
>>>>>>>
-----------------------------------------------------------------------
Richard Barrett http://www.openinfo.co.uk
More information about the Mailman-Users
mailing list