Re: intermittent long "log file sync" waits

From: Stefan Koehler <contact_at_soocs.de>
Date: Tue, 28 Jan 2020 22:14:30 +0100 (CET)
Message-ID: <1914992928.116548.1580246070994_at_ox.hosteurope.de>


Hello Chris,
here we go with the full profile: http://www.soocs.de/public/tmp/lsst2db2_ora_8486_tvd.html

Latency bucket histogram of "log file sync" looks not good ;-)

No need to enable 10046 for LGWR if this is a LGWR problem (e.g. AIO syscalls take so long). You should see entries (in LGWR or its LG<nn> slaves .trc depending on which mode is used) like "Warning: log write elapsed time <X> ms, size <X> KB". However as Nenad already mentioned these lines do not necessarily indicate an IO problem but you can figure this out with strace or Tanel's psnapper when you hit this problem again :-)

P.S.: There are also cases where sporadic "log file sync" issues are related to the LGWR post/wait or polling implementation but you should see this mode switches in LGWR trace as well.

Best Regards
Stefan Koehler

Independent Oracle performance consultant and researcher Website: http://www.soocs.de
Twitter: _at_OracleSK

> Chris Stephens <cstephens16_at_gmail.com> hat am 28. Januar 2020 um 22:02 geschrieben:
>
> I figured it would. :) This application is going to be the death of me...but I guess I'll be learning all the way to the grave. ha.
>  
> maybe i should enable 10046 for lgwr? luckily this isn't production. I'm not too worried about unintended consequences of tracing lgwr.

--
http://www.freelists.org/webpage/oracle-l
Received on Tue Jan 28 2020 - 22:14:30 CET

Original text of this message