[squid-users] Significant memory leak with version 5.x (not with 4.17)

Praveen Ponakanti pponakanti at roblox.com
Fri Jan 7 05:12:43 UTC 2022


Hi Alex,

We are fine with version 4.17 as of now. I can try out the fix sometime
next week if you need further data. Is there a build with the fix, or do
you have some recommended steps to manually pull the source, patch the fix
and then recompile?


On Thu, Jan 6, 2022 at 8:16 PM Alex Rousskov <
rousskov at measurement-factory.com> wrote:

> On 1/6/22 2:50 AM, Praveen Ponakanti wrote:
> > Hi Alex/Amos,
> >
> > Do you still need memory logs from version 5.3 after stopping traffic
> > through the squid?
>
> I cannot answer for Amos who asked for those logs, but you may want to
> try a fix posted at
> https://bugs.squid-cache.org/show_bug.cgi?id=5132#c27
>
>
> HTH,
>
> Alex.
>
>
> > We have disabled traffic to the 5.3 version squid
> > about 6 hours ago and have not seen any memory being freed up since.
> > This node has used up ~50G more memory compared with 4.17 squid taking
> > similar traffic over the last 3+ weeks. I am collecting hourly memory
> > logs on 5.3 after stopping traffic. Let me know and I can attach the
> > log tomorrow morning.
> >
> > Thanks
> > Praveen
> >
> > On Mon, Dec 27, 2021 at 4:58 PM Praveen Ponakanti <pponakanti at roblox.com
> > <mailto:pponakanti at roblox.com>> wrote:
> >
> >     I cant make any changes to our prod squids this week. I have a squid
> >     instance (5.3v) in a test env but could not reproduce the leak by
> >     starting & stopping traffic with a bulk http req generator (wrk).
> >     Was able to send 175k rps @ 20k concurrent sessions (each doing a
> >     get on a 1KB object) through the 30-worker squid. This initially
> >     caused a 3G increase in memory usage and then flattened out after
> >     stopping the requests. If I restart the bulk reqs, the memory usage
> >     only goes up ~0.5GB and then drops back down. Live traffic is
> >     probably exercising a different code path within squid's memory
> pools.
> >
> >     On Mon, Dec 27, 2021 at 2:26 AM Lukáš Loučanský
> >     <loucansky.lukas at kjj.cz <mailto:loucansky.lukas at kjj.cz>> wrote:
> >
> >         After one day of running without clients my squid memory is
> stable
> >
> >         29345 proxy     20   0  171348 122360  14732 S   0.0   0.7
> >         0:25.96 (squid-1) --kid squid-1 -YC -f /etc/squid5/squid.conf
> >         29343 root      20   0  133712  79264   9284 S   0.0   0.5
> >         0:00.00 /usr/sbin/squid -YC -f /etc/squid5/squid.conf
> >
> >         Storage Mem size: 3944 KB Storage Mem capacity: 0.2% used, 99.8%
> >         free Maximum Resident Size: 489440 KB Page faults with physical
> >         i/o: 0 Memory accounted for: Total accounted: 15741 KB
> >         memPoolAlloc calls: 1061495 memPoolFree calls: 1071691 Total
> >         allocated 15741 kB So this does not seem to be the problem... L
> >
> >         Dne 26.12.2021 v 10:02 Lukáš Loučanský napsal(a):
> >>         ok - as it seems my squid quacked on low memory again today -
> >>
> >>         Dec 26 00:04:25 gw (squid-1): FATAL: Too many queued store_id
> >>         requests; see on-persistent-overload.#012    current master
> >>         transaction: master4629331
> >>         Dec 26 00:04:28 gw squid[15485]: Squid Parent: squid-1 process
> >>         15487 exited with status 1
> >>         Dec 26 00:04:28 gw squid[15485]: Squid Parent: (squid-1)
> >>         process 28375 started
> >>
> >>         2021/12/26 00:01:20 kid1| helperOpenServers: Starting 5/64
> >>         'storeid_file_rewrite' processes
> >>         2021/12/26 00:01:20 kid1| ipcCreate: fork: (12) Cannot
> >>         allocate memory
> >>         2021/12/26 00:01:20 kid1| WARNING: Cannot run
> >>         '/lib/squid5/storeid_file_rewrite' process.
> >>         2021/12/26 00:01:20 kid1| ipcCreate: fork: (12) Cannot
> >>         allocate memory
> >>
> >>         I'm going to reroute my clients (which are on their days off
> >>         anyway) to direct connections and run it "dry" - on it's own.
> >>         But I'm not able to to test it before "lack of memory issues
> >>         occur" - because my clients are offline. So I'll watch squid
> >>         for it's own memory consuption. It's all I can do right now -
> >>         my squid already restarted and it's memory has been freed - so
> >>         I think just now I have no power to fill it up again :-]
> >>
> >>         L
> >>
> >>         Dne 26.12.2021 v 7:41 Amos Jeffries napsal(a):
> >>>
> >>>         If possible can one of you run a Squid to get this behaviour,
> >>>         then stop new clients connecting to it before lack of memory
> >>>         issues occur and see if the memory usage disappears or
> >>>         reduces after a 24-48hr wait.
> >>>
> >>>         A series of regular mempools report dumps from across the
> >>>         test may help Alex or whoever works on the bug eliminate
> >>>         further which cache and client related things are releasing
> >>>         properly.
> >>>
> >>>
> >>>         Amos
> >>>
> >>>         _______________________________________________
> >>>         squid-users mailing list
> >>>         squid-users at lists.squid-cache.org
> >>>         <mailto:squid-users at lists.squid-cache.org>
> >>>         http://lists.squid-cache.org/listinfo/squid-users
> >>>         <http://lists.squid-cache.org/listinfo/squid-users>
> >>
> >
> >         <
> https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=emailclient
> >
> >               Bez virů. www.avast.com
> >         <
> https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=emailclient
> >
> >
> >
> >
>  <#m_-6622557068709516458_m_9217020348889694418_DAB4FAD8-2DD7-40BB-A1B8-4E2AA1F9FDF2>
> >         _______________________________________________
> >         squid-users mailing list
> >         squid-users at lists.squid-cache.org
> >         <mailto:squid-users at lists.squid-cache.org>
> >         http://lists.squid-cache.org/listinfo/squid-users
> >         <http://lists.squid-cache.org/listinfo/squid-users>
> >
> >
> > _______________________________________________
> > squid-users mailing list
> > squid-users at lists.squid-cache.org
> > http://lists.squid-cache.org/listinfo/squid-users
> >
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.squid-cache.org/pipermail/squid-users/attachments/20220106/8e149523/attachment.htm>


More information about the squid-users mailing list