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

[Praveen Ponakanti]

Hi Alex,

BTW, where do I send the files with the memory logs? I had sent a few
attachments in an email reply to this thread but got back a
message saying it is too large and will need to be reviewed by an admin.
Thanks
Praveen

On Wed, Dec 22, 2021 at 10:58 PM Praveen Ponakanti <pponakanti at roblox.com>
wrote:

> Hi Alex,
>
> Thanks, please see inline below.
>
> On Wed, Dec 22, 2021 at 6:45 AM Alex Rousskov <
> rousskov at measurement-factory.com> wrote:
>
>> On 12/21/21 7:48 PM, Praveen Ponakanti wrote:
>>
>> > We are running the squid proxy for servicing outbound HTTP quests from
>> > our network and have observed a significant memory leak with 5.x
>> > versions. While there are several discussions about memory leaks with
>> > recent versions, just wanted to list out what we have observed in case
>> > this is an unknown leak.
>>
>> I recommend sharing a log with 48+ hourly mgr:mem snapshots. These
>> snapshots help compare your leak with others we know about and may help
>> isolate some of the memory leaks:
>> https://bugs.squid-cache.org/show_bug.cgi?id=5132#c8
>>
>>
> Attached the current memory stats from the 2 nodes that have been taking a
> similar rate of requests, with the node running 5.3 already using up 20G
> more than the 4.17 version. The memory leak increases over a few hours in
> almost a step-like curve each day on the 5.x versions; I will continue
> collecting these stats hourly.
>
> Note, I am using a curl to http://localhost:3128/squid-internal-mgr/mem
> to collect the memory stats as we dont have the squidclient built into the
> docker container that runs the squid proxy. The output appears to be quite
> large with data for each kid process (we have 30 workers configured).
>
>
>> > I have attempted to build squid with -with-valgrind-debug, and run it in
>> > a test env. However valgrind appears to collect some data from the parse
>> > config functions and then the squid proxy restarts.
>>
>> FWIW, valgrind works as expected in my environment. If your Squid proxy
>> is killed by an assertion or crashes, even in a test environment, please
>> consider reporting that bug (after checking that it has not been
>> reported already, of course). If your test proxy is misconfigured, then
>> please fix the configuration before proceeding with tests.
>>
>> The Valgrind image isnt crashing, and it uses the same config that works
> in prod without valgrind. I might have to fix how it is launched from the
> docker entrypoint script in my test env.
>
> root     117051 117005  0 Dec11 ?        00:00:00 /usr/bin/valgrind.bin
> --leak-check=full --show-leak-kinds=all --verbose
> --log-file=/squid/var/logs/valgrind-out.txt /squid/sbin/squid -f
> /squid/etc/squid.conf
>
> gdb attach 1
> GNU gdb (Ubuntu 9.2-0ubuntu1~20.04) 9.2
>
> (gdb) monitor
> "monitor" command not supported by this target.
> (gdb)
>
>
>> Thank you,
>>
>> Alex.
>> _______________________________________________
>> 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/20211222/3a007051/attachment.htm>