Possible memory leak in [1.2]

[do-not-set-2fa]

What version of Janus is this happening on?
master (from 2024-07-18)
const char *janus_version_string = "1.2.4";

Have you tested a more recent version of Janus too?
Yes

Was this working before?
Yes

Is there a gdb or libasan trace of the issue?
libasan report

Additional context
Simple Peer connection via lua plugin, cleanup of all sessions and shutdown of janus

[lminiero]

I see references to incoming requests, which makes me think there were still active connections to Janus when you shut it down.

[do-not-set-2fa]

Ok i will check it again, but first vacation;)

[do-not-set-2fa]

I repeated the same steps, and killed all sessions (made sure that session_list admin request return empty list) and got same leaks from incoming request, can it be that session recovery timer keeps it in memory for that amount of time till it is fully cleared?

[atoppi]

Just test by terminating after the session reclaim timeout. E.g. if it's 30 seconds wait 60 seconds since the last destroy, then kill the server.

[do-not-set-2fa]

I have 120sec reclame, and waited +3 min but same libasan output

[lminiero]

You can try uncommenting the REFCOUNT_DEBUG define in refcount.h and recompile, which should show more info on the references Janus still has at shutdown.

[atoppi]

I suspect that janus_lua_session_free is never being called because the lua session is the only thing getting malloc'ed while creating the plugin session.

[atoppi]

@do-not-set-2fa are you able to repro with one of the stock lua plugins?

[atoppi]

I was trying to repro with echotest.lua and bumped into a memory leak that seems different

==52251==ERROR: LeakSanitizer: detected memory leaks

Direct leak of 640 byte(s) in 8 object(s) allocated from:
    #0 0x58f97aeac743 in malloc (/usr/local/janus/bin/janus+0x27b743) (BuildId: 1a85cdaa67c38c85f1006f74bd2cbbe7db8a5234)
    #1 0x7c5ae62e8af9 in g_malloc /usr/src/glib2.0-2.80.0-6ubuntu3.1/debian/build/deb/../../../glib/gmem.c:100:13
    #2 0x7c5ae63416f1 in g_system_thread_new /usr/src/glib2.0-2.80.0-6ubuntu3.1/debian/build/deb/../../../glib/gthread-posix.c:1265:12
    #3 0x7c5ae275c1fb in janus_lua_method_pushevent /home/atoppi/src/janus-gateway/src/plugins/janus_lua.c:554:3
    #4 0x7c5ae1c56f7d in luaD_precall /build/lua5.3-LmUA4D/lua5.3-5.3.6/src/ldo.c:434:12
    #5 0x7c5ae1c617f8 in luaV_execute /build/lua5.3-LmUA4D/lua5.3-5.3.6/src/lvm.c:1134:13
    #6 0x7c5ae1c4f888 in luaD_rawrunprotected /build/lua5.3-LmUA4D/lua5.3-5.3.6/src/ldo.c:142:3
    #7 0x7c5ae1c522e7 in lua_resume /build/lua5.3-LmUA4D/lua5.3-5.3.6/src/ldo.c:664:12
    #8 0x7c5ae1c6e97f in auxresume /build/lua5.3-LmUA4D/lua5.3-5.3.6/src/lcorolib.c:39:12
    #9 0x7c5ae1c6eba9 in luaB_coresume /build/lua5.3-LmUA4D/lua5.3-5.3.6/src/lcorolib.c:60:7
    #10 0x7c5ae1c56f7d in luaD_precall /build/lua5.3-LmUA4D/lua5.3-5.3.6/src/ldo.c:434:12
    #11 0x7c5ae1c617f8 in luaV_execute /build/lua5.3-LmUA4D/lua5.3-5.3.6/src/lvm.c:1134:13
    #12 0x7c5ae1c57237 in luaD_call /build/lua5.3-LmUA4D/lua5.3-5.3.6/src/ldo.c:499:5
    #13 0x7c5ae1c572ab in luaD_callnoyield /build/lua5.3-LmUA4D/lua5.3-5.3.6/src/ldo.c:509:3
    #14 0x7c5ae1c572ab in lua_callk /build/lua5.3-LmUA4D/lua5.3-5.3.6/src/lapi.c:925:5
    #15 0x7c5ae2783c5b in janus_lua_scheduler /home/atoppi/src/janus-gateway/src/plugins/janus_lua.c:2533:4
    #16 0x7c5ae6311c81 in g_thread_proxy /usr/src/glib2.0-2.80.0-6ubuntu3.1/debian/build/deb/../../../glib/gthread.c:835:20
    #17 0x58f97aeaa22c in asan_thread_start(void*) asan_interceptors.cpp.o

This is a leak for sure, since for any echotest session started and closed, the amount of leaked bytes grows. Wondering if the root cause might be similar.

[atoppi]

#3409 should fix the leak mentioned in the message above

[atoppi]

With #3409 I can't reproduce memory leaks anymore with echotest.lua

[do-not-set-2fa]

Great thanks, I will test too in a week or so, after vacation. Regards, Mirko

…

On Wed, 24 Jul 2024, 12:18 Alessandro Toppi, ***@***.***> wrote: With #3409 <#3409> I can't reproduce memory leaks anymore with echotest.lua — Reply to this email directly, view it on GitHub <#3408 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BIALZIKJVPTUHM4EYCTGLA3ZN55OPAVCNFSM6AAAAABLEUTVWSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDENBXGUYTCOBUGQ> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[do-not-set-2fa]

@atoppi what transport plugin fif you use?

Have you tried with http?

[atoppi]

@do-not-set-2fa it's the same with both ws and http. No leak when using this patch and the official echotest.lua.

[do-not-set-2fa]

Ok this would explain the slow leak that i am experiencing over the long period of time (even though i can't seem to get the same libasan report as you do, probably doing something wrong :)

I'll wait for the merge and new tag and then start to deploy it to see the effect over time.

Thanks a lot @atoppi and @lminiero

[atoppi]

@do-not-set-2fa the PR has been merged and will soon be backported to 0.x too.

[do-not-set-2fa]

Great thanks

…

On Wed, 4 Sept 2024, 11:37 Alessandro Toppi, ***@***.***> wrote: @do-not-set-2fa <https://github.com/do-not-set-2fa> the PR has been merged and will soon be backported to 0.x too. — Reply to this email directly, view it on GitHub <#3408 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BIALZIKU27UOXI6SWMQZTZ3ZU3IFJAVCNFSM6AAAAABLEUTVWSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMRYGM4DIMJYGY> . You are receiving this because you were mentioned.Message ID: ***@***.***>