NC21: Memleak when do "occ db:add-missing-indices"

[TheRaven500]

Steps to reproduce

1. sudo -u www-data ./occ db:add-missing-indices

Expected behaviour

Database should be fixed

Actual behaviour

Memory grow up and up until it crashes

Server configuration

Operating system:
Debian 10
Web server:
Abache2
Database:
Mariadb
PHP version:
7.4
Nextcloud version: (see Nextcloud admin page)
21.0.0.18
Updated from an older Nextcloud/ownCloud or fresh install:
Updated from 20.0.0.8

Never had problems before. I think it is a problem with NC21.

[kesselb]

Logs?

[TheRaven500]

What for logs are needed?
I don't think that you see a "memleak" in the "normal" logs. Maybe in a debug log?

[TheRaven500]

The bug is still present in NC21.0.1 RC1.
What logs are needed?
In dmesg i see:

[660693.440580] php-fpm7.4 invoked oom-killer: gfp_mask=0x6200ca(GFP_HIGHUSER_MOVABLE), nodemask=(null), order=0, oom_score_adj=0
[660693.440582] php-fpm7.4 cpuset=/ mems_allowed=0
[660693.440586] CPU: 0 PID: 574 Comm: php-fpm7.4 Not tainted 4.19.0-11-amd64 #1 Debian 4.19.146-1
[660693.440588] Call Trace:
[660693.440595]  dump_stack+0x66/0x90
[660693.440599]  dump_header+0x6b/0x283
[660693.440601]  oom_kill_process.cold.30+0xb/0x1cf
[660693.440603]  ? oom_badness+0x23/0x140
[660693.440604]  out_of_memory+0x1a5/0x450
[660693.440606]  __alloc_pages_slowpath+0xbd8/0xcb0
[660693.440608]  __alloc_pages_nodemask+0x28b/0x2b0
[660693.440610]  filemap_fault+0x333/0x780
[660693.440612]  ? alloc_set_pte+0x49e/0x560
[660693.440614]  ? __schedule+0x2a7/0x840
[660693.440615]  ? filemap_map_pages+0x139/0x3a0
[660693.440640]  ext4_filemap_fault+0x2c/0x40 [ext4]
[660693.440642]  __do_fault+0x36/0x130
[660693.440644]  __handle_mm_fault+0xe6c/0x1270
[660693.440646]  handle_mm_fault+0xd6/0x200
[660693.440648]  __do_page_fault+0x249/0x4f0
[660693.440650]  ? page_fault+0x8/0x30
[660693.440652]  page_fault+0x1e/0x30
[660693.440654] RIP: 0033:0x7fa0190fe7b7
[660693.440657] Code: Bad RIP value.
[660693.440658] RSP: 002b:00007ffc0c066288 EFLAGS: 00010246
[660693.440660] RAX: 0000000000000000 RBX: 0000000000000001 RCX: 00007fa0190fe7b7
[660693.440660] RDX: 0000000000000001 RSI: 000055605966d320 RDI: 0000000000000008
[660693.440661] RBP: 0000000000000064 R08: 000055605966d320 R09: 00000000000a57f5
[660693.440662] R10: 0000000000000064 R11: 0000000000000246 R12: 00007ffc0c0662c0
[660693.440663] R13: 0000556057617a60 R14: 20c49ba5e353f7cf R15: 0000000000000000

And:

[660693.440865] Out of memory: Kill process 13170 (php) score 892 or sacrifice child
[660693.440874] Killed process 13170 (php) total-vm:9204764kB, anon-rss:7308052kB, file-rss:0kB, shmem-rss:0kB
[660693.914196] oom_reaper: reaped process 13170 (php), now anon-rss:0kB, file-rss:0kB, shmem-rss:0kB

[vorlon]

I stumbled upon the same issue after a fresh upgrade to 21.0.1 from 20.x. Then I noticed that even just calling occ without arguments leads to an OOM situation.
This might be the same as described in #26184 . In my case commenting out
'memcache.local' => '\\OC\\Memcache\\APCu' in config.php helped to succesfully run occ again.

[TheRaven500]

Ha! Thx! Has also worked for me. Job completet in few seconds after commenting out the cache option in config.php!

[caugner]

ping @rullzer and @MorrisJobke as top committers in the Memcache namespace. Do you have any guess what might cause memory leaks when running occ with enabled APCu cache?

[kesselb]

Closing in favour of #26184 #25742