daemon/install: file-descriptors leakage

[lucab]

Host system details

# rpm-ostree status
State: idle
AutomaticUpdatesDriver: Zincati
  DriverState: active; periodically polling for updates (last checked Thu 2022-08-25 08:29:02 UTC)
Deployments:
● fedora:fedora/s390x/coreos/testing
                  Version: 36.20220806.2.0 (2022-08-08T16:10:51Z)
               BaseCommit: 2b52abcb312724e21b49f31d5270e3109aae4e9f79dbec00d9f3ce105bb19d21
             GPGSignature: Valid signature by 53DED2CB922D8B8D9E63FD18999F7CBF38AB71F4

Expected vs actual behavior

After a few rounds of looping rpm-ostree install foo && rpm-ostree cleanup -p, I'm seeing quite a few FDs that apparently leaked:

# ls -la /proc/<PID>/fd/
[...]
rwx------. 1 root root 64 Aug 25 08:22 361 -> 'socket:[136876]'
lrwx------. 1 root root 64 Aug 25 08:22 362 -> /sysroot/ostree/repo/.lock
lr-x------. 1 root root 64 Aug 25 08:22 363 -> /sysroot/ostree/repo
lr-x------. 1 root root 64 Aug 25 08:22 364 -> /sysroot/ostree/repo/objects
lr-x------. 1 root root 64 Aug 25 08:22 365 -> /sysroot/ostree/repo/tmp
lrwx------. 1 root root 64 Aug 25 08:22 366 -> /sysroot/ostree/repo/.lock
lr-x------. 1 root root 64 Aug 25 08:22 367 -> /sysroot/ostree/repo/tmp/cache
lrwx------. 1 root root 64 Aug 25 08:22 368 -> /sysroot/ostree/repo/.lock
lrwx------. 1 root root 64 Aug 25 08:22 369 -> 'socket:[138358]'
lr-x------. 1 root root 64 Aug 25 08:22 370 -> /sysroot/ostree/repo
lr-x------. 1 root root 64 Aug 25 08:22 371 -> /sysroot/ostree/repo/objects
lr-x------. 1 root root 64 Aug 25 08:22 372 -> /sysroot/ostree/repo/tmp
lr-x------. 1 root root 64 Aug 25 08:22 373 -> /sysroot/ostree/repo/tmp/cache
lrwx------. 1 root root 64 Aug 25 08:22 374 -> 'socket:[138375]'
lr-x------. 1 root root 64 Aug 25 08:22 375 -> /sysroot/ostree/repo
lr-x------. 1 root root 64 Aug 25 08:22 376 -> /sysroot/ostree/repo/objects
lr-x------. 1 root root 64 Aug 25 08:22 377 -> /sysroot/ostree/repo/tmp
lr-x------. 1 root root 64 Aug 25 08:22 378 -> /sysroot/ostree/repo/tmp/cache
lrwx------. 1 root root 64 Aug 25 08:22 380 -> /sysroot/ostree/repo/.lock
lrwx------. 1 root root 64 Aug 25 08:22 384 -> /sysroot/ostree/repo/.lock
lrwx------. 1 root root 64 Aug 25 08:22 386 -> /sysroot/ostree/repo/.lock
[...]

Not a big deal as eventually the daemon goes to idle and stops (or it can be stopped/restarted manually).
But this FDs leakage makes reproducing non-deterministic bugs a bit more cumbersome; the first time I tried it in a tight loop I encountered a weird forking error, which turned out to be a FD exhaustion.

[jlebon]

This might've been fixed by ostreedev/ostree#2692.

[lucab]

Ah, very likely. Although it seems we are also leaking a socket FD. I'm not looking into this for the moment, but let's do another check when ostree 2022.6 is out.

[cgwalters]

I'd still like eventually to fork off the transaction as a separate subprocess...