[Bug 1890313] Re: GFileMonitor deadlocks

Wed Jan 27 11:52:45 UTC 2021

I renamed the bug to better describe the issue now that it has been root
caused. Feel free to adjust if you disagree.

> Although this is a bug that could arguably affect users too (the fix
is not limited to test code). So it may fix a real hang bug which we
just didn't notice - or get reported - yet. I think it *could* be
released to users.

I agree, and leave that to your discretion. Users affected seem unlikely
to me to be in a position to track it back to this bug, so they won't be
able to report themselves as affected. And the fix is trivial and
limited in scope. So I think it's OK to release to users on the basis
that it's much more likely to be useful to them when compared to the
cost of the update to them and the risk of regressing them.

Feel free to untag block-proposed-{focal,groovy} if you think it's
appropriate.

** Summary changed:

- libxmlb/0.1.15-2 autopkgtest fails or is flaky 
+ GFileMonitor deadlocks

-- 
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to libxmlb in Ubuntu.
Matching subscriptions: foundations-bugs-libxmlb
https://bugs.launchpad.net/bugs/1890313

Title:
  GFileMonitor deadlocks

Status in libxmlb package in Ubuntu:
  Fix Released
Status in libxmlb source package in Focal:
  Fix Committed
Status in libxmlb source package in Groovy:
  Fix Committed

Bug description:
  [ Description ]

  https://autopkgtest.ubuntu.com/packages/libx/libxmlb/groovy/amd64

  https://objectstorage.prodstack4-5.canonical.com/v1/AUTH_77e2ada1e7a84929a74ba3b87153c0ac
  /autopkgtest-
  groovy/groovy/amd64/libx/libxmlb/20200803_212158_ebe4f@/log.gz :

  ...
  (/usr/libexec/installed-tests/libxmlb/xb-self-test:1066): XbSilo-DEBUG: 21:16:46.741: /tmp/temp.xml changed, invalidating
  Executing: libxmlb/libxmlb.test
  Executing: libxmlb/libxmlb.test
  ...
  Executing: libxmlb/libxmlb.test
  Executing: libxmlb/libxmlb.test
  Executing: libxmlb/libxmlb.test
  Executing: libxmlb/libxmlb.test
  Test timed out after 300 seconds
  Executing: libxmlb/libxmlb.test
  FAIL: libxmlb/libxmlb.test (Child process killed by signal 9)
  SUMMARY: total=1; passed=0; skipped=0; failed=1; user=0.0s; system=0.0s; maxrss=5900
  FAIL: libxmlb/libxmlb.test (Child process killed by signal 9)
  ...

  The backtrace while the test is hanging looks like:

  (gdb) bt
  #0  syscall () at ../sysdeps/unix/sysv/linux/x86_64/syscall.S:38
  #1  0x00007fc9150d57cc in g_mutex_lock_slowpath (mutex=mutex at entry=0x7fc9153a0318 <g.inotify_lock_lock>)
      at ../../../glib/gthread-posix.c:1458
  #2  0x00007fc9150d6256 in g_mutex_lock (mutex=mutex at entry=0x7fc9153a0318 <g.inotify_lock_lock>) at ../../../glib/gthread-posix.c:1482
  #3  0x00007fc915309de4 in _ih_sub_cancel (sub=0x55a566f5be10) at ../../../gio/inotify/inotify-helper.c:121
  #4  0x00007fc91530a06e in g_inotify_file_monitor_cancel (monitor=0x55a566f501a0) at ../../../gio/inotify/ginotifyfilemonitor.c:75
  #5  0x00007fc91523680b in g_file_monitor_cancel (monitor=0x55a566f501a0) at ../../../gio/gfilemonitor.c:241
  #6  g_file_monitor_cancel (monitor=0x55a566f501a0) at ../../../gio/gfilemonitor.c:235
  #7  0x00007fc915236890 in g_file_monitor_dispose (object=0x55a566f501a0) at ../../../gio/gfilemonitor.c:123
  #8  0x00007fc9152fe8b3 in g_local_file_monitor_dispose (object=0x55a566f501a0) at ../../../gio/glocalfilemonitor.c:814
  #9  0x00007fc915179c93 in g_object_unref (_object=<optimized out>) at ../../../gobject/gobject.c:3461
  #10 g_object_unref (_object=0x55a566f501a0) at ../../../gobject/gobject.c:3391
  #11 0x000055a5666dce9d in xb_silo_file_monitor_item_free (item=0x55a566f53810) at ../src/xb-silo.c:1220
  #12 0x00007fc915075172 in g_hash_table_remove_all_nodes (hash_table=hash_table at entry=0x55a566f62b60, notify=notify at entry=1, 
      destruction=destruction at entry=1) at ../../../glib/ghash.c:706
  #13 0x00007fc915076533 in g_hash_table_remove_all_nodes (destruction=1, notify=1, hash_table=0x55a566f62b60)
      at ../../../glib/ghash.c:1461
  #14 g_hash_table_unref (hash_table=0x55a566f62b60) at ../../../glib/ghash.c:1461
  #15 0x000055a5666dd445 in xb_silo_finalize (obj=0x55a566f47a00) at ../src/xb-silo.c:1320
  #16 0x00007fc915179d0e in g_object_unref (_object=<optimized out>) at ../../../gobject/gobject.c:3499
  #17 g_object_unref (_object=0x55a566f47a00) at ../../../gobject/gobject.c:3391
  #18 0x000055a5666c4864 in xb_builder_finalize (obj=0x55a566f46da0) at ../src/xb-builder.c:1027
  #19 0x00007fc915179d0e in g_object_unref (_object=<optimized out>) at ../../../gobject/gobject.c:3499
  #20 g_object_unref (_object=0x55a566f46da0) at ../../../gobject/gobject.c:3391
  #21 0x000055a5666db999 in glib_autoptr_clear_GObject (_ptr=0x55a566f46da0) at /usr/include/glib-2.0/gobject/gobject-autocleanups.h:27
  #22 glib_autoptr_clear_XbBuilder (_ptr=0x55a566f46da0) at ../src/xb-builder.h:20
  #23 glib_autoptr_cleanup_XbBuilder (_ptr=<synthetic pointer>) at ../src/xb-builder.h:20
  #24 xb_builder_ensure_watch_source_func () at ../src/xb-self-test.c:501
  #25 0x00007fc9150b13ce in test_case_run (tc=0x55a566f43b00) at ../../../glib/gtestutils.c:2633
  #26 g_test_run_suite_internal (suite=suite at entry=0x55a566f42440, path=path at entry=0x0) at ../../../glib/gtestutils.c:2721
  #27 0x00007fc9150b1174 in g_test_run_suite_internal (suite=suite at entry=0x55a566f42420, path=path at entry=0x0)
      at ../../../glib/gtestutils.c:2733
  #28 0x00007fc9150b18ba in g_test_run_suite (suite=0x55a566f42420) at ../../../glib/gtestutils.c:2808
  #29 0x00007fc9150b18d5 in g_test_run () at ../../../glib/gtestutils.c:2043
  #30 0x000055a5666c38bb in main (argc=<optimized out>, argv=<optimized out>) at ../src/xb-self-test.c:2423
  (gdb) 

  [ Fix ]

  This is a hang when disposing the GFileMonitor.

  See https://gitlab.gnome.org/GNOME/glib/-/issues/1941 for more
  details.

  The fix is to explicitly cancel the GFileMonitor when when unreffing
  it.

  [ QA ]

  We don't have a testcase for the bug in real usage. It's quite easy to
  check in autopkgtest though. Check the autopkgtests stop being
  flaky/racy with the new version.

  [ What could go wrong ]

  We're cancelling the file monitor explicitly now. Previously this was
  happening internally. This is a pattern used by other glib-using
  projects, but if it's bad then it may be that something goes wrong
  with the file monitor.

  [ Other information ]

  I'll stage this following
  https://wiki.ubuntu.com/StableReleaseUpdates#Staging_an_upload.

  Although this is a bug that could arguably affect users too (the fix
  is not limited to test code). So it may fix a real hang bug which we
  just didn't notice - or get reported - yet. I think it *could* be
  released to users.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/libxmlb/+bug/1890313/+subscriptions