Extra forks on automount initialisation & HUP signal
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
autofs (Debian) |
Fix Released
|
Unknown
|
|||
autofs (Ubuntu) |
Fix Released
|
Medium
|
Dimitri John Ledkov |
Bug Description
Imported from Debian bug http://
Package: autofs
Version: 5.0.6-2
Severity: important
User: <email address hidden>
Usertags: origin-ubuntu quantal ubuntu-patch
After merging 5.0.6-2 for Ubuntu I have noticed that upstart init daemon
was tracking the wrong PID of the daemon. [1]
This resulted me into digging what has changed and I noticed the three
patches, which I have attached to this email.
These patches [0001-0003] do three things:
* Check mount.nfs version to be >= 1.1.1
* Check kernel version is >= 2.6.22
* Do above by calling nfs_version_check at pre-demonisation, hence
upstart ending up tracking `mount.nfs -V` call
* all for the sake of figuring out if probing optimisation can be used
(or something, not too sure):
"""
The change to have the kernel process text based mount options can
introduce lengthy timeout waits when attempting a mount to a host
that is not available.
To avoid these waits autofs should probe singleton mounts if it
thinks mount.nfs will pass text options to the kernel (which of
course implies the kernel supports this).
"""
But, mount.nfs 1.1.1 is acient and so is linux kernel 2.6.22... so I'm
thinking to make this a no-op. and simply remove those checks. See mine
Remove-*.patch attached.
With that patch I want to revert the upstream patches [0001-0003].
Setting priority to important as this is change of daemon behaviour and
resource waste to fork calls like this. Feel free to lower severity, if
you wish.
[1] Upstart does not rely on pidfiles, instead it counts forks/clones
and chooses the correct pid to track, such that it notices if it dies /
gets killed. This is more reliable than pidfiles. Upstart can also run
daemons in foreground mode.
--
Regards,
Dmitrijs.
Changed in autofs (Ubuntu): | |
status: | New → Triaged |
status: | Triaged → Fix Committed |
importance: | Undecided → Medium |
assignee: | nobody → Dmitrijs Ledkovs (dmitrij.ledkov) |
Changed in autofs (Debian): | |
importance: | Undecided → Unknown |
Changed in autofs (Debian): | |
status: | New → Fix Released |
This bug was fixed in the package autofs - 5.0.6-2ubuntu1
---------------
autofs (5.0.6-2ubuntu1) quantal; urgency=low
* Merge from Debian unstable (LP: #1006509), remaining changes: autofs. upstart - upstart job patches/ 16group_ buffer_ size.patch:
- debian/
- debian/
+ prevents package to eat the cpu if you have large groups.
* Dropped changes: with-lpthread. patch libs-correctly. patch
- 17ld.patch, applied in debian as link-daemon-
and save-hesiod-
* This upload renames autofs5 back to autofs, following debian's name
change.
* Added patch descriptions/ headers to 16group_buffer_size
* debian/ patches/ 0001-Remove- kernel- mount.nfs- version- checks- on-Debian- Ubu.patch:
Remove kernel & mount.nfs version checks on Debian/Ubuntu. (LP: #1016673)
-- Dmitrijs Ledkovs <email address hidden> Wed, 20 Jun 2012 12:34:50 +0100