Thanks to the careful and detailed analysis of Stephane in
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=527623, it seems that
there are two issues:
==========================================================
SIGHUP, SIGQUIT, SIGINT ignored after package installation
After an apt-get install SIGHUP (as well as SIGQUIT and SIGINT) is
ignored by mysqld_safe. This is due to the fact that apt blocks SIGHUP,
SIGQUIT and SIGINT which is propagated to dpkg, the maintainer shell
scripts, the mysql init shell script and finally mysqld_safe (a shell
script). However SIGTERM is not.
What are the consequences? A mysql refresh command cannot be sent to
mysqld by sending a SIGHUP to mysqld_safe.
Since SIGTERM is not ignored mysqld_safe can properly shutdown mysqld.
It seems that this issue is not related to the bug described here and
new bug should be opened instead.
===============================================================
SIGTERM may be ignored when bash is the default shell (/bin/sh)
If /bin/sh points to bash (not the default configuration), SIGHUP (and
other signals) will *always* be ignored by mysqld_safe after the first
SIGHUP has been sent to mysqld_safe. This is a change in behavior
brought by the new upload.
The consequences are similar to the ones mentionned above with the added
case that SIGTERM will also be ignored. While this is more problematic,
it should be noted that the mysql init script tries *first* to shutdown
mysqld via mysqladmin, and uses SIGTERM on mysqld (and not mysqld_safe)
if it fails, and then SIGKILL on mysqld (and not mysqld_safe) if SIGTERM
fails.
Before the upload the behavior was that on every SIGHUP (regardless of
the shell), mysqld_safe would kill mysqld (via SIGKILL) and restart
mysqld. Since mysqld was killed with SIGKILL a proper shutdown was *not*
done, which could lead to data corruption.
The proposed upload removes a potential data corruption for mysqld
except in the case where bash is used as the default shell (/bin/sh).
In this non-default configuration (bash as the default shell):
* sending a SIGHUP to mysqld_safe would lead to:
- before (ubuntu10): mysqld killed with SIGKILL and restarted (potential
data corruption)
- after (ubuntu10.2): refresh the first time, nothing on subsequent
signaling.
This is the bug reported here and the upload provides an improvement.
* sending a SIGTERM to mysqld_safe would lead to:
- before (ubuntu10): proper shutdown of mysqld
- after (ubuntu10.2): proper shutdown of mysqld if no SIGHUP has been sent
before, nothing if a SIGHUP has already been sent to mysqld_safe.
This is a regression.
However it should be noted that the mysql init script doesn't send
signals to mysqld_safe when stopping mysql and is run before the
sendsigs script during the system shutdown process (rc6).
** Changed in: mysql-dfsg-5.0 (Ubuntu Karmic)
Status: In Progress => Triaged
--
mysqld_safe thinks mysqld has crashed when it hasn't
https://bugs.launchpad.net/bugs/326768
You received this bug notification because you are a member of Ubuntu
Server Team, which is subscribed to mysql-dfsg-5.0 in ubuntu.
--
Ubuntu-server-bugs mailing list
Ubuntu-server-bugs@lists.ubuntu.com
Modify settings or unsubscribe at:
https://lists.ubuntu.com/mailman/listinfo/ubuntu-server-bugs
