Description:
This is a followup of BUG#14106. The problem is that
signals are not handled (or dispatched) correctly
in Instance Manager, so it can not shutdown gracefully.
This problem appears at least on FreeBSD 5.4
(buildb.mysql.com, freebsd5.4-i386).
"Good" platforms at least are:
- Linux 2.6 i386/amd64;
- Sun Solaris 10 (sol10-amd64-a.mysql.com, solaris2.10-i386);
How to repeat:
1. Start Instance Manager as a daemon;
2. Send TERM signal to the IM process (not Angel);
3. Investigate log-files and find that IM didn't handled signal.
-----------------------------------------------------------------
Proper behaviour (on linux):
1. Start Instance Manager:
$ ./mysqlmanager --defaults-file=../../mysql-test/var/im.cnf --run-as-service --log=/tmp/im.log
2. /tmp/im.log contains:
060628 22:46:58 IM pid file: '/home/alik/MySQL/devel/5.0-rt-build/mysql-test/var/run/im.pid'; PID: 6343.
060628 22:46:58 Angel pid file: '/home/alik/MySQL/devel/5.0-rt-build/mysql-test/var/run/im.angel.pid'; PID: 6342.
060628 22:46:58 guardian: starting instance mysqld1
060628 22:46:58 starting instance mysqld1
060628 22:46:58 accepting connections on ip socket
060628 22:46:58 accepting connections on unix socket /home/alik/MySQL/devel/5.0-rt-build/mysql-test/var/tmp/im.sock
3. Send TERM signal:
$ kill 6343
4. /tmp/im.log contains:
060628 22:47:21 Listener_thread::run(): shutdown requested, exiting...
So, IM recognised request to stop and shutted down gracefully.
-----------------------------------------------------------------
Buggy behaviour:
1. Start Instance Manager:
$ ./mysqlmanager --defaults-file=../../mysql-test/var/im.cnf --run-as-service --log=/tmp/im.log
2. /tmp/im.log contains:
060628 20:46:14 Angel pid file: '/usr/home/mysqldev/alik/mysql-5.0.23/mysql-test/var/run/im.angel.pid'; PID: 19850.
060628 20:46:14 loaded user im_admin
060628 20:46:14 IM pid file: '/usr/home/mysqldev/alik/mysql-5.0.23/mysql-test/var/run/im.pid'; PID: 19851.
060628 20:46:14 accepting connections on ip socket
060628 20:46:14 guardian: starting instance mysqld1
060628 20:46:14 starting instance mysqld1
060628 20:46:14 accepting connections on unix socket /usr/home/mysqldev/alik/mysql-5.0.23/mysql-test/var/tmp/im.sock
3. Send TERM signal:
$ kill -TERM 19581
4. /tmp/im.log contains:
060628 20:46:35 angel(): mysqlmanager exited abnormally: respawning...
060628 20:46:36 Angel pid file: '/usr/home/mysqldev/alik/mysql-5.0.23/mysql-test/var/run/im.angel.pid'; PID: 19850.
060628 20:46:36 loaded user im_admin
060628 20:46:36 IM pid file: '/usr/home/mysqldev/alik/mysql-5.0.23/mysql-test/var/run/im.pid'; PID: 19862.
060628 20:46:36 accepting connections on ip socket
060628 20:46:36 accepting connections on unix socket /usr/home/mysqldev/alik/mysql-5.0.23/mysql-test/var/tmp/im.sock
So, Angel treated killing of IM-process, as fault and respawned the process.
