The cause of im_daemon_life_cycle.imtest random failures was the following
behaviour of some implementations of LINUX threads: let's suppose that a
process has several threads (in LINUX threads, there is a separate process for
each thread). When the main process gets killed, the parent receives SIGCHLD
before all threads (child processes) die. In other words, the parent receives
SIGCHLD, when its child is not completely dead.
In terms of IM, that means that IM-angel receives SIGCHLD when IM-main is not dead
and still holds some resources. After receiving SIGCHLD, IM-angel restarts
IM-main, but IM-main failed to initialize, because previous instance (copy) of
IM-main still holds server socket (TCP-port).
Another problem here was that IM-angel restarted IM-main only if it was killed
by signal. If it exited with error, IM-angel thought it's intended / graceful
shutdown and exited itself.
So, when the second instance of IM-main failed to initialize, IM-angel thought
it's intended shutdown and quit.
The fix is
1. to change IM-angel so that it restarts IM-main if it exited with error code;
2. to change IM-main so that it returns proper exit code in case of failure.
- Starting time of a query sent by bootstrapping wasn't initialized
and starting time defaulted to 0. This later used value by NOW-
item and was translated to 1970-01-01 11:00:00.
- Marketing the time with thd->set_time() before the call to
mysql_parse resolves this issue.
- set_time was refactored to be part of the thd->init_for_queries-
process.
- Starting time of a query sent by file bootstrapping wasn't initialized
and starting time defaulted to 0. This later used value by the Now-
item and is translated to 1970-01-01 11:00:00.
- marking the time with thd->set_time() before the call to
mysql_parse resolves this issue.
im_daemon_life_cycle fails randomly.
1. Move IM-angel functionality into a separate file, create Angel class.
2. Be more verbose;
3. Fix typo in FLUSH INSTANCES implementation;
4. Polishing.
- The condition variable implementation "lost" a signal to
WaitOnSingleObject when a semaphore was released.
- The signal could be consumed by a new call to pthread_cond_wait
before all waiting threads had awoken.
- The new implementation of pthread_cond_* uses events
instead of semaphores. It also uses an extra lock to protect entry
into new cond wait before the broadcast has finished.
Showstopper and regression against 5.0.24.
Previously, we ignored seek() errors (see Bug#22828) and let seek()s
against pipes fail. Now, since we check that a seek didn't fail,
and return without reading, this bug popped up.
This restores the behavior for file-ish objects that could never be
seek()ed.
