In file_fsync(), for the mmap mode of fileio operation, a successful
return sentence was missing, and the catch-all error return sentence was
being activated for this mode. This is fixed by adding a proper return
value for this mode.
Similar to multiple hosts in the --mysql-host option, --mysql-socket can
now accept multiple comma-separated sockets. In which the value of
--mysql-host is ignored, and connections are created to all specified
sockets on the round-robin basis.
Fixed a race condition in the test termination code with respect to
--report-interval: the reporting thread might read a non-zero
report_interval value and call print_stats(). If report_interval is
changed to 0 by the master thread and the test's done() method called,
print_stats() could access some shared resources deallocated in
done().
Fixed by protecting report_interval with a separate mutex which is
locked on updates and during the check by reporting thread and
print_stats().
stats
When --forced-shutdown is used, it prints the currently collected
statistics and terminates without waiting for worker threads to
complete. This was implemented to be able to complete benchmark even
if we have some long running or stuck transactions.
The problem is that unfinished transactions are not reflected in
statistics. So, for example, if we do have stuck transactions that
have been stuck for the entire test duration, their timers will still
be running on forced shutdown and thus, statistics will only show
transactions performed by other threads, if any.
The fix is stop all timers on forced shutdown, so even stuck
transactions are accounted with (forced_shutdown_time -
trx_start_time) duration. Also printing the number of currently
running transactions at the time of forced shutdown would also make
sense.
The problem was in a hard-coded library (-ldl) being added to LIBADD.
Fixed by using the appropriate Libtool/Autoconf macros to check if using
-ldl is necessary.
The problem was a stack overflow in the event generation thread due to a
large array allocated on stack.
Fixed by changing the array to a global static variable.
Changed the fileio prepare command so that any existing files left from
the previous runs are now reused rather than recreated from
scratch. This is is mostly useful for SSD drives, but may speed up the
prepare stage for regular drives too.
intervals.
Added a new --report-checkpoints option that allows to dump full
statistics and reset all counters at specified points in time. The
argument is a list of comma-separated values representing the amount of
time in seconds elapsed from start of test when report checkpoint(s)
must be performed. Report checkpoints are off by Report checkpoints
are off by default.
* advancing position for sequential tests is actually a good idea
* replaced LOG_INFO with LOG_NOTICE for test setup messages to make them
visible by default
Due to an incorrect merge from 0.4, some code paths in
file_get_seq_request() did not properly release the exec mutex before
returning. This led to deadlocks in sequential fileio tests.
