opensuse 10.3, gcc 4.8.3
compiled with ./BUILD/compile-pentium64-valgrind-max

I was unable to repeat the failure. The only reason for the failure that I can think of is that the server is crashing for some unexpected reason, before this statement in ibuf0ibuf.cc is executed:

			DBUG_EXECUTE_IF(

				"crash_after_log_ibuf_upd_inplace",

				log_buffer_flush_to_disk();

				ib_logf(IB_LOG_LEVEL_INFO,

					"Wrote log record for ibuf update in "

					"place operation");

				DBUG_SUICIDE();

			);

Next time this occurs, please save the datadir and the server error log for analysis. It should be easy to analyze because during the time of the failure, the server is not running, and we could also use the InnoDB redo log for figuring out what happened.

The problem is that in xtradb there is a LOT of log entries of type:
innodb_change_buffering_debug flush 4 45
innodb_change_buffering_debug evict 4 45

So many that the my_restart.err files becomes more than 50,000 characters, which causes search_pattern_in_file.inc to fail.

Fix: Don't flood the buffer or increase the search limit to 70,000 characters.

The error log messages are part of the `innodb_change_buffering_debug` instrumentation that was added to MySQL 5.5 by me in 2010.
There is no real need to flood the server error log with those messages. We can remove them.

However, it is interesting that XtraDB is writing much more of these messages than InnoDB (1064 vs 532). That could be a sign of some real problem.