This code change has 0 effects as it's about the case where innobase_very_fast_shutdown!=0,

which is always false. In a very fast InnoDB shutdown, we just ensure that
no more transactions are running, flush InnoDB log, signal InnoDB threads to die,
and then return from InnoDB (from innobase_end()) without waiting for those threads
to actually die. I have tested on a 4CPU machine that even with --innodb_flush_log_at_trx_commit=0,
this optimized InnoDB very fast shutdown loses no committed transactions. Patch pre-approved by Heikki.


innobase/log/log0log.c:
  In an InnoDB very fast shutdown, we just need to wait for no more transactions to be happening
  and then we can flush the InnoDB log and don't need to wait for the
  signaled-to-die InnoDB threads to finish (saves seconds).
innobase/srv/srv0start.c:
  In an InnoDB very fast shutdown, once we have forced a flush of the InnoDB
  log to disk, and signalled InnoDB threads to die, we needn't wait
  for these threads to die.

1 files changed, 7 insertions, 0 deletions

diff --git a/innobase/srv/srv0start.c b/innobase/srv/srv0start.c
index 210739b26fd..c65b7d3e141 100644
--- a/innobase/srv/srv0start.c
+++ b/innobase/srv/srv0start.c
@@ -1740,6 +1740,13 @@ innobase_shutdown_for_mysql(void)
 
 	srv_shutdown_state = SRV_SHUTDOWN_EXIT_THREADS;
 
+        /* In a 'very fast' shutdown, we do not need to wait for these threads
+        to die; all which counts is that we flushed the log; a 'very fast'
+        shutdown is essentially a crash. */
+
+        if (srv_fast_shutdown)
+          return((int) DB_SUCCESS);
+
 	/* All threads end up waiting for certain events. Put those events
 	to the signaled state. Then the threads will exit themselves in
 	os_thread_event_wait(). */