Comments on: Tuning ‘log file sync’ Event Waits

By: shamsudeen

shamsudeen — Mon, 30 Jun 2008 23:08:37 +0000

Hi Alex
Good catch. Yes, you are correct about that line. semnum should be 118 in that line. Thanks.

Thanks
Riyaj

By: Polarski Bernard

Polarski Bernard — Mon, 30 Jun 2008 06:47:12 +0000

Beside playing with commit_write, there is very little action that a production DBA can do about logfile synch. Usually, once you diagnose too much commit, the lead developer replies that that the application is built like that and amen.

By: Alex Gorbachev

Alex Gorbachev — Sun, 29 Jun 2008 13:28:35 +0000

Riyaj, Excellent post indeed! My favorite part is LGWR & semaphores overview. Hey, I think I noticed a wrong trace line in step 4 of Signals, Semaphores and LGWR section...

After successful completion of the write(s), LGWR posts a semaphore of the waiting process using the semctl command.
22459/1: semctl(9, 15, SETVAL, 1)                   = 0

If I understand correctly, the semaphore with sem_num 118 -- the one that server process is waiting on inside log file sync.

By: Syed Jaffar Hussain

Syed Jaffar Hussain — Sat, 28 Jun 2008 06:37:49 +0000

Nice explanation to tackle the ‘log file sync’ wait event.
Long time back, we had this issue on AIX os (Oracle v10g) on one of the business critical database with 500+ transaction per second. After we enable CIO option on the filesystem, this wait event has been disappeared.

Jaffar

By: Doug Burns

Doug Burns — Fri, 27 Jun 2008 23:42:09 +0000

Excellent post. It’s funny that the various causes of log file sync cropped up during a presentation the other week. I wish I could have pointed people at this blog post. I will next time.