A short time ago I was asked to checkout the performance of an 4 node 64 bit linux RAC cluster on 9.2.0.7 RDBMS. After a short investigation I found out that the system was waiting for a very rare wait event, called: "ksu process alloc latch yield". This wait event was comming up on all nodes (statspack and current session waits).
Looking at Metalink the issue is only mentioned as part of bug 4658188 and the wait event was introduced in 9.2.0.7. The problem is only occurring in combination with RAC, because according the bug LMON (Lock Monitor) is waiting on the event. I was able to make sure, by mapping the session waiting on the event, to the LMON process (shows up as ora_lmon_<name>).
Because of the bug the LMON process is slow and in that way the whole RAC cluster. For this issue a one of patch is available on top of 9.2.0.7 (one-off patch for Bug 4658188).
Kind Regards,
Gerwin
thanks Gerwin for sharing this experience to the online community;
Karl
The note 343180.1 is an ALERT note from Oracle. The patch 4947798 you mention is a patch also including two other severe issues, so patch 4658188 is part of patch 4947798. The other fixed bugs are:
– Bug 4771380 – 9207 Upgrd Rac Db – 3 Instances All Crash Ora-4031, Core Dumps
– Bug 4923667 – Ora-27300, Ora-27301, Ora-27302 Err after apply patch 4778081
Regards, Gerwin Hendriksen
I ran into this bug lately on tru64 when using a TCP interconnect (not memorychannel) and guess what: the patch you mention is not availlable for tru64. Therefore I was pointed to patch 4947798 (via note 343180.1).
I am still wondering wat is the main difference though (yes it is on my shortlist for follow-ups, but so far other issues have more priority)