High Avw Wait latch: row cache objects
In a recent stress test, we encountered "latch: row cache objects" as a top wait event, with avg wait of 343ms!!! R10.3, on RAC, heavy INSERT OLTP on dedicated single node.
In AWR, I see very high sleeps/sleep waits on
Sleeps Sleep Waits
row cache objects kqrpre: find obj 0 218,359 119,592
row cache objects kqreqd: reget 0 68,523 114,190
row cache objects kqreqd 0 54,203 107,286
While I would normally assume expanding SGA, the shared pool is 10GB (and haven't seen this problem before).
I'd also look for high dynamic SQL parsing or hard parsing, but don't see it (unless they snuck something in on the test that doesn't show in AWR).
Execute to Parse Ration ranges from 61% on 1st run, to 80% on 3rd run, no restart/shared pool flush.
Any suggestions would be greatly appreciated.