D
Dapeng Wu
Guest
The messages in the .lg file do look very strange or out of order. In the normal case, we should see (562) first, then (2252), then (2253). The mechanism to back out a transaction depends on the transaction entry (trid) in the transaction table in shared memory, but it looks the backout didn't do anything while online. And the real backout happened in the crash recovery phase, where the transaction table will be reestablished from the BI files, and this transaction table seems more reliable than the previous one. Message (1077) cannot be explained either, it may be picked up from a wrong location. Once we see (2256), the watch dog will set a FATAL flag in shared memory and the database will start an emergency shutdown process. In most cases, the messages following that are not reliable since we are in an emergency. So although the ABL process was already dead, the broker may still try to clean up something for it. Put all these together, it looks like a memory stomp to me. And I won't rule out any OS issues either. Dapeng
Continue reading...
Continue reading...