G
George Potemkin
Guest
The article missed the scenario that happend in this case and that I saw many times reported by our customers on Unix. It's easy to reproduce if you have a rather large database: kill -9 a session that reads data from disk. The chances to get "Wrong dbkey error" are higher than "Usr died holding latch". There are no corruptions: neither on disk nor in memory. "Writing block to log file" will dump a block contents that is correct for a "found" dbkey. You can compare it with a dump of the same block on disk in database. "Should be" dbkey is correct as well. "Wrong dbkey" reported by watchdog after a dead session that did NOT report the "wrong dbkey" error is always not a real corruption. It's confirmed many times.
Continue reading...
Continue reading...