Whenever I have some time to kill (boring meeting, waiting for something
to finish, etc) sometimes I will start up a test environment in the clown.
I have a subscription to a training site that allows free time-limited
sandboxes
in the three big hyperscale clowns. And I'll run a Citadel Server and max
out the load tester on it.
Today's test ran for four hours. I had it "kill -11" the server process
every few minutes on a randomly timed basis. The load tester ran 240 threads
and was configured to automatically restart whenever it lost the server.
By the end of the sandbox's lifetime, it had completed close to a million
transactions, with a few hundred "induced" server crashes and NO unintentional
server crashes ... and not one tiny bit of database corruption. None. Nada.
Nunca. Jamais. I know I've said this a number of times before, but I am
just so damn satisfied with
how stable the database layer is in server >= 993 after last summer's effort
to make it more robust.
I'd be interested in hearing any feedback y'all might have on this. My test
systems have been on AMD64 with both Linux and FreeBSD, and on ARM with Linux.
Production is 32-bit x86 with Linux, and has also been completely stable.
Does anyone have experience to report, either confirming or refuting that
we made an improvement?
