On 30/09/2010 14:46, Phil Mayers wrote:
On 30/09/10 14:36, Chris Withers wrote:
Indeed, but, as I keep saying, I need to build a scheduler that's resilient to errors in the tasks its running ;-)
Sure.
Most prominent is reactor.callLater.
...which I'm not actually using, it was just a small-as-possible way I could simulate the failure mode (rather than the specific failure) I need to protect against.
Ah.
Actually, what appears to work is simply changing `loop` to not be an async fuction:
def loop(): try: doStuff() except Exception,e: log.err(None,'Unhandled scheduled exception')
looper = task.LoopingCall(loop) looper.start(1.0) reactor.run()
This appears to solve most of the problems: - the schedule now keeps running regardless - exceptions in doStuff and below get logged
However, it now has the problem that if doStuff *does* return a deferred and it errbacks, the I get the ugly:
Well, you could do this, which is more Twist-y, and Glyph suggested:
def loop(): d = defer.maybeDeferred(doStuff, *args) d.addErrback(log.err)
Okay, but further down in the call stack of the "real app", I already have: @inlineCallbacks def sendTransmission(self...): ... try: yield maybeDeferred(feed.initiateTransmission, ...) except Exception, ex: if isinstance(ex,GeneratorExit): raise ... Is that not doing what you're talking about? If it is, it didn't help...
FWIW, you say "if doStuff does return a deferred"; I presume you don't really have a single call returning both normal and deferred values,
Correct, doStuff *should* always return a deferred, the problem is that the deferred returned never fires (errback or callback) when an exception is raised in the code that should be firing it. Chris