HTTPS clone URL
Subversion checkout URL
The getRawObjects call can throw (eg, if you can't connect to the mongo server for too long). A few pieces of state were being corrupted in that case: - self._results was being set too early, leading to 'first' not being set on future _pollMongo calls, and_multiplexer.ready() never being called. This had two effects: - The observe (and thus any subscription) would never become ready(). Due to deduping, *no observe on this query* would ever become ready either. This also implies that the observeChanges that are part of _publishCursor would never return, so the sub.onStop would never get called, so the observeHandle would never stop, leading not only to leaks, but for an inability for that query to ever stop being deduped with the corrupted PollingObserveDriver! - The onFlush calls would throw a "not ready" error instead of calling the callback, so (a) errors would be logged and (b) write fences would never be closed Fixed this by not writing to self._results at the top of the function. - writesForCycle was being lost, so those write fences would never close. Fixed this by pushing writesForCycle back onto _pendingWrites if getRawObjects throws.