On Feb 26, 2014, at 02:45 PM, Stephen J. Turnbull wrote:
I have no idea what to do next,
This is clearly a bug, although I think it's relatively recent, so it might be worth seeing if earlier revisions avoid the problem. Yes, I can reproduce it.
The interesting thing is that the test is in rest/docs/membership.rst so these are multiprocess related bugs. Typically when this happens (and it will only be with the default SQLite database, as observed by others), it's a bug in the test, not necessarily a bug in the core.
Tests which involve multiple processes, as the REST tests do (i.e. the foreground testing process and a background runner process) have to be careful to release the database lock when they expect background processes to access the database. Releasing the lock means calling .commit() or .abort() at the appropriate time. The thing to keep in mind is that with Storm, even doing an operation that results in a database query opens a transaction and thus acquires the lock.
In the context of membership.rst, what this probably means is that somewhere in the doctest there's a database query with a missing explicit .commit() or .abort() before the background REST runner process executes. Tracing through the doctest to find out exactly where it hangs usually helps isolate where the missing commit/abort should go.
Of course, it's possible that there's a missing commit/abort in the core, but I rather doubt it, since that's pretty well tied into the REST runner's HTTP transaction machinery, and other REST tests don't exhibit the hang.
Tom, if you're up for debugging it, that would be great. If not, no worries. The test suite hangs for me, so I'll find some time this weekend to take a look.
-Barry