With a stopwatch I'm seeing test_websocket.html take ~13 sec on aurora, and around 22 secs on mozilla-central. There's one additional test on m-c, but disabling it (test-47) doesn't make any difference. This could well be a websockets bug, but I also stopwatched http://websocketstest.com/ and both aurora/m-c clock in at ~10 seconds. Both the mochitest and the site do a lot of websocket connections, but the mochitest does more. Do we know of any changes to the mochitest proxy server or anything else that might be driving this? Figured I'd ask before I wade in...
6 years ago
The pywebsocket stuff hasn't changed in a long time. I don't think there have been any other changes that would make a difference here. My memory is fuzzy, but I think we direct websocket connections directly to the pywebsocket server, so no other Mochitest changes should matter.
(In reply to Ted Mielczarek [:ted] from comment #1) > fuzzy, but I think we direct websocket connections directly to the > pywebsocket server, +1
Well then this is probably websockets. I guess we should consider adding some websockets component to Talos or otherwise start keeping tabs on WS performance so we can catch these more easily.
Component: Mochitest → Networking: WebSockets
Product: Testing → Core
QA Contact: mochitest → networking.websockets
I'm not seeing this now. M-c's performance is the same as aurora's--around 13sec. I've tried a bunch of things to try to repro--including fiddling with proxy settings, winding back my repo to the time of my initial report, and running the microbenchmarks I'm writing for bug 770718 (no differences there either). So marking this WORKSFORME for now, and I'll keep an eye on it. The Talos tests should hopefully catch stuff too.
Status: NEW → RESOLVED
Last Resolved: 6 years ago
Resolution: --- → WORKSFORME
You need to log in before you can comment on or make changes to this bug.