don't use external service for testing websockets #1138
Conversation
|
Hmmm .. I technically did directly in #1117 but I can cherry-pick it in a different PR and rebase it when this get merged ... |
mstoykov
force-pushed
the
dontUseExternalSiteForTestingWebsockets
branch
from
3f54c7f
to
e108a97
Compare
|
After a bunch more tries at fixing the panics and getting ... them rarer and rarer I am of the opinion that it's a lot better if I just remove the logs and assert/require's in the ws handlers. I don't think that the handlers themselves should check for this things I think that the test should be able to detect whether the handle did what they wanted and if this is not possible a custom handler and that is synchronized to exist before the test ends will need to be written. |
Codecov Report
@@ Coverage Diff @@
## master #1138 +/- ##
==========================================
- Coverage 73.58% 73.55% -0.03%
==========================================
Files 144 144
Lines 10527 10527
==========================================
- Hits 7746 7743 -3
- Misses 2323 2326 +3
Partials 458 458
Continue to review full report at Codecov.
|
mstoykov
force-pushed
the
dontUseExternalSiteForTestingWebsockets
branch
from
3e4daa7
to
2c27c56
Compare
This _should_ fix flaky test `TestSentReceivedMetrics`. See https://circleci.com/gh/loadimpact/k6/6474, https://circleci.com/gh/loadimpact/k6/6484 . This issue was introduced in a63bb58 (PR #1138), where the previous implementation of `getWebsocketEchoHandler()` closed the connection after writing, so this change brings that back. It seems that closing the connection creates additional metrics which `TestSentReceivedMetrics` takes into account _sometimes_, hence the flakiness.
This _should_ fix flaky test `TestSentReceivedMetrics`. See https://circleci.com/gh/loadimpact/k6/6474, https://circleci.com/gh/loadimpact/k6/6484 . This issue was introduced in a63bb58 (PR #1138), where the previous implementation of `getWebsocketEchoHandler()` closed the connection after writing, so this change brings that back. It seems that closing the connection creates additional metrics which `TestSentReceivedMetrics` takes into account _sometimes_, hence the flakiness. From discussions with @na--, we agreed to remove the persistent connection implementation of a63bb58 (the `for` loop here) since that matches the previous version, but we'll create a new WS test that tests multiple message passing, since none of our current tests seem to do this.
part of #537