-
Notifications
You must be signed in to change notification settings - Fork 523
Flaky test: MaxRequestBufferSizeTests.LargeUpload #2225
Comments
Still failing, same error. |
cc @halter73 |
@Tratcher What OS? Have you noticed it fail on the 2.0 or 2.1 runtime? Both? |
You can see the history of this test failure here. |
Another failure, but a different error this time:
|
@Tratcher did you get a chance to look into this yet? |
No, I should have some time to look at tests tomorrow. |
RE the InRange failure, it does not appear to be TFM specific, and it reproduces both on Sockets and Libuv, though more readily on Sockets. Only the connectionAdapter: True variation fails with the above error. |
Re:
|
Mitigation for the original issue has been checked into dev for 2.2. If we see it again in 2.1 we can consider backporting it. If we see any more reports of the other errors we can open a separate issue. |
Nevermind, it's still failing. |
This fails most frequently on our Win10 build, on which it has failed 4 times today. |
@Tratcher did you look into this one yet? |
I wasn't able to get a consistent repro so I moved on to tests with more failures. I'll revisit it when I get that far down the list. |
Just failed on my dev machine, same error as the first post in this issue:
It's possible the client and/or server buffers on Win10 have been increased, so we may need to change the test parameters: KestrelHttpServer/test/Kestrel.FunctionalTests/MaxRequestBufferSizeTests.cs Lines 49 to 52 in de98f8c
KestrelHttpServer/test/Kestrel.FunctionalTests/MaxRequestBufferSizeTests.cs Lines 222 to 225 in de98f8c
Specifically, we may need to increase the data sent from 20MB to something larger until the test is reliable on all platforms. |
I can take this issue since I wrote these tests. |
On my dev machine, the failure was in The next step is to determine if the flakiness is caused by test code (e.g. the 15MB difference between upload size and max request buffer is not large enough) or product code (e.g. sometimes the server fails to apply backpressure, which would be a product bug). |
I've been running the LargeUpload tests in a loop on my dev machine:
The following test (and only this specific test) fails about 25-30% of the time:
It may be noteworthy that I also tried running only the two tests with
However, when running this subset I was unable to repro any failures. |
I was also able to reproduce this in a Win10 Hyper-V VM on my dev machine. It fails significantly less often than my physical machine, but I was able to repro some failures with
|
- Increase _dataLength from 20MB to 40MB to improve test reliability when using Sockets transport on Windows - Addresses #2225
New occurance: https://ci3.dot.net/job/aspnet_KestrelHttpServer/job/dev/job/linux-Configuration_Release_prtest/307/
|
This has failed 50% of the time for me on recent local builds, always with the same actual result.
The text was updated successfully, but these errors were encountered: