HugeResourceTest failing

[gregw]

Jetty version(s)
12

Description
The HugeResourceTest was failing for me by never completing on the 1G size. I tried making a 100M size, but whilst that passed it actually failed.
However, the CI is green, so either this test is not run there or it is working.

For now, I have disabled the test as it is failing for me in 12 HEAD and delaying other work.

[olamy]

This test is tagged at class level as @Tag("large-disk-resource") so this doesn't run on CI.
tags excluded from the CI build are external, large-disk-resource, stress, slow, flaky

[gregw]

@olamy thanks for confirming that.
It feels like we've re-introduced something non-linear. Probably the buffer pool issue we are already looking at. @lorban can you try this test out with your latest pools.

[lorban]

(It is ee10 HugeResourceTest that is being discussed here.)

My investigation has shown this test suffers from two bugs:

• Performance of ArrayRetainableByteBufferPool.acquire() can degenerate pathologically as the buckets grow in size #9311 which explains why this test takes so long. This can be worked around by configuring a ByteBufferPool.NonPooling pool while waiting for the fix.
• Using a large part size makes MultiPartFormData perform a double buffer release which makes the part be silently dropped. That explains why the test fails. I suspect that the test passes with small part sizes because of a buffer leak. There seems to be something wrong with the buffer tracking in this code.

[gregw]

Yes this is org.eclipse.jetty.ee10.webapp.HugeResourceTest
So you are already working on something for #9311, so let's not change anything in the test for that and we can use it to test the better pools.
As for the double buffer release, that sounds like a bug, so we should keep this open, even after #9311, until we fix that.