"Failed to fetch test runs" on prod with only 6 runs

[foolip]

I'm not able to load a filtered list of the Interop 2022 tests across 6 runs on prod today: https://wpt.fyi/results/?view=subtest&q=label%3Ainterop-2021-aspect-ratio%20or%20label%3Ainterop-2021-flexbox%20or%20label%3Ainterop-2021-grid%20or%20label%3Ainterop-2021-position-sticky%20or%20label%3Ainterop-2021-transforms%20or%20label%3Ainterop-2022-cascade%20or%20label%3Ainterop-2022-color%20or%20label%3Ainterop-2022-contain%20or%20label%3Ainterop-2022-dialog%20or%20label%3Ainterop-2022-forms%20or%20label%3Ainterop-2022-scrolling%20or%20label%3Ainterop-2022-subgrid%20or%20label%3Ainterop-2022-text%20or%20label%3Ainterop-2022-viewport%20or%20label%3Ainterop-2022-webcompat&run_id=5652738877685760&run_id=5680388870504448&run_id=5635446534045696&run_id=5131133808279552&run_id=5134205785210880&run_id=5125533909123072

What I see is this:

This is common on staging (https://github.com/web-platform-tests/wpt.fyi/issues/2617) but I haven't seen it on prod much.

cc @KyleJu

[KyleJu]

Looking at the logs, I don't see any obvious errors/issues

I am able to load the test runs when I was experimenting with fewer search atoms. It is probably a API timeout issue because it is taking too long to process, as it runs (the number of tests) * (the number of browsers) * (the number of label atoms) operations, then search cache will have to combine all the test results because these are or operands. I have two ideas:

• Label is filed at the test-level, regarding of the browsers. You can split this 6 browsers into two groups, where each group has the Chrome-Firefox-Safari, as shown below. It does work but it defeats the purpose of this specific search for 6 browsers.

https://wpt.fyi/results/?view=subtest&q=label%3Ainterop-2021-aspect-ratio%20or%20label%3Ainterop-2021-flexbox%20or%20label%3Ainterop-2021-grid%20or%20label%3Ainterop-2021-position-sticky%20or%20label%3Ainterop-2021-transforms%20or%20label%3Ainterop-2022-cascade%20or%20label%3Ainterop-2022-color%20or%20label%3Ainterop-2022-contain%20or%20label%3Ainterop-2022-dialog%20or%20label%3Ainterop-2022-forms%20or%20label%3Ainterop-2022-scrolling%20or%20label%3Ainterop-2022-subgrid%20or%20label%3Ainterop-2022-text%20or%20label%3Ainterop-2022-viewport%20or%20label%3Ainterop-2022-webcompat&run_id=5652738877685760&run_id=5680388870504448&run_id=5635446534045696

• The way we are searching label atoms is inherently inefficient, because it is independent of browsers. We could only run label atoms on the tests, instead of tests on one browser at a time, which will cut the number of operations to (the number of tests) * (the number of label atoms). However, we need a smart way to combine tests, as each test run could run a different sets of tests, e.g. deletion.

[foolip]

I just hit this again a few times, this time with the query (label:interop-2021-aspect-ratio or label:interop-2021-flexbox or label:interop-2021-grid or label:interop-2021-position-sticky or label:interop-2021-transforms or label:interop-2022-cascade or label:interop-2022-color or label:interop-2022-contain or label:interop-2022-dialog or label:interop-2022-forms or label:interop-2022-scrolling or label:interop-2022-subgrid or label:interop-2022-text or label:interop-2022-viewport or label:interop-2022-webcompat) exists(status:!pass) seq(status:!missing status:!missing status:!missing status:pass status:pass status:pass) with 3 runs from early March and 3 recent ones: https://wpt.fyi/results/?view=subtest&q=%28label%3Ainterop-2021-aspect-ratio%20or%20label%3Ainterop-2021-flexbox%20or%20label%3Ainterop-2021-grid%20or%20label%3Ainterop-2021-position-sticky%20or%20label%3Ainterop-2021-transforms%20or%20label%3Ainterop-2022-cascade%20or%20label%3Ainterop-2022-color%20or%20label%3Ainterop-2022-contain%20or%20label%3Ainterop-2022-dialog%20or%20label%3Ainterop-2022-forms%20or%20label%3Ainterop-2022-scrolling%20or%20label%3Ainterop-2022-subgrid%20or%20label%3Ainterop-2022-text%20or%20label%3Ainterop-2022-viewport%20or%20label%3Ainterop-2022-webcompat%29%20exists%28status%3A%21pass%29%20seq%28status%3A%21missing%20status%3A%21missing%20status%3A%21missing%20status%3Apass%20status%3Apass%20status%3Apass%29&run_id=5694898400395264&run_id=5671385434161152&run_id=5747379813744640&run_id=4818947338141696&run_id=4882567984054272&run_id=5158468428759040

When reloading I eventually get it to load.

@KyleJu this query also uses labels a lot, so if label queries could be made more efficient that would probably help, yeah.

[KyleJu]

Fixed in prod. Future improvement in https://github.com/web-platform-tests/wpt.fyi/issues/3323