[common] Correct definition of alternate host name

[jugglinmike]

The value of the "alternate" host is completely at the discretion of the
server operator (via the alternate_hosts configuration). It is not
guaranteed to have any resemblance to the primary host name. Despite
this, the shared get-host-info.sub.js file attempted to derive the
alternate host name in terms of the primary host name.

Replace that faulty derivation with the value which is present in the
server configuration.

[jugglinmike]

@Hexcles this patch failed Chrome's stability check. I'm trying to determine if it's due to the content of the change, the number of affected tests, a glitch in the infrastructure.

The logs report

Identified 333 affected tests

And finally

TEST-START | /signed-exchange/sxg-referrer-policy-header.tentative.https.html
Run 1/10
mem avail: 4277 MiB (57 %), swap free: 0 MiB (0 %)

[taskcluster:error] Task timeout after 7200 seconds. Force killing container.
[taskcluster 2019-07-17 04:05:31.541Z] === Task Finished ===
[taskcluster 2019-07-17 04:05:31.542Z] Unsuccessful task run with exit code: -1 completed in 7201.772 seconds

We don't have timestamps, but I think the runner was actually hanging here. That's because we only intermittently print memory information, so it's unlikely to have occurred during a normal test execution.

Do you think we should just retry?

[jugglinmike]

...then again, if the runner was hanging for a significant length of time, then we'd expect to see a number of messages about memory availability.

[Hexcles]

I think the task is most likely timing out because of too many tests affected, but there could an infra glitch that led to incomplete (unflushed?) logs.

[jugglinmike]

@Hexcles Further research supports the theory that the timeout is due to the number of tests being run, not the change itself (see below). Would you be up for admin-merging? Or would you like me to check something else, first?

---

This second attempt timed out on a different test:

TEST-OK | /signed-exchange/sxg-location.tentative.html | took 3478ms
TEST-START | /signed-exchange/sxg-location.tentative.html
Run 8/10

[taskcluster:error] Task timeout after 7200 seconds. Force killing container.
[taskcluster 2019-07-17 18:02:55.315Z] === Task Finished ===
[taskcluster 2019-07-17 18:02:55.315Z] Unsuccessful task run with exit code: -1 completed in 7201.861 seconds

So this patch isn't causing a deterministic timeout. It may be exposing some flaky behavior, but I can't reproduce that locally. I've run the stability checker on each of the tests that were running when the previous two builds timed out:

$ python ./wpt run --channel=dev --verify --verify-log-full --no-pause --no-restart-on-unexpected --install-fonts --no-headless chrome signed-exchange/sxg-location.tentative.html signed-exchange/sxg-referrer-policy-header.tentative.https.html

That command produces the same result on master and the pull request branch:

$ python ./wpt run --channel=dev --verify --verify-log-full --no-pause --no-restart-on-unexpected --install-fonts --no-headless chrome signed-exchange/sxg-location.tentative.html signed-exchange/sxg-referrer-policy-header.tentative.https.html
Ran 30 tests finished in 14.8 seconds.
  • 0 ran as expected. 0 tests skipped.
  • 15 tests timed out unexpectedly
  • 30 tests had unexpected subtest results

$ echo $?
0

I opened a draft pull request with an inconsequential change to the same file, and the stability checks timed out in Chrome there, as well.

[Hexcles]

I stumbled upon #15819 which seems like an easy fix (add one more field to the returned object). While we are at it, shall we fix that as well?

[jugglinmike]

Reading over that request, I have differing opinions about the technical implementation (I'd rather define that as ALT_HOST to reuse the terminology from the configuration file and make their relation clear). Sorry to say, but my own objection makes it less of an easy fix.

[Hexcles]

Yeah that's cool. I'll merge this now. Thanks!

[jugglinmike]

Thank you! Following up on gh-15819 presently