One change was that the builder was recently bumped to Node 13.

I think that's unrelated. When I've seen this, the error is also as Bryan quoted above: a timeout doing a gRPC call to maintner, asking whether we should even do this build. We used to need to skip builds for js/wasm if the git history of the rev to be built didn't include js-wasm support. But now that we've supported js/wasm for a number of releases (since Go 1.11) and we don't even support Go 1.10 (or even Go 1.11) any more, we can just remove that condition on that builder.

But also: maintner shouldn't be so slow at that query.

Change https://golang.org/cl/203757 mentions this issue: dashboard: stop checking whether the branch/rev supports js/wasm for js/wasm

Not very scientific, but the has-ancestor RPCs are taking only 100ms for me. (includes TCP+TLS setup, etc) Not anywhere near 30 seconds.

$ time maintq has-ancestor 0ae9389609f23dc905c58fc2ad7bcc16b770f337 3dced519cbabc213df369d9112206986e62687fa
has_ancestor:true

real    0m0.105s
user    0m0.076s
sys     0m0.028s

@dmitshur - Would you be able to check if this is still happening ?

@dmitshur Do you know if this is still an issue?

I feel like I've seen some discussion about this still happening occasionally quite recently, more so than the comments in this issue would suggest. Perhaps it was in another thread. I'll keep an eye out for this.

Yes, it is still happening. Last week I caught the js-wasm trybot taking 25 minutes when most were done in 15,
and just now I caught it taking 34 minutes when everything else was done in 16.
That one was https://go-review.googlesource.com/c/go/+/266357 and I gave up waiting and just submitted the CL.

Any advice about what to do when js-wasm looks like it is stuck/very slow and how to debug further would be greatly appreciated.

Any advice about what to do when js-wasm looks like it is stuck/very slow and how to debug further would be greatly appreciated.

I suggest trying to determine if there is a specific place where the build progresses more slowly than others. I.e., is it perhaps that time to first test is delayed, then it's fast, or is one of the package tests very slow, while the rest is fast, or is slowness equally distributed across all packages?

I'll also try to look into logs for the two occurrences you've shared and see if I can determine it from that.

From CL 266374 where I recently started trybots:

In this instance, the js-wasm builder is in waiting_for_machine state for over 10 minutes. That is very unexpected—it's a GCP builder that we should be able to spin up without running out of physical machines. Perhaps we're running out of some quota. Not sure why only js-wasm and not other GCP builders. But this is a lead.

Edit: Here's the scheduler state at the time:

Issues #31282, #36088 may be related.

In several occurrences that I saw, the js-wasm TryBot started running and partially completed, then descheduled and started over again from the beginning. I think there is some sort of failure being buried by retry logic, rather than just a slow test.

If we don't have the bandwidth to diagnose this at the moment, perhaps we should temporarily demote js-wasm to an opt-in SlowBot?

I'm also experiencing js-wasm slowdowns for my trybot runs. Here is an example:

Log (https://farmer.golang.org/temporarylogs?name=js-wasm&rev=ff065bd219cd7a9df01466e338b28f1891671da3&st=0xc0184e91e0):

  builder: js-wasm
      rev: ff065bd219cd7a9df01466e338b28f1891671da3
 buildlet: (nil *buildlet.Client)
  started: 2020-10-30 16:46:48.188839196 +0000 UTC m=+727935.299326955
   status: still running

Events:
  2020-10-30T16:46:48Z checking_for_snapshot 
  2020-10-30T16:46:48Z finish_checking_for_snapshot after 23.9ms
  2020-10-30T16:46:48Z get_buildlet 
 +2093.2s (now)

Build log:

(buildlet still starting; no live streaming. reload manually to see status)

Seems as though it is not getting past "get buildlet".

[Edit: did finally make progress after 40 min wait.]

I don't know much about farmer, but on https://farmer.golang.org/#sched it seems to me as if multiple trybot runs are waiting on the js-wasm tests, but I can only see very few buildlets with the prefix GCE VM: buildlet-js-wasm-. Is something preventing these buildlets from spawning?

Is something preventing these buildlets from spawning?

It seems so. I don't have an idea about what it might be yet. It may be related to #42285. I'm going to look more later today.

All this is to say that more work needs to be done to get closer to finding and resolving the root issue(s) here. I think we have confirmation that this problem is intermittent and skipping a deterministically slow test is not an available option.

Change https://golang.org/cl/347295 mentions this issue: dashboard: disable reboot test on js-wasm builder

Is anyone still observing this issue?

Is something preventing these buildlets from spawning?

A new idea by now is that it might've been hitting the GCE VM quota. Some of those quotas are visible in the first line of https://farmer.golang.org/#pools.

If there are no frequent timeouts affecting this builder by now, perhaps we should close this.

I haven't seen any of these recently, but I also haven't been looking for them.

I suspect that the “failure being buried by retry logic” that I mentioned in #35170 (comment) may have been the one you fixed in CL 407555.

Since we can identify a specific change that plausibly could have fixed this issue, I'd be ok with closing it until / unless we observe it again.

Sounds good, closing since I haven't seen reports of this in a long while, can reopen or file another issue if this comes up again.

From looking at the initial report, there's also a chance that the same problem as in issue #55947 (fixed as of Oct 2022) may have contributed here too.