1505037 - Worker page takes a long time to load

Status: RESOLVED INCOMPLETE

(Taskcluster :: UI, enhancement)

Description

[Kartikaya Gupta (email:[e-mail])]

Going to https://tools.taskcluster.net/provisioners/localprovisioner/worker-types/webrender-ci-osx for example takes a really long time to load. Given there's only two workers of this time it seems like it's doing an excessive amount of work.

Comment 1

[Chris Cooper [:coop] (he/him)]

Hassan: you may want to have a look at this. I tried loading it in the new tc-web version and it was also longer than expected. Would be good to figure out where the slowdown is coming from.

Comment 2

[Hassan Ali (:hassan)]

I can't seem to replicate on tc-web. Kartikaya, could you try and see if it takes you a long time to load https://taskcluster-web.netlify.com/provisioners/localprovisioner/worker-types/webrender-ci-osx, please? Let me know how long it takes before you see the workers table.

Comment 3

[Kartikaya Gupta (email:[e-mail])]

I loaded the netlify.com URL twice - first time took around 6 seconds and the second time about 2 seconds. Then I went back to the taskcluster.net URL and that also loaded quite fast, maybe 1-2 seconds. Does the data get cached somewhere? I'll try again in a bit and see if it's slow again.

Comment 4

[Hassan Ali (:hassan)]

I looked into it. The UI is not caching the result. The call to Azure is returning empty results with continuation tokens. I believe this is related to Bug 1497560.

Comment 5

[Hassan Ali (:hassan)]

This should hopefully be better once we move to postgres. See https://bugzilla.mozilla.org/show_bug.cgi?id=1497560#c3 for more information.

Comment 6

[Kartikaya Gupta (email:[e-mail])]

I had to load this worker list a bunch today and it was slow on both tools.taskcluster.net and the netlify.com version. Took about 22 seconds on netlify.com just now when I timed it.

Comment 7

[Dustin J. Mitchell [:dustin] (he/him)]

It's known that this is slow -- there's not much we can do about it because of the backend datastore we use.  Azure table storage is pretty slow when searching for matching rows, as it basically searches partition-by-partition and returns an empty dataset for partitions with no matching rows, with a continuation token.  The new site is handling the sequential fetching on the server side, but that doesn't really make a difference in total duration.

Postgres is obviously much more flexible in its querying abilities, but transitioning all of TC to postgres is not something that is going to happen overnight.  That work is tracked in bug 1436478 and its dependency tree.

Until that's done, the slowness is tolerable (and this isn't the only place it appears).  I'll call this INCOMPLETE since the page remains slow, but aside from completing bug 1436478 there's no action to take to improve the situation.

Comment 8

[Kartikaya Gupta (email:[e-mail])]

I was under the impression that the netlify.com version was using postgres, but my experience today indicates that it's not helping. At least not consistently.

Comment 9

[Dustin J. Mitchell [:dustin] (he/him)]

Sorry, no, still Azure-backed.

Comment 10

[Kartikaya Gupta (email:[e-mail])]

Ah ok. In that case I'll wait for the postgres version. Thanks!