tools: extend timeout for doc build version fetch

[rvagg]

5s is too short, one of our CI CentOS 7 machines fails on this, @MylesBorins noted that it regularly happens when he's doing releases. I don't know why this machine has problems (I'm updating it and restarting it atm) with a 5s timeout but it does.

Confirmed on that machine that extending the timeout gets it through doc-only.

Is 30s too much? The timeout is there to handle the no-internet case.

[rvagg]

https://ci.nodejs.org/job/node-test-commit-linux/33761/nodes=centos7-64-gcc6/console failure example

[addaleax]

Looks good, although imo we really really should not be doing this at all, or at least cache the results for a minute or two so that make test/make doc doesn’t request the same files 30 times from GitHub over and over.

[rvagg]

yes, I was going to open another issue about that, it's really tedious to watch this in action and when you imagine all those requests spread across our CI for each file then it's not really fair to GitHub (although I'm sure they can handle it!)

[MylesBorins]

LGTM as a temporary solution to at least stop CI from failing.

Honestly seems like we could avoid this problem all together by keeping a list of supported versions in the source rather than hitting the nework for this

[addaleax]

Honestly seems like we could avoid this problem all together by keeping a list of supported versions in the source rather than hitting the nework for this

@MylesBorins That’s what the tool falls back to when no network is available – but it intentionally tries to fetch from upstream first because the local copy in the source tree might be outdated (e.g. when on older branches)

[richardlau]

As I have a day off work today I've had some time to finish off a refactoring of the previous versions logic and submitted #32518 which should help (also bumped the timeout in that to match what's been PR'ed here).

[rvagg]

fixed in #32518 I think