Bazel: make git_lfs_probe.py try all available endpoints

[redsun82]

It is advisable to exclude b63bd2a from the diff view: that changes only formatting according to my current IDE settings (and the top level black configuration).

This was confirmed to help a user with a specific git remote configuration in ql.

[criemen]

LGTM! I wonder if we could somehow persist which endpoint we chose at least for the duration of a single bazel invocation, so that we don't run into timeouts over and over.
But maybe that's
a) overkill to do up-front, and this should be deferred until we observe a need for this in practice
b) quite tricky, as we'd probably need a repository rule to figure this out and store the result (plus, if we don't rerun it every time, that needs cache invalidation too).

[redsun82]

LGTM! I wonder if we could somehow persist which endpoint we chose at least for the duration of a single bazel invocation, so that we don't run into timeouts over and over. But maybe that's a) overkill to do up-front, and this should be deferred until we observe a need for this in practice b) quite tricky, as we'd probably need a repository rule to figure this out and store the result (plus, if we don't rerun it every time, that needs cache invalidation too).

Indeed. Also, consider that while we might encounter timeouts or delays multiple times in a bazel invocation, we will still encounter them just once per file: after a download succeeds, it will be in the repository cache and be fetched instantaneously from then on. And CI does not have any such problem (the first and only endpoint is good). So I don't think it's worth to complicate it further (considering it already covers a bit of a corner case).

[criemen]

Ah that's a good point indeed - where we download a lot, we only have one endpoint, so no confusion possible, and locally we ought to not download a lot at all - understood!