Move away from deprecated bors-ng

[michaelsproul]

Description

The Bors bot that we're using has been deprecated and will soon stop working: https://github.com/apps/bors.

They suggest using Github's own "merge queue" implementation instead, which seems to me like a good solution.

Steps to resolve

According to Github's docs someone with the highest level of privileges (@paulhauner) needs to require merge queues on the unstable branch: https://docs.github.com/en/repositories/configuring-branches-and-merges-in-your-repository/configuring-pull-request-merges/managing-a-merge-queue#managing-a-merge-queue

I think for the settings we want:

• Merge method: squash
• Only merge non-failing pull requests: false (so we can batch two fixes that might pass together but not individually)
• Timeout: 3 hours, per:

lighthouse/bors.toml

Line 28 in 693886b

timeout_sec = 10800

We also need to add the merge_group event to the test-suite YAML (I'll raise a PR to do this now).

• Enable merge queues
• Merge Prepare CI for merge queues #4252 using queue
• Merge Add mergify merge queue configuration file #4917 using mergify queue
• Delete bors.toml 🫡

[michaelsproul]

PR here: #4252. I reckon we should enable the merge queue and then try to merge that PR via the queue.

[michaelsproul]

I'm taking a break from this for now because of the issue where Github's merge queue requires CI to finish before queueing (more here: #4254 (comment)).

If GitHub fix this soon we'll be able to switch over. If not we might want to host Bors ourselves

I found some more information about the Bors deprecation from its creator here: https://bors.tech/newsletter/2023/05/01/tmib-76/

[michaelsproul]

I commented on the public feedback: https://github.com/orgs/community/discussions/46757#discussioncomment-5786544

[michaelsproul]

Another workaround would be to skip the long-running jobs on pull requests, then run them only on the merge queue.

I was having a poke around some other people's workflow YAML to see what they do, and it seems this is a common approach, e.g.

• determine job to calculate what to run: https://github.com/bytecodealliance/wasmtime/blob/85cbaa5220fd4028dfc5203171ebd92375949d16/.github/workflows/main.yml#L110-L120
• status job to report a unified status based on everything that did run: https://github.com/bytecodealliance/wasmtime/blob/85cbaa5220fd4028dfc5203171ebd92375949d16/.github/workflows/main.yml#L661-L713

[jimmygchen]

It was mentioned in this blog post that the bors server will be shutdown once Github merge queue is out of beta. GitHub merge queue went out of beta in July as well, so probably worth looking into this soon.

@michaelsproul suggested another option is to self host our own Bors instance.