x/build: adding linux-riscv64-unmatched builder

[mengzhuo]

According to #41956 and @cagedmantis suggestion.

I’m requesting builder linux-riscv64-unmatched

[cagedmantis]

/cc @golang/release

[gopherbot]

Change https://golang.org/cl/329489 mentions this issue: dashboard: add linux-riscv64-unmatched

[4a6f656c]

It would be good to understand what the plan is for linux/riscv64 builders moving forward - currently there are two already (although the linux-riscv64-unleashed builder was absent for a long time and appears to have disappeared again). There seems to be limited value in running multiple builders for the same platform, if they're doing the same work. Obviously the unmatched hardware will be slightly faster CPU-wise and a lot faster I/O-wise - if this hardware is available perhaps we should run this and turn off the existing two once stable?

(the other option would be to look at sharding the workloads, but that has its own sets of challenges)

[dmitshur]

It's hard to have a concrete plan, since these builders are reverse and depend on people running them. Having multiple builders for the same platform can help if one goes missing, another can still provide coverage. Having them share workload would likely be better, but we don't have good support for that.

if this hardware is available perhaps we should run this and turn off the existing two once stable?

That seems reasonable to me.

Also CC @bradfitz who's the owner of host-linux-riscv64-unleashed.

[4a6f656c]

It's hard to have a concrete plan, since these builders are reverse and depend on people running them. Having multiple builders for the same platform can help if one goes missing, another can still provide coverage. Having them share workload would likely be better, but we don't have good support for that.

Running two might make sense, but I certainly doubt three is beneficial - even with one, if that disappears for a period of time, we can readily get a replacement online (especially now that the hardware is more readily available).

if this hardware is available perhaps we should run this and turn off the existing two once stable?

That seems reasonable to me.

Sounds like a plan.

Also CC @bradfitz who's the owner of host-linux-riscv64-unleashed.

[dmitshur]

If this offer still stands @mengzhuo, I think we can proceed. Has a builder key already been provided, or does that need to happen? I can review CL 329489 this week.

[mengzhuo]

@dmitshur @golang/release I'd happy to proceed or I can add this builder to @4a6f656c 's linux-riscv64-jsing's pool

[dmitshur]

@mengzhuo Excellent, let's proceed then, thank you again. CL 329489 is reviewed, can you address the remaining comment on it, and submit it? After that, I'll redeploy coordinator, which will allow the reverse builder to connect (I understand you have a builder key, but if not, I can provide it via your Gerrit email if you ask).

[mengzhuo]

@dmitshur CL 329489 is broken and needs fixed, I will updated it today.
I don't have key for linux-riscv64-unmatched please send the key to me, thanks.

[dmitshur]

@mengzhuo A key's been provided via email. Patch set 2 of CL 329489 looks good, please submit it when you're ready, and after the following cmd/coordinator deploy the reverse builder will be able to connect.

[dmitshur]

cmd/coordinator has been redeployed, so the reverse builder should be able to connect. It’s visible at https://farmer.golang.org now:

host-linux-riscv64-unmatched: 0/0 (1 missing)

[mengzhuo]

linux-riscv64-unmatched is up and running now. Thanks @dmitshur !