Repository navigation
Conversation
|
Review requested:
|
nodejs-github-bot
added
the
meta
3e63215 to
4548c8d
Compare
inoway46
changed the title
inoway46
changed the title
|
As I said before, I don't think this is the right approach, IMO a solution à la #66228, where we reduce the load instead of paying for a more powerful runner, would be much preferable. I suggested in #64972 (comment) to add cache, that would be a preferable outcome |
|
Makes sense. I’ll close this. Thanks! |
Move lint-js-and-md to ubuntu-26.04-arm to reduce runtime and provide more headroom when the ESLint cache cannot be restored. Assisted-by: Codex Signed-off-by: inoway46 <inoueyuya416@gmail.com>
4548c8d to
41f27fc
Compare
|
As discussed here, timeouts still occur despite caching. I've reopened this PR with updated benchmarks, including a two-CPU ARM approximation to help assess the private-repository cost trade-off. |
|
If costs in the private repo used for security releases are still a concern, then I'd suggest checking if the workflow can check if it's running in a public / private repository and then choose the runner accordingly. It would however be simpler to use one runner in both instances. According to the assessment in the original post, the incremental cost in the private repo would be around $5 / 1000 runs. I can't judge whether that would be significant or not. |
Move
lint-js-and-mdtoubuntu-26.04-armfor faster execution and more timeout headroom.Private ARM runners have two CPUs, so we also restricted the public runner to two CPUs to approximate private execution costs.
Warm cache
ubuntu-slimCold cache
ubuntu-slimUsing the ~75% cache restoration rate, the 2-CPU approximation estimates private costs at $13.65 versus $9.02 per 1,000 runs (+51%).
Benchmarks: slim / ARM, 2 CPU ARM.
Refs: #64972