More EC2 instance usage - #161562
Conversation
This also cleans up some legacy runner configurations that aren't currently used. We can always restore these if needed and it's distracting when trying to identify what GitHub large runners are getting used. dist-alt.sh does less (no opt-dist) so we run it on a less powerful EC2 instance.
|
|
|
@bors try jobs=dist-x86_64-linux,dist-x86_64-linux-quick,dist-x86_64-linux-alt |
This comment has been minimized.
This comment has been minimized.
More EC2 instance usage try-job: dist-x86_64-linux try-job: dist-x86_64-linux-quick try-job: dist-x86_64-linux-alt
f915b6e to
c882349
Compare
|
Warning If you are changing how CI LLVM is built or linked, make sure to bump cc @jieyouxu |
|
Nice! With this, we could reinstate doing PGO + BOLT in the @bors r+ rollup |
| free_disk: true | ||
| <<: *base-job | ||
|
|
||
| - &job-linux-4c-largedisk |
There was a problem hiding this comment.
I think we sometimes need to quickly switch to this one if space starts running out on the free 4 core runners. But I don't remember seeing it being used for some time, so let's try to remove it.
…uwer Rollup of 6 pull requests Successful merges: - #159434 (fix: suggest removing a trailing semicolon when a closure argument fails a trait bound) - #160428 (Add more context to dyn-compatibility diagnostic for receiver-less associated functions) - #161166 (add crashtests [5/N], remove unused aux files ) - #161310 (Add regression test for extern crate alias shadowed by a module of the same name) - #161509 (Add regression test for private fields lint) - #161562 (More EC2 instance usage)
Rollup merge of #161562 - Mark-Simulacrum:alt-ec2, r=Kobzol More EC2 instance usage This extends our usage of EC2 to dist-x86_64-linux-alt builders (1h23m with m8a.2xlarge, https://github.com/rust-lang/rust/actions/runs/32137297331/job/95711706485). This is moving from GHA credits to EC2 credits, so we should confirm we want that, but in dollar terms this is cheaper: at 1h55m (last auto build) on [GHA $1.32/hr](https://docs.github.com/en/billing/reference/actions-runner-pricing) = $2.53/run, vs. 1h23m on EC2 at $0.48688/hr = $0.67/run. My primary goal is to try to free up GHA credits so we can move Windows and/or macOS jobs to large runners, since I suspect us trying to host those ourselves is going to be more painful. As a drive-by change this also adds support for EC2 aarch64 machines to the bors config (but not CI config). My suspicion is that if/when we have aarch64 perf we may want faster aarch64 try builds, and in any case dist-aarch64-linux is one of our slower runners -- at 2h24m -- so it may benefit from getting a faster machine. I don't know yet how EC2 will compare but extending the bors config is cheap so I'd rather just do that now to enable easier testing. I also deleted some of the old unused runner templates, I don't see much point in keeping dead ones around. r? Kobzol
…uwer Rollup of 6 pull requests Successful merges: - rust-lang/rust#159434 (fix: suggest removing a trailing semicolon when a closure argument fails a trait bound) - rust-lang/rust#160428 (Add more context to dyn-compatibility diagnostic for receiver-less associated functions) - rust-lang/rust#161166 (add crashtests [5/N], remove unused aux files ) - rust-lang/rust#161310 (Add regression test for extern crate alias shadowed by a module of the same name) - rust-lang/rust#161509 (Add regression test for private fields lint) - rust-lang/rust#161562 (More EC2 instance usage)
Fix toml syntax in bors config bors production is logging "Could not deserialize repository config: missing comma between key-value pairs, expected `,`" for at least a week (since landing #161562, I think). I assume it would fail to start if restarted? (Reproducible here: https://play.rust-lang.org/?version=stable&mode=debug&edition=2024&gist=9e7d78b0ae002de5d73c9c9f8f518f38). I think we'll want to follow up to gate bors *merging* a PR breaking the toml syntax, I cut rust-lang/bors#829 to track that. r? @Kobzol
Fix toml syntax in bors config bors production is logging "Could not deserialize repository config: missing comma between key-value pairs, expected `,`" for at least a week (since landing #161562, I think). I assume it would fail to start if restarted? (Reproducible here: https://play.rust-lang.org/?version=stable&mode=debug&edition=2024&gist=9e7d78b0ae002de5d73c9c9f8f518f38). I think we'll want to follow up to gate bors *merging* a PR breaking the toml syntax, I cut rust-lang/bors#829 to track that. r? @Kobzol
This extends our usage of EC2 to dist-x86_64-linux-alt builders (1h23m with m8a.2xlarge, https://github.com/rust-lang/rust/actions/runs/32137297331/job/95711706485).
This is moving from GHA credits to EC2 credits, so we should confirm we want that, but in dollar terms this is cheaper: at 1h55m (last auto build) on GHA $1.32/hr = $2.53/run, vs. 1h23m on EC2 at $0.48688/hr = $0.67/run.
My primary goal is to try to free up GHA credits so we can move Windows and/or macOS jobs to large runners, since I suspect us trying to host those ourselves is going to be more painful.
As a drive-by change this also adds support for EC2 aarch64 machines to the bors config (but not CI config). My suspicion is that if/when we have aarch64 perf we may want faster aarch64 try builds, and in any case dist-aarch64-linux is one of our slower runners -- at 2h24m -- so it may benefit from getting a faster machine. I don't know yet how EC2 will compare but extending the bors config is cheap so I'd rather just do that now to enable easier testing.
I also deleted some of the old unused runner templates, I don't see much point in keeping dead ones around.
r? Kobzol