update support-workers to java21 runtime #5792
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Size Change: 0 B Total Size: 1.78 MB ℹ️ View Unchanged
|
|
Seen on PROD (merged by @johnduffell 13 minutes and 20 seconds ago)
Sentry Release: support-client-side, support |
This was referenced Mar 13, 2024
|
in PROD you can see the improvement in cold starts, |
|
still looking good overnight |
rtyley
added a commit
to guardian/gha-scala-library-release-workflow
that referenced
this pull request
Apr 30, 2024
This allows projects to specify, with an `asdf` (https://asdf-vm.com/)-formatted `.tool-versions` file, the Java version to be used by the workflow for building the project- `gha-scala-library-release-workflow` has always used a recent LTS version Java for the build (eg Java 17), and this has sometimes been _too recent_ (guardian/atom-maker#94) for some projects at the Guardian. We _want_ projects to update to Java 21 (see guardian/support-frontend#5792, guardian/scala-steward-public-repos#67 etc), but tying dozens of projects tightly to what `gha-scala-library-release-workflow` is using will make updating that version pretty hard. If, at some later point, _some_ projects want to experiment with Java 25, we shouldn't have to force all other projects to be compatible with that first. ## How to express the version of Java required... ### Configuration proliferation is a problem... One option would have been to simply add a new input parameter to the workflow, specifying the required Java version, but that has a downside: it proliferates the number of places in a project where the desired Java version is declared - and there are many in use at the Guardian, with no well-agreed canonical source-of-truth: * GHA `ci.yml`, in the [`java-version`](https://github.com/guardian/etag-caching/blob/7ecc04981f5a42a0f2ecb10631f28da571a49836/.github/workflows/ci.yml#L22) field of `actions/setup-java` - this has been my favourite in the past, as whatever CI runs with is usually pretty close to the truth * In sbt, the `scalacOptions` of `-target`, `-release`, `-java-output-version` (currently `-release` preferred, tho' once [support](scala/scala#10654) is pervasive, `-java-output-version` is probably best) and the `javacOptions` of `-target` & `-source` - these all effectively set lower-bounds on the version of Java supported, rather than guaranteeing a minimum upper bound of Java version the way CI does. * In apps running on Amigo AMI images; the Java version baked into the AMI, usually [referenced](https://github.com/guardian/mobile-apps-api/blob/3231e6bf064163c6d0e72c8dc862678c68eb0b62/mobile-fronts/conf/riff-raff.yaml#L10) by `riff-raff.yaml` * In AWS Lambdas; the cloudformation [`Runtime`](https://docs.aws.amazon.com/AWSCloudFormation/latest/UserGuide/aws-resource-lambda-function.html#cfn-lambda-function-runtime) parameter, often set [by CDK](https://github.com/guardian/mobile-save-for-later/blob/1ac12e4c0100edb976ebae9e2a9975ad2321e26e/cdk/lib/mobile-save-for-later.ts#L44) ### ...`asdf`'s `.tool-versions` flle offers some consolidation Benefits of using `.tool-versions`: * Developers will automatically get the right Java version if they have `asdf` installed * actions/setup-java#606 has added early support for reading the version of Java used in CI by `setup-java` from the `.tool-versions` flle - unfortunately, it's of limited use to us at the moment because of actions/setup-java#615, but it's a promising start _(`setup-java` also has support for `jEnv` files, but they are not commonly used at the Guardian)_ * Format of the file is simple enough that it can be easily understood and used, even if `asdf` is not in use - **including the file doesn't _force_ developers to use `asdf`** (I know some devs have been having some problems with it, and maybe there are/will be better alternatives out there), but it clearly documents the Java version to be used. This does mean that **all library repos need to add a `.tool-versions` file** before this PR is merged. Helpful error messaging has been added with this PR to handle cases where the file is missing or invalid: #### Only Java _major_ version is guaranteed Note that although `asdf` requires a fully-specified Java version (eg `21.0.3.9.1`) in the `.tool-versions` file, currently the workflow will only match the *major* version of Java specified in the file (eg `21`), and will _always_ use the AWS Corretto distribution of Java. This is due to [limitations](actions/setup-java#615) in [`actions/setup-java`](https://github.com/actions/setup-java).
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR bumps support workers to java 21. This is supposed to give us performance improvements.
https://aws.amazon.com/blogs/compute/java-17-runtime-now-available-on-aws-lambda/
https://aws.amazon.com/about-aws/whats-new/2023/11/aws-lambda-support-java-21/
We are using scala 2.13.12 which is compatible with JRE21
https://docs.scala-lang.org/overviews/jdk-compatibility/overview.html
I have tested in CODE and looking at one lambda (create payment method) you can clearly see the performance improvement for the cold start
This is replicated for the whole state machine execution, although it is diluted by the different paths it can take through the state machine.
