ci: Use Linkage Checker to test Binary Compatibility #3650
Conversation
| - uses: actions/setup-java@v4 | ||
| with: | ||
| java-version: 17 | ||
| # Use Java 11 for this as Linkage Checker is only compatible with Java 11 or below |
There was a problem hiding this comment.
Curious why Linkage Checker can not be run with Java 11+?
There was a problem hiding this comment.
Not sure yet. I suspect there may be older incompatible dependency versions, but there isn't much info given in the error message. GoogleCloudPlatform/cloud-opensource-java#2395
There was a problem hiding this comment.
This is a slight concern when we drop Java 11, maybe it's worthwhile to at least put the issue into our backlog.
| # Create two mappings of possible API names (Key: Maven Artifact ID Prefix, Value: Maven Group ID) | ||
| # for the libraries that should be tested. | ||
| # 1. These are special handwritten libraries in google-cloud-java that should be tested | ||
| declare -A monorepo_handwritten_libraries |
There was a problem hiding this comment.
We probably want to include all libraries with handwritten layers. AFAIR, java-translate has a heavy handwritten layer, java-dns as well. Not sure if there is a good way to find all of them though.
There was a problem hiding this comment.
Adding Translate. IIRC, DNS is apiary based (or wraps an Apiary library)
| popd | ||
|
|
||
| echo "Artifact List: ${artifact_list}" | ||
| # Only run Linkage Checker if the repo has any relevant artifacts to test for |
There was a problem hiding this comment.
I guess this is not a realistic scenario, it is in case someone passes in an invalid repo name?
There was a problem hiding this comment.
I meant to include a check to fail the CI if there are no artifacts found
There was a problem hiding this comment.
Is there a scenario that could cause "no artifacts found"?
There was a problem hiding this comment.
Currently, no. But adding new downstream tests for repos relies on versions.txt's accuracy. We have few hardcoded values in the script that always assume cloud/ google-cloud/ or com.google.cloud.
Adding something like auth to be tested would lead to no artifacts found as it no longer follows google-cloud prefix (rather google-auth).
I think it would be harder to figure out the cause if the error message from linkage checker is Exception in thread "main" java.lang.IllegalArgumentException: Bad artifact coordinates, rather than Unable to find any matching artifacts to test in ...
|
|
||
| # Builds a string output to `artifact_list`. It contains a comma separate list of Maven GAV coordinates. Parses | ||
| # the `versions.txt` file by searching for the matching artifact_id_prefix to get the corresponding version. | ||
| function build_artifact_list() { |
There was a problem hiding this comment.
For future enhancements: we could extract this function to a separate workflow, so that the artifact list can be passed to the binary check in a matrix, which can then be run in parallel. Similar to how we generate Apiary libraries.
There was a problem hiding this comment.
This function only builds a string of packages for the linkage checker to test again. The artifactList corresponds to modules found in a repo and every repo is already built using a matrix.
ie.. bigtable -> artifactList: google-cloud-bigtable
google-cloud-java -> artifactList: translate,resourcemanager,vertexai
There was a problem hiding this comment.
Maybe I'm misunderstanding something, I thought there is still a for loop that runs each repo one by one?
Maybe I was not clear, when I mentioned "we could extract this function to a separate workflow", I meant that "we could extract the functionality of building a list of artifact for all repos to a separate workflow".
i.e.
google-cloud-java, java-bigtable, java-bigquery, java-bigquerystorage etc.
->
google-cloud-bigtable, translate,resourcemanager,vertexai etc.
Then we put all of the artifacts to a Github matrix so that they can run in parallel.
There was a problem hiding this comment.
This job is already running in parallel and each downstream repo has it's own job in the matrix. The for loop is only building a list of source artifacts for the linkage checker to use since each repo can have multiple relevant artifacts to test
There was a problem hiding this comment.
Discussed offline that we are going to maintain a hardcode list of repo -> list of artifacts mapping. This makes the shell script much easier to maintain.
|
|
||
| # Builds a string output to `artifact_list`. It contains a comma separate list of Maven GAV coordinates. Parses | ||
| # the `versions.txt` file by searching for the matching artifact_id_prefix to get the corresponding version. | ||
| function build_artifact_list() { |
There was a problem hiding this comment.
Discussed offline that we are going to maintain a hardcode list of repo -> list of artifacts mapping. This makes the shell script much easier to maintain.
|
|
2 participants
Create a Binary Compatibility Check for Protobuf versions for all downstream libraries. Binary compatiblity is done via the linkage checker tool.