We had a pipeline issue (in stage) this week due to an errant metadata generation change:
#219
#220
Fortunately when we deploy these kinds of changes SRE is generally made aware, and well-formed parts of the metadata in this case explicitly blocked the pipeline from deploying without operator approval. However, in the normal case (since this wasn't a BQ schemas error), this change might have auto-deployed to production even though the stage decoder wound up in a broken (but running) state.
We can probably either have tests to verify metadata is valid or else have the application run json schema validation checks after generating schemas, refusing to push schemas if there are issues.
We had a pipeline issue (in stage) this week due to an errant metadata generation change:
#219
#220
Fortunately when we deploy these kinds of changes SRE is generally made aware, and well-formed parts of the metadata in this case explicitly blocked the pipeline from deploying without operator approval. However, in the normal case (since this wasn't a BQ schemas error), this change might have auto-deployed to production even though the stage decoder wound up in a broken (but running) state.
We can probably either have tests to verify metadata is valid or else have the application run json schema validation checks after generating schemas, refusing to push schemas if there are issues.