The second commit (127429d) changes the authentication workflow to rely on GitHub as an OpenID Connect provider (instead of using AssumeRole with static AWS credentials, which can not expose in the CI, even as GitHub secrets), and removes the AssumeRole authentication test.
The remaining commits fix issues that I discovered while working on this, and could be extracted out to separate PRs if needed:
babe657 fixes a bug when streamChanges is enabled but no static credentials were provided. We used to fail with a NullPointerException. We now fallback to the default AWS credentials provider.
1ab3a51 fixes the shutdown hooks to effectively terminate the application instead of only dumping the migration state and keeping the Spark application running forever.
b31e52f properly uninstalls the savepoints manager when the Migrator completed the migration of the initial snapshot.
8bfe80d increases the timeout of some test suites to reduce their risk of flakiness.
Alternative to #205. Fixes #113.
The first commit (4945055) is taken from #205.
The second commit (127429d) changes the authentication workflow to rely on GitHub as an OpenID Connect provider (instead of using AssumeRole with static AWS credentials, which can not expose in the CI, even as GitHub secrets), and removes the AssumeRole authentication test.
The remaining commits fix issues that I discovered while working on this, and could be extracted out to separate PRs if needed:
streamChangesis enabled but no static credentials were provided. We used to fail with aNullPointerException. We now fallback to the default AWS credentials provider.