Yicong-Huang opened a new pull request, #6205:
URL: https://github.com/apache/texera/pull/6205
### What changes were proposed in this PR?
After #6201 upgraded Hadoop to 3.4.3, `texera-web` crashes on boot with
`NoClassDefFoundError:
com/sun/jersey/core/provider/jaxb/AbstractRootElementProvider`.
Hadoop 3.4 ships Jersey 1's JSON module via the fork
`com.github.pjfanning:jersey-json` instead of `com.sun.jersey:jersey-json`,
so the
existing `ExclusionRule(organization = "com.sun.jersey")` rules no longer
catch it and
the jar lands on the runtime classpath. Its
`META-INF/services/javax.ws.rs.ext.MessageBodyReader|Writer`
entries register Jersey 1.x providers (e.g. `JSONRootElementProvider`) whose
superclasses
live in jersey-core 1.x — which *is* excluded — so Dropwizard's Jersey 2
`ServiceFinder`
dies loading them during startup.
This PR:
- adds an exclusion for `com.github.pjfanning:jersey-json` to the
`hadoop-common`
(and `hadoop-mapreduce-client-core`) dependencies in
`common/workflow-core/build.sbt`
and `amber/build.sbt`. The module only serves hadoop's own web endpoints;
the
Iceberg `HadoopCatalog`/`HadoopFileIO` and `HDFSRecordStorage` code paths
don't use it.
- removes the two jars that consequently leave the dists —
`com.github.pjfanning.jersey-json-1.22.0.jar` and its transitive
`com.sun.xml.bind.jaxb-impl-2.2.3-1.jar` — from the
`LICENSE-binary`(-java) manifests
of the four affected services (amber, file-service,
workflow-compiling-service,
computing-unit-managing-service). `jettison` stays: hadoop-common declares
it directly.
### Any related issues, documentation, discussions?
Fixes #6204. Regression introduced by #6201.
### How was this PR tested?
- Rebuilt all four affected dists (`sbt <svc>/dist`) and verified via `unzip
-l` that
`jersey-json` and `jaxb-impl` are no longer bundled in any of them (before
the fix,
all four bundled both).
- `./bin/licensing/check_binary_deps.py --ignore-transitive-version jar
--license-binary ...`
(the CI PR-mode check) passes for all four services against the freshly
built dists.
- `./bin/licensing/generate_notice_binary.py` regenerated all four
`NOTICE-binary` files
from the same dists: no drift (the removed jars carry no NOTICE entries).
- Brought the stack up from this branch with `bin/local-dev.sh`:
`file-service` and
`workflow-compiling-service` — Dropwizard services that previously bundled
the jar via
workflow-core and exhibit the same startup path — start cleanly (Jetty up,
graceful
shutdown on `down`). On unmodified `main`, `texera-web` reproducibly
crashes with the
`NoClassDefFoundError` above; with the jar excluded there is nothing left
on the
classpath for `ServiceFinder` to trip over.
### Was this PR authored or co-authored using generative AI tooling?
Generated-by: Claude Code (Claude Fable 5)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
