My vote is: -1 (non-binding)

Verification results:

- PASS: Checksums and PGP signatures are valid
- PASS: LICENSE and NOTICE files are correct
- PASS: All files have ASF license headers where appropriate (Apache RAT: 0
unknown licenses)
- PASS: Source tarball matches the Git tag
- PASS: Docker Compose deploys successfully, but with the functional issues
described below

Steps 1–4 all pass cleanly. My -1 is for the Docker Compose deployment.

I can reproduce the Python UDF operator error already reported in
https://github.com/apache/texera/issues/5963. While investigating, I found
the root cause and a second instance of the same problem.

Root cause: the deployment's nginx.conf has drifted out of sync with
frontend/proxy.config.json. Several routes that the frontend relies on are
missing from nginx.conf, so those requests fall through to the catch-all
route, dashboard-service, and return 404.

Confirmed broken in the RC2 Docker images (
docker.io/apache/...:1.2.0-incubating-rc2):

1. GET /pve/pves?cuid=... -> 404
   This is the issue reported in #5963. The frontend calls /pve/*, but the
backend serves it as /api/pve/* on the runtime coordinator. nginx.conf is
missing the /pve -> /api/pve rewrite that exists in proxy.config.json. This
happens every time a Python UDF operator is selected.

2. GET /api/access/computing-unit/{list,owner,...}/<id> -> 404
   This affects the "Share access" feature for computing units. The
endpoint exists on workflow-computing-unit-managing-service:8888 and
returns 401 when hit directly. However, nginx.conf has no
/api/access/computing-unit route, so the request is routed to
dashboard-service and returns 404.

Suggested fix: add these blocks to nginx.conf before the location /
fallback:

location /pve/ {
    rewrite ^/pve/(.*)$ /api/pve/$1 break;
    proxy_pass http://workflow-runtime-coordinator-service:8085;
    proxy_set_header Host $host;
    proxy_set_header X-Real-IP $remote_addr;
}

location /api/access/computing-unit/ {
    proxy_pass http://workflow-computing-unit-managing-service:8888;
    proxy_set_header Host $host;
    proxy_set_header X-Real-IP $remote_addr;
}

I verified both fixes live on the RC2 stack: /pve/pves went from 404 to
200, and /api/access/computing-unit/owner/1 went from 404 to reaching the
intended service.

Since #5963 is not isolated, I suggest auditing nginx.conf against
frontend/proxy.config.json for any other missing routes before cutting the
next release candidate.

Best regards,
Kary

Reply via email to