[PG1204]

What changes were proposed in this PR?

/api/huggingface/media-proxy checked only the URL it was given, then let Unirest follow redirects on its own. So an allowlisted host could redirect the server anywhere, including internal addresses (#9103). This applies the fix #7013 made on the Python side:

• Redirects are followed by hand. The proxy now has its own Unirest instance with followRedirects(false). The global config used by the Hub endpoints is unchanged. Each redirect comes back to the resource, and the next hop is checked before it is requested, up to 5 hops. A relative Location is resolved against the current URL. A missing or unparsable Location, or a sixth redirect, returns 502.
• Every hop gets the same checks: https only (the scheme is compared case-insensitively, so HTTPS:// now works), an allowlisted host, and every address the host resolves to must be public. The address check rejects private, loopback, link-local (including 169.254.169.254), shared (100.64.0.0/10), unique-local, multicast, reserved and documentation ranges. IPv4-mapped, IPv4-compatible and NAT64 IPv6 addresses are judged by the IPv4 address inside them.
• hf.co is added to the allowlist. Hub file URLs (https://huggingface.co/<repo>/resolve/main/<file>) redirect to us.aws.cdn.hf.co. Without hf.co, re-checking each hop would break the most common Hugging Face media URL.

Behaviour changes worth knowing:

• http:// media URLs are now refused (400), and the result panel shows such a value as text instead of media. http://huggingface.co and http://replicate.delivery already redirect to https, so no Hugging Face URL is lost.
• As in fix(workflow-operator): re-validate redirect hops in HF remote-URL fetches to close an SSRF bypass #7013, the address check runs before the request, and the HTTP client resolves the host again to connect. That narrows DNS rebinding but does not close it.
• Not changed here: for a non-redirect error, the upstream status is still passed back to the caller. HuggingFace media proxy: host allowlist can be bypassed via HTTP redirects #9103 mentions this as related.

Any related issues?

Closes #9103.

How was this PR tested?

30 new tests in HuggingFaceModelResourceSpec, 114 in total, all passing. No existing test was changed.

• Red-green: the tests were written first, against a version that only added the test seams. 26 failed, each for the expected reason. For example, the media client followed a real 302 instead of returning it, and a redirect to an off-allowlist host came back as 302 instead of 403. With the fix, all pass. Reverting the fix makes exactly those 26 fail again.
• The root cause is pinned with the real HTTP client against a loopback server: the media client must hand back a 302 rather than follow it. The other proxy tests use the suite's existing stub client plus a fake resolver, so nothing leaves the machine.
• Mutation check: switching off each defence in turn (no auto-follow, per-hop check, hf.co, embedded-IPv4 handling, the empty-resolution guard, the hop limit, https-only) is caught by its own tests.
• Live check against the real network: a Hub file URL follows its redirect to us.aws.cdn.hf.co and returns the PNG. The same URL over http:// returns 400, an off-allowlist host 403, and an unresolvable allowlisted name 502.
• scalafmtCheck and scalafix --check are clean for both main and test.

Was this PR authored or co-authored using generative AI tooling?

Yes, this PR was co-authored with Claude in compliance with ASF policy.

[github-actions]

Backport auto-label report

This fix: PR was checked against each actively-supported release branch. A release/* label nominates a backport target; the branch's release manager approving this PR is what sends the fix there. The required Backport Approvals check stays red until every label below is approved, so each manager either approves or removes their own label — which is why the labels left on a merged PR are exactly the branches it reached.

Release branch	Analysis
✅ release/v1.3	Change detected on this branch — label added; this fix is queued to backport here. @mengw15 decides: approving sends the fix here, removing this label declines it. The merge waits on one or the other. Review requested.
✅ release/v1.2	Change detected on this branch — label added; this fix is queued to backport here. @xuang7 decides: approving sends the fix here, removing this label declines it. The merge waits on one or the other. Review requested.

Auto-label run.

[github-actions]

Automated Reviewer Suggestions

Based on the git blame history of the changed files, we recommend the following reviewers:

• Contributors with relevant context: @aglinxinyuan
  You can notify them by mentioning @aglinxinyuan in a comment.

[github-actions]

⚠️ Benchmark changes need a look

🟢 4 better · 🔴 4 worse · ⚪ 7 noise (<±5%) · 0 without baseline

Compared against main 98b875f benchmarked on this same runner, so the delta is largely free of cross-runner hardware noise. The "7d avg" column still reflects the gh-pages dashboard. Treat <±5% as noise unless repeated.

Dashboard · Run

	config	throughput	MB/s	latency	max Δ latest / 7d
🔴	bs=10 sw=10 sl=64	356	0.217	27,667/33,008/33,008 us	🔴 +23.3% / 🔴 +207.7%
🔴	bs=100 sw=10 sl=64	821	0.501	121,346/133,985/133,985 us	🟢 -19.0% / 🔴 +79.0%
⚪	bs=1000 sw=10 sl=64	964	0.588	1,043,052/1,087,416/1,087,416 us	⚪ within ±5% / 🔴 +55.8%

Baseline details

Latest main 98b875f from same runner

config	metric	PR	latest main	7d avg	Δ latest	Δ 7d
bs=10 sw=10 sl=64	throughput	356 tuples/sec	400 tuples/sec	1,119 tuples/sec	-11.0%	-68.2%
bs=10 sw=10 sl=64	MB/s	0.217 MB/s	0.244 MB/s	0.683 MB/s	-11.1%	-68.2%
bs=10 sw=10 sl=64	p50	27,667 us	22,438 us	8,992 us	+23.3%	+207.7%
bs=10 sw=10 sl=64	p95	33,008 us	38,253 us	11,255 us	-13.7%	+193.3%
bs=10 sw=10 sl=64	p99	33,008 us	38,253 us	13,808 us	-13.7%	+139.1%
bs=100 sw=10 sl=64	throughput	821 tuples/sec	846 tuples/sec	1,518 tuples/sec	-3.0%	-45.9%
bs=100 sw=10 sl=64	MB/s	0.501 MB/s	0.517 MB/s	0.927 MB/s	-3.1%	-45.9%
bs=100 sw=10 sl=64	p50	121,346 us	114,694 us	68,504 us	+5.8%	+77.1%
bs=100 sw=10 sl=64	p95	133,985 us	165,363 us	74,866 us	-19.0%	+79.0%
bs=100 sw=10 sl=64	p99	133,985 us	165,363 us	89,939 us	-19.0%	+49.0%
bs=1000 sw=10 sl=64	throughput	964 tuples/sec	974 tuples/sec	1,569 tuples/sec	-1.0%	-38.5%
bs=1000 sw=10 sl=64	MB/s	0.588 MB/s	0.595 MB/s	0.957 MB/s	-1.2%	-38.6%
bs=1000 sw=10 sl=64	p50	1,043,052 us	1,023,112 us	669,501 us	+1.9%	+55.8%
bs=1000 sw=10 sl=64	p95	1,087,416 us	1,090,151 us	706,487 us	-0.3%	+53.9%
bs=1000 sw=10 sl=64	p99	1,087,416 us	1,090,151 us	737,560 us	-0.3%	+47.4%

Raw CSV

config_idx,batch_size,schema_width,string_len,num_batches,total_ms,total_tuples,total_bytes,tuples_per_sec,mb_per_sec,lat_p50_us,lat_p95_us,lat_p99_us
0,10,10,64,20,561.97,200,128000,356,0.217,27667.16,33007.92,33007.92
1,100,10,64,20,2434.85,2000,1280000,821,0.501,121346.36,133984.79,133984.79
2,1000,10,64,20,20745.27,20000,12800000,964,0.588,1043051.62,1087415.52,1087415.52

[codecov-commenter]

Codecov Report

❌ Patch coverage is 90.80460% with 8 lines in your changes missing coverage. Please review.
✅ Project coverage is 91.80%. Comparing base (2407d79) to head (862675b).
⚠️ Report is 30 commits behind head on main.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
...texera/web/resource/HuggingFaceModelResource.scala	90.80%	0 Missing and 8 partials ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##               main    #9134      +/-   ##
============================================
- Coverage     92.33%   91.80%   -0.54%     
- Complexity     5045     5077      +32     
============================================
  Files          1261     1264       +3     
  Lines         54120    54772     +652     
  Branches       6768     6879     +111     
============================================
+ Hits          49974    50282     +308     
- Misses         2486     2785     +299     
- Partials       1660     1705      +45     

Flag	Coverage Δ		*Carryforward flag
access-control-service	77.38% <ø> (ø)		Carriedforward from 2407d79
agent-service	99.16% <ø> (ø)		Carriedforward from 2407d79
amber	85.97% <90.80%> (-1.34%)	⬇️
computing-unit-managing-service	60.48% <ø> (ø)		Carriedforward from 2407d79
config-service	87.37% <ø> (ø)		Carriedforward from 2407d79
file-service	82.00% <ø> (ø)		Carriedforward from 2407d79
frontend	96.56% <ø> (ø)		Carriedforward from 2407d79
notebook-migration-service	83.73% <ø> (ø)		Carriedforward from 2407d79
pyamber	98.57% <ø> (ø)		Carriedforward from 2407d79
workflow-compiling-service	72.43% <ø> (ø)		Carriedforward from 2407d79

*This pull request uses carry forward flags. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.