(fix) deep context intelligent assistant model vision verification no cache

[JslYoon]

Summary

Fixes deep context model vision verification to not cache false results when the upstream returns a 5xx error.

Why

Previously, when the vision verification endpoint received a 5xx from the upstream model server (e.g., temporary outage, overloaded), the result was cached as supportsVision: false. This meant the model was permanently marked as non-vision-capable until cache expiry, even though the failure was transient. Users would lose deep context (image attachment) functionality for that model until the cache cleared.

What changed

• Removed the else if (status >= 500) and else branches from the vision verification handler in router.ts — the endpoint now only caches and responds for successful (vision-confirmed) results. All other outcomes (4xx, 5xx, network errors) fall through to the existing catch block, which logs but does not cache.
• Removed 3 tests from router.test.ts that validated the old caching-on-failure behavior:
  • returns false when model lacks vision
  • returns false when model is not found
  • returns 502 without caching when upstream returns 5xx

Impact

• Vision verification failures are no longer permanently cached — retries will re-check the upstream
• Eliminates false negatives for vision support caused by transient upstream errors

✔️ Checklist

• A changeset describing the change and affected packages. (more info)
• Added or Updated documentation
• Tests for new functionality and regression tests for bug fixes
• Screenshots attached (for UI changes)

[rhdh-gh-app]

Important

This PR includes changes that affect public-facing API. Please ensure you are adding/updating documentation for new features or behavior.

Changed Packages

Package Name	Package Path	Changeset Bump	Current Version
@red-hat-developer-hub/backstage-plugin-intelligent-assistant-backend	workspaces/intelligent-assistant/plugins/intelligent-assistant-backend	minor	v3.2.0

[rhdh-qodo-merge]

PR Summary by Qodo

Avoid caching model-vision verification failures for upstream 5xx responses

🐞 Bug fix ⚙️ Configuration changes 🕐 40+ Minutes

AI Description

• Add a changeset for a minor backend release documenting the cache behavior change.
• Simplify /v1/validate-model-vision handling to only cache positive (supportsVision=true) results.
• Remove explicit non-OK (4xx/5xx) response handling to prevent caching false on upstream errors.

Diagram

graph TD
  A["Client"] --> B["POST /v1/validate-model-vision"] --> C{{"LCS /v1/responses"}}
  B -- "cache true" --> D[(ModelCapabilitiesCache)]
  subgraph Legend
    direction LR
    _api["API Route"] ~~~ _ext{{"External API"}} ~~~ _cache[("Cache")]
  end

Loading

High-Level Assessment

The following are alternative approaches to this PR:

1. Keep explicit non-OK handling; only skip caching on 5xx

• ➕ Preserves existing contract: return supportsVision=false for expected 4xx cases (model lacks vision/not found)
• ➕ Avoids hanging requests when fetch returns non-OK without throwing
• ➕ Still prevents poisoning the cache during transient upstream failures
• ➖ Slightly more branching in the route handler

2. Tri-state cache (true/false/unknown) with separate TTLs

• ➕ Avoids repeated upstream calls for known negatives while preventing long-lived caching of transient failures
• ➕ Makes transient error behavior explicit (unknown)
• ➖ More complexity (data model + tests) for a small endpoint

3. Cache per-status class (2xx true, 4xx false, 5xx uncached) with metrics

• ➕ Clear policy; easier to reason about and observe via metrics/logs
• ➕ Keeps cache useful while avoiding bad entries
• ➖ Requires adding metrics/logging and potentially configuration knobs

Recommendation: If the goal is strictly “don’t cache false results when upstream returns 5xx”, the best approach is to restore explicit handling for non-OK responses: treat 5xx as a 502 without caching, and treat 4xx as supportsVision=false (optionally cached). The current diff removes all non-OK response handling, which risks leaving requests without a response when fetch returns 4xx/5xx (since fetch typically does not throw on non-2xx).

Files changed (2) +5 / -12

Bug fix (1) +0 / -12

router.tsAdjust validate-model-vision caching to only persist successful results +0/-12

Adjust validate-model-vision caching to only persist successful results

• Updates the /v1/validate-model-vision route to cache and return results only when the upstream vision test succeeds (OK). Removes prior branches that handled upstream 5xx with a 502 response and cached supportsVision=false for other non-OK responses.

workspaces/intelligent-assistant/plugins/intelligent-assistant-backend/src/service/router.ts

Other (1) +5 / -0

late-snails-battle.mdAdd changeset for backend minor release (vision cache behavior) +5/-0

Add changeset for backend minor release (vision cache behavior)

• Introduces a changeset bumping the intelligent-assistant backend plugin as a minor release. Documents that model vision verification results for upstream 500s should no longer be cached.

workspaces/intelligent-assistant/.changeset/late-snails-battle.md

[rhdh-qodo-merge]

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 📎 Requirement gaps (0)

Great, no issues found!

Qodo reviewed your code and found no material issues that require review

_{To customize comments, go to the Qodo configuration screen, or learn more in the docs.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 58.03%. Comparing base (58144ee) to head (4ea7697).
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4147      +/-   ##
==========================================
- Coverage   58.12%   58.03%   -0.09%     
==========================================
  Files        2422     2411      -11     
  Lines       96490    96287     -203     
  Branches    26854    26780      -74     
==========================================
- Hits        56081    55877     -204     
- Misses      40203    40214      +11     
+ Partials      206      196      -10     

Flag	Coverage Δ		*Carryforward flag
adoption-insights	84.55% <ø> (ø)		Carriedforward from baee808
ai-integrations	69.61% <ø> (-0.16%)	⬇️	Carriedforward from baee808
app-defaults	69.79% <ø> (ø)		Carriedforward from baee808
augment	46.67% <ø> (ø)		Carriedforward from baee808
boost	76.77% <ø> (ø)		Carriedforward from baee808
bulk-import	72.56% <ø> (ø)		Carriedforward from baee808
cost-management	13.55% <ø> (ø)		Carriedforward from baee808
dcm	60.72% <ø> (ø)		Carriedforward from baee808
extensions	56.59% <ø> (ø)		Carriedforward from baee808
global-floating-action-button	71.18% <ø> (ø)		Carriedforward from baee808
global-header	66.50% <ø> (ø)		Carriedforward from baee808
homepage	47.50% <ø> (-0.09%)	⬇️	Carriedforward from baee808
install-dynamic-plugins	59.95% <ø> (ø)		Carriedforward from baee808
intelligent-assistant	74.59% <ø> (-0.03%)	⬇️
konflux	91.98% <ø> (ø)		Carriedforward from baee808
lightspeed	69.02% <ø> (ø)		Carriedforward from baee808
mcp-integrations	83.40% <ø> (ø)		Carriedforward from baee808
orchestrator	66.87% <ø> (ø)		Carriedforward from baee808
quickstart	63.67% <ø> (-0.08%)	⬇️	Carriedforward from baee808
sandbox	79.56% <ø> (ø)		Carriedforward from baee808
scorecard	85.27% <ø> (-0.71%)	⬇️	Carriedforward from baee808
theme	88.52% <ø> (-0.25%)	⬇️	Carriedforward from baee808
translations	5.12% <ø> (ø)		Carriedforward from baee808
x2a	79.20% <ø> (ø)		Carriedforward from baee808

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

---

Continue to review full report in Codecov by Harness.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 58144ee...4ea7697. Read the comment docs.

🚀 New features to boost your workflow:

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

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud