fix: add classic histogram buckets to perf insights metric

[ivanauth]

Description

Perf insights histogram was defined with Buckets: nil + NativeHistogramBucketFactor: 1.1, which produces no classic buckets - only le="+Inf". This broke histogram_quantile queries and the Perf Insights UI.

Adds explicit classic bucket boundaries matching defaults.go and updates the test to assert buckets are populated.

Testing

1. Start the stack
   docker compose up --build -d

2. Wait for healthy
   docker compose ps
   Both spicedb-1 and spicedb-2 should show (healthy).

3. Write a schema and generate traffic
   zed context set local localhost:50051 foobar --insecure
   zed schema write schema.zed --insecure
   //send some check requests
   for i in $(seq 1 20); do zed permission check document:1 view user:1 --insecure; done

4. Verify classic buckets on metrics endpoint
   curl -s http://localhost:$(docker compose port spicedb-1 9090 | cut -d: -f2)/metrics | grep
   api_shape_latency_seconds_bucket
   Should show 20 le= boundaries (0.001 through 10) plus +Inf.

5. Verify histogram_quantile works in Prometheus (localhost:9091)
   histogram_quantile(0.99, sum(rate(spicedb_perf_insights_api_shape_latency_seconds_bucket[5m])) by (le,
   api_kind))
   Should return real values (e.g. ~0.009 for CheckPermission), not NaN or 5.

6. Verify native histograms aren't broken
   Check Prometheus logs for errors:
   docker compose logs prometheus | grep -i error
   Should be clean — no histogram-related errors.

7. Cleanup
   docker compose down

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 73.57%. Comparing base (bd7cb9c) to head (72cc8c8).
⚠️ Report is 1 commits behind head on main.

❌ Your project status has failed because the head coverage (73.57%) is below the target coverage (75.00%). You can increase the head coverage or adjust the target coverage.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3027      +/-   ##
==========================================
- Coverage   73.62%   73.57%   -0.04%     
==========================================
  Files         497      497              
  Lines       59888    59888              
==========================================
- Hits        44085    44059      -26     
- Misses      12630    12648      +18     
- Partials     3173     3181       +8     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[miparnisari]

did you copy these numbers from somewhere? if so, can you leave a comment?

[ivanauth]

Yes, they are from

spicedb/pkg/cmd/server/defaults.go

Line 508 in c069f97

.001, .003, .006, .010, .018, .024, .032, .042, .056, .075, .100, .178, .316, .562, 1, 5,

. Added the comment.

[miparnisari]

Note that native histograms are enabled in the docker-compose file:

spicedb/docker-compose.yaml

Line 132 in bd7cb9c

- "--enable-feature=native-histograms"

[miparnisari]

LGTM. Tested with --enable-feature=native-histograms commented out.

[vroldanbet]

@ivanauth @miparnisari these metrics have very high cardinality and are expensive to run in Prometheus. It was already an issue for native histograms, despite our knowingly accepting it because native histograms lead to fewer time series than classic. Now we have the worst of both worlds, and it's going to lead to a fundamental increase in time-series and added pressure to prometheus.

I don't understand this change; it doesn't seem correct to me. Why would one need to add both native and classic histograms? What would be the point of native histograms then? Native Histograms do support quantile computations.

EDIT: I'm actually unsure whether both will be emitted by default - better double-check. If what this is trying to solve is to support environments that have no native-histogram support (older prom versions), I'd argue this has to be opt-in, since again, it's going to be a very expensive histogram to store.