fix(multinode): WARMUP_S=60 + grouping_labels + queries-e2e wave queries - #397
Merged
Merged
Conversation
…ries-e2e wave queries Three harness gaps closed (against asap arm post-#396 baseline 332/413): 1. WARMUP_S 30 → 60s in `topology.env`. The 30s warmup wasn't long enough for the agent's first 30s tumbling window to close + flush to backend before replay started. Result: 14 leading "error/200" on 4 queries per arm. 2. `deploy/mvp-multinode/configs/asap/mvp-workload.yaml` was missing `grouping_labels: [zone]` on all 8 entries. ASAPCollector PR #390 added these to the singlenode workload but the multinode equivalent was never swept. Without grouping_labels, the controller's `attributes/keep` processor isn't injected, the agent emits full-attr sketches, the backend stores per-(zone,rack,node,pod) sids, and queries return per-series instead of per-zone. 3. Replaced broken `quantile by (zone) (0.99, http_requests_total_latency_ms)` in `queries-e2e.json` with `max by (zone) (quantile_over_time(0.99, http_requests_total_latency_ms[5m]))`. Original used PromQL's instant `quantile()` aggregator which expects scalar values across series — incompatible with DDSketch state. The replacement is valid PromQL on both raw and sketch tiers: inner computes per-series p99 over 5min (or per-zone p99 directly on asap thanks to the new grouping_labels), outer `max by (zone)` is the per-zone worst-case rollup (SLO-relevant tail). `histogram_quantile` was considered but doesn't apply — the fake-exporter emits a Float64Gauge per fake-exporter/main.go:446, not a Prometheus histogram with `_bucket{le=...}` series. Results post-fix on asap arm: 6/7 wave queries now 100% success (was 4/7 with 14-error leading pattern on the other 3). Baselines b0/b1 unchanged at 100%+empty on the queries the producer emits data for. Known follow-up (asap engine limitation, not harness debt): the new `max by (zone) (quantile_over_time(...))` query still errors on asap because the engine doesn't compose outer aggregation operators on function results — even when the inner already returns per-zone (which makes the outer a no-op identity). The same query succeeds 118/118 on b0/b1 via VictoriaMetrics. Worth a separate ASAPQuery-backend PR. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Three harness gaps closed against the post-#396 multinode baseline (asap arm 332/413 success):
WARMUP_S30 → 60s intopology.env. The 30s warmup wasn't long enough for the agent's first 30s tumbling window to close + flush sketch state to backend before replay started. Result: ~14 leadingerror/200per query on 4 queries each run.grouping_labels: [zone]added to all 8 entries indeploy/mvp-multinode/configs/asap/mvp-workload.yaml. PR fix(deploy): add grouping_labels: [zone] to mvp-workload entries (B3 end-to-end) #390 added these to the singlenode equivalent but the multinode workload was never swept. Without it, the controller'sattributes/keepupstream-of-sketch processor isn't injected → agent emits full-attr sketches → backend stores per-(zone,rack,node,pod) sids → queries return per-series instead of per-zone.Replaced broken
quantile by (zone) (0.99, http_requests_total_latency_ms)inqueries-e2e.jsonwithmax by (zone) (quantile_over_time(0.99, http_requests_total_latency_ms[5m])). Original used PromQL's instantquantile()aggregator which expects scalar values across series — incompatible with DDSketch state on asap tier. Replacement is valid PromQL on both tiers: inner returns per-zone p99 directly on asap (thanks to fix 2's grouping_labels), per-series p99 on b0/b1 raw scan. Outermax by (zone)rolls per-series up to worst-case per-zone tail.histogram_quantilewas considered but doesn't apply — fake-exporter/main.go:446 emits aFloat64Gauge, not a Prometheus histogram with_bucket{le=...}series.Validation (full all-arms run post-fix)
quantile_over_time(0.99, ...[5m])quantile_over_time(0.5, ...[5m])sum by (zone) (http_requests_total)sum by (zone) (rate(...))sum_over_time(...[5m])topk(5, sum by (zone) (rate(...)))max by (zone) (quantile_over_time(...))6/7 wave queries now 100% on asap (was 4/7 with 14-error leading pattern on the other 3 in PR #396's baseline).
Follow-up (ASAPQuery-backend, not harness debt)
max by (zone) (quantile_over_time(0.99, ...))errors on asap tier — the engine doesn't compose outer aggregation operators on function results, even when the inner already returns per-zone (no-op identity). The same query succeeds 118/118 on b0/b1 via VictoriaMetrics. Worth a separate ASAPQuery-backend PR.Test plan
bash deploy/mvp-multinode/scripts/run_demo.sh all→ MVP_REPORT.md shows the table above