fix: warm-tier SimpleEngine answers canonical quantile_over_time + rate queries - #102
Merged
Merged
Conversation
…te queries Add `DDSketch` to `compatible_agg_types(Statistic::Quantile)` and `HLL` to `compatible_agg_types(Statistic::Cardinality)` so capability matching resolves runtime sketch-emitted aggs that the planner's canonical map (`map_statistic_to_precompute_operator`) does not emit. The MVP demo's controller plans `http_requests_total_latency_ms` as a `DDSketch` directly from `mvp-workload.yaml` and routes the delta payloads through the modified-OTLP wire format; pre-fix, `quantile_over_time(0.99, http_requests_total_latency_ms[1m])` would miss capability matching whenever the inference YAML's exact-string match didn't cover the live phi/range pair, and the warm tier returned `EngineError::CapabilityMiss`. The runtime accumulator surface (`DDSketchAccumulator::query_statistic`, `HllSketchAccumulator::query_statistic`) already supports these statistics — this PR just enumerates the types that match. Two regression tests pin the fix: - `asap_types::capability_matching::ddsketch_resolves_quantile_query_post_fix` unit-test for the matcher. - `query_engine_rust::tests/inference_yaml_pattern_coverage::canonical_mvp_demo_quantile_over_time_resolves_via_capability_matching` end-to-end through `SimpleEngine::handle_query_promql` with a DDSketch agg whose `query_config` deliberately mismatches the live request — forces the capability-matching branch. `five_sketch_canonical_statistics_in_compat_list` extended to assert the new DDSketch / HLL membership inline so future divergence fails the build. Pre-existing test failures (lib: 33 / asap_types: 1 flaky `avg_finds_sum_and_count` HashMap-iteration-order test / integration: `kll_envelope_round_trip_through_backend_adapter`) unchanged. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
4 tasks
zzylol
added a commit
that referenced
this pull request
May 8, 2026
…hapes (#108) Adds three regression tests in `asap-query-engine/src/tests/datafusion/warm_engine_replay_regression_tests.rs` that pin the warm-tier `SimpleEngine` matcher + capability dispatch for the two query shapes ProjectASAP/ASAPCollector#46 reported as 500/500 + 500/1000 `status=error` after the PR #333 demo run: * `quantile_over_time(0.99, http_requests_total_latency_ms[1m])` against a `DDSketchAccumulator` — exercises `OnlyTemporal` pattern + `Statistic::Quantile` + capability matching now that PR #102 added `DDSketch` to `compatible_agg_types(Quantile)`. * `quantile_over_time(0.5, ...)` variant of the above so the fix is not implicitly hard-coded to the 0.99 phi. * `sum by (zone) (http_requests_total)` instant — exercises the `OnlySpatial` pattern + `Statistic::Sum` against a `SumAccumulator` window. Tests use a windowed `(timestamp - 60s, timestamp)` insert (via an inline `build_engine_with_window` helper) so the store's `[query_start, query_end)` overlap filter selects them — the `create_engine_single_pop` factory's zero-duration `(timestamp, timestamp)` window misses that filter and produces a spurious `No precomputed outputs found` for `OnlySpatial` queries whose range is `[end - scrape*1000, end)`. Audit conclusion: SimpleEngine's `controller_patterns` (`engines/simple/engine.rs:213-343`) cover both query shapes verbatim, and capability matching in `asap-common/dependencies/rs/asap_types/src/capability_matching.rs:160-164` already lists `DDSketch` for `Statistic::Quantile` (PR #102). No matcher additions or fixes were needed; the demo's 500/500 + 500/1000 failures are downstream of the matcher path (likely data-flow: the agent's DDSketch processor renames `http_requests_total_latency_ms` → `_latency_ms_quantile` before backend ingest, so the unsuffixed query name finds no data; and the warm tier ingests counters as `IncreaseAccumulator`, which does not support `Statistic::Sum` for the bare-counter spatial sum). Those data-flow gaps are out of scope for this PR but the regression tests here pin the matcher contract so a future refactor that breaks the working pattern paths surfaces in unit-test rather than 25-min-demo time. Refs: ProjectASAP/ASAPCollector#46 Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
4 tasks
zzylol
added a commit
that referenced
this pull request
May 8, 2026
…PromQL queries The agent's DDSketch processor renames raw input metrics (`http_latency_ms`) to a sketched-form wire name (`http_latency_ms_quantile`) before emitting to the warm tier — by design, so warm-tier routing distinguishes raw from sketched form. But the replay client / PromQL caller still references the conceptual unsuffixed metric in `quantile_over_time(q, X[range])`, so the warm engine sees a query for `X` while its sketch store only holds `X_quantile`. Result: the lookup misses and the engine returns `status=error` to a query that is logically answerable (matchers + capability dispatch already work given the right name, per PR #108). This change picks Option 1 from the design alternatives — backend self-heals stale routing tables — because it requires no controller change and keeps the wire-side rename convention intact: * New helper `resolve_quantile_metric_alias` in `SimpleEngine` parses the query, classifies its shape, and rewrites `<metric>` → `<metric>_quantile` only when (a) shape is `Quantile` (only shapes whose data lives behind the rename), (b) the bare metric is NOT registered locally but the suffixed variant IS, and (c) the rewrite re-parses cleanly. Deployments without the rename pass through unchanged. * `handle_query_promql` calls the resolver once at entry so every downstream stage (pattern match, `QueryConfig` lookup, capability matching, `StoreQueryParams.metric`, schema label lookup) sees the same name. * `replace_metric_token` does identifier-aware substitution ([A-Za-z0-9_:] boundary check) so a hypothetical `http_latency_ms_total` elsewhere in the query is left alone. Three new regression tests in `warm_engine_replay_regression_tests` pin the contract: 1. `quantile_over_time_resolves_unsuffixed_metric_to_quantile_state` — the production failure case from ProjectASAP/ASAPCollector#46: query `quantile_over_time(0.99, http_latency_ms[1m])` against a store that only holds `http_latency_ms_quantile` returns the DDSketch quantile, not `status=error`. 2. `non_quantile_query_does_not_rewrite_metric` — `sum_over_time` against the same store does NOT rewrite (shape gating). 3. `quantile_query_without_ingest_rename_passes_through` — when the bare metric IS registered locally (deployment without rename), the resolver leaves the query untouched. Refs: ProjectASAP/ASAPCollector#46 Builds on: PR #102 (DDSketch in `compatible_agg_types(Quantile)`), PR #108 (warm-tier matcher coverage tests). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
zzylol
added a commit
that referenced
this pull request
May 8, 2026
…PromQL queries (#110) The agent's DDSketch processor renames raw input metrics (`http_latency_ms`) to a sketched-form wire name (`http_latency_ms_quantile`) before emitting to the warm tier — by design, so warm-tier routing distinguishes raw from sketched form. But the replay client / PromQL caller still references the conceptual unsuffixed metric in `quantile_over_time(q, X[range])`, so the warm engine sees a query for `X` while its sketch store only holds `X_quantile`. Result: the lookup misses and the engine returns `status=error` to a query that is logically answerable (matchers + capability dispatch already work given the right name, per PR #108). This change picks Option 1 from the design alternatives — backend self-heals stale routing tables — because it requires no controller change and keeps the wire-side rename convention intact: * New helper `resolve_quantile_metric_alias` in `SimpleEngine` parses the query, classifies its shape, and rewrites `<metric>` → `<metric>_quantile` only when (a) shape is `Quantile` (only shapes whose data lives behind the rename), (b) the bare metric is NOT registered locally but the suffixed variant IS, and (c) the rewrite re-parses cleanly. Deployments without the rename pass through unchanged. * `handle_query_promql` calls the resolver once at entry so every downstream stage (pattern match, `QueryConfig` lookup, capability matching, `StoreQueryParams.metric`, schema label lookup) sees the same name. * `replace_metric_token` does identifier-aware substitution ([A-Za-z0-9_:] boundary check) so a hypothetical `http_latency_ms_total` elsewhere in the query is left alone. Three new regression tests in `warm_engine_replay_regression_tests` pin the contract: 1. `quantile_over_time_resolves_unsuffixed_metric_to_quantile_state` — the production failure case from ProjectASAP/ASAPCollector#46: query `quantile_over_time(0.99, http_latency_ms[1m])` against a store that only holds `http_latency_ms_quantile` returns the DDSketch quantile, not `status=error`. 2. `non_quantile_query_does_not_rewrite_metric` — `sum_over_time` against the same store does NOT rewrite (shape gating). 3. `quantile_query_without_ingest_rename_passes_through` — when the bare metric IS registered locally (deployment without rename), the resolver leaves the query untouched. Refs: ProjectASAP/ASAPCollector#46 Builds on: PR #102 (DDSketch in `compatible_agg_types(Quantile)`), PR #108 (warm-tier matcher coverage tests). 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
DDSketchtocompatible_agg_types(Statistic::Quantile)andHLLtocompatible_agg_types(Statistic::Cardinality)so capability matching resolves runtime sketch-emitted aggs that the planner's canonical map (map_statistic_to_precompute_operator) doesn't emit. DDSketch / HLL are wired in via the modified-OTLP path from the agent processors and theirquery_statisticimpls already cover the corresponding statistics — this PR just enumerates them so the matcher's filter doesn't reject the configs.quantile_over_time(0.99, http_requests_total_latency_ms[1m])was hitting this gap: the controller planshttp_requests_total_latency_msas aDDSketchdirectly frommvp-workload.yaml, but when the inference YAML's exact-stringfind_query_configmatch doesn't cover the live (phi, range) pair, the engine falls into capability matching and pre-fix returnedNone→EngineError::CapabilityMiss→ 404. Same pattern unblockscount(http_requests_total{...})against an HLL-only deploy.Root cause
asap_types::capability_matching::compatible_agg_types(Statistic::Quantile)listed only[DatasketchesKLL, HydraKLL]— the planner's canonical-pick types. The five-sketch comment at the top ofcapability_matching.rsnotes that HLL / DDSketch / CountSketch route via modified-OTLP and aren't in the canonical map, but the runtime accumulator surface still resolves them, so they need to appear in the compat list. KLL membership was tested; DDSketch / HLL membership wasn't, so the gap shipped.Fix shape
Option (a) per the task plan: extend
compatible_agg_typesto enumerate the runtime sketch types whose accumulators answer the statistic natively. No router-level fall-through-to-archive auto-retry.Test plan
cargo test --release -p asap_types --lib(pre-fix flakeavg_finds_sum_and_countunchanged — HashMap iteration ordering, not in our blast radius)cargo test --release -p query_engine_rust --test inference_yaml_pattern_coverage— 11 passed including the newcanonical_mvp_demo_quantile_over_time_resolves_via_capability_matchingcargo build --release— cleancargo test --release -p query_engine_rust --lib— 879/33 (same asorigin/mainbaseline; the 33 pre-existing failures are unchanged)Files
asap-common/dependencies/rs/asap_types/src/capability_matching.rs(+87/-3): compat list expansion + unit regression test + extended five-sketch agreement testasap-query-engine/tests/inference_yaml_pattern_coverage.rs(+57): integration regression throughSimpleEngine::handle_query_promql🤖 Generated with Claude Code