Repository navigation
An exact result is not reused for the same query with a looser accuracy target #903
Description
Workload
The same PromQL query twice on 4535c957, every 10 s: quantile_over_time(0.9, data[5m]), with Q1 exact and Q2 at ε=0.01, δ=0.001. data is a gauge with 10,000 series sampled every 15 s.
Save as wl.json:
{"query_workload": {"language": "promql", "query_batch": null, "repeating_queries": [{"query": "quantile_over_time(0.9, data[5m])", "demand": {"fixed_interval": 10000}, "requirements": {}, "predictability": {"predictable": {"known_at": null}}}, {"query": "quantile_over_time(0.9, data[5m])", "demand": {"fixed_interval": 10000}, "requirements": {"accuracy": {"max_error": 0.01, "failure_probability": 0.001}}, "predictability": {"predictable": {"known_at": null}}}]}, "data_workload": {"arrival": {"ingesting": {"rows_per_second": 666.67}}, "sources": {"data": {"series_count": 10000, "metric_type": "gauge", "sample_interval": 15000}}}}cargo run -p asap-devtools --bin stage_pipeline -- --workload wl.json --exhaustive --out out.jsonObservation
| Plan | Q1 | Q2 | Cost per second |
|---|---|---|---|
P6-m1 (selected) |
exact Aggregate |
KLL | 0.166 |
P3-m1 |
exact Aggregate |
a second, identical exact Aggregate |
0.184 |
Both plans share Scan and TimeRange. In P3-m1, the two exact Aggregate nodes are identical but are not shared. An exact answer meets any ε, so Q2 could read Q1's result.
Expected: a candidate in which Q2 reads Q1's exact aggregate. Its cost would be 0.084 (scan) + 0.020 (time range) + 0.040 (aggregate) = 0.144, cheaper than the selected 0.166.
Activity
zzylol commentedon Oct 23, 2026
Deferred until the memo search (#879, #888) lands. This sharing/reuse rule will be added as a memo transformation rather than to the legacy search.
devtools: planner_fuzz, a metamorphic fuzz tester for missed reuse an…