kll: report asap_sketchlib's allocation as the lib rows' footprint by zzylol · Pull Request #192 · ProjectASAP/sketch-bench

zzylol · 2026-10-09T21:37:13Z

Closes #191.

The kll-percall lib rows now report what asap_sketchlib::KLL allocates at construction, kll_lib_bytes::<T>(k):

the retained-item slots, compute_max_capacity(k, m) (a copy of the library's private function);
the level index, (MAX_LEVELS + 1) usizes;
the merge buffer's capacity, k items.

The size is still independent of the item count, because the library allocates once. At k = 50, 200, 800 with i64 it is 5536, 10128 and 29136 B, against 1600, 6400 and 25600 B before.

What changed

wrappers/kll/mod.rs: the capacity replica and its constants moved here from hydra_kll/mod.rs (renamed kll_lib_slots, KLL_LIB_*). kll_lib_bytes was added, and kll_lib_k reproduces the library's floor and cap on k.
wrappers/kll/sketchlib.rs: both lib rows (per-call and cdf) use kll_lib_bytes.
hydra_kll: imports the moved replica. Its footprint is unchanged: slots plus level index per cell, without the merge buffer. Hydra is out of scope here.
The oxide rows keep the nominal 4·k, because sketch_oxide's KLL grows its level vectors as it fills, so there is no fixed allocation to report. The doc comment now says so.

Tests

New lib_footprint_is_the_library_allocation. At k = 269 it gives 12296 B, the figure aqpbm-planeval derives independently for the same library. It also pins kll_lib_slots(50) = 580 and the floor below LIB_K_MIN.
cargo fmt --check, clippy --workspace --all-targets -D warnings (without rqe-optimizer, which needs libclang locally), cargo test -p sketch-bench (96 passed) and -p aqpbm-cli pass.

What this PR doesn't count

The cdf row's prepared CDF table (about 16 B per retained item), and the oxide cdf row's, built for queries.
Each hydra-kll cell's merge buffer. hydra-kll is out of scope; a comment at memory_hydra_kll notes the difference.

Follow-up (not in this PR): committed data with the old KLL memory (1600, 6400, 25600 B), to regenerate together

The optimizer cost table, rqe-optimizer/results/autosketch-vs-asap-inputs/saturation/optimizer_cost/rqe_atomic_costs.json (mem_bytes_per_instance), via study_saturation.py --phase optimizer-cost. Then rerun the AutoSketch-vs-ASAP results from eval: AutoSketch vs. ASAP on the synthetic mixed set (cost by use; frontier and SLA versions) #138; only the memory-weighted (Fargate) ones change.
rqe-optimizer/data/autosketch-eval/table.json, the KLL rows.
docs/figures/saturation/grid_cost/crossover.csv and recommendations.csv, and docs/figures/saturation/merge_grid/recommendations_with_merge.csv (the sketch_memory column).
The rqe-optimizer doc's KLL paragraph (rqe-optimizer: let mergeable finer-group deployments serve coarser RAQEs #190) will need its numbers updated once this lands.

🤖 Generated with Claude Code

https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis

The kll-percall lib rows reported a nominal 4·k·sizeof(T), which is not what asap_sketchlib::KLL allocates: init_internal boxes compute_max_capacity(k, m) slots once, plus a level index and a merge buffer of capacity k (5536 B at k = 50, not 1600 B). Report that, from the capacity replica the hydra-kll wrapper already had, now shared in the kll module. The oxide rows keep the nominal size: sketch_oxide's KLL grows as it fills. Closes #191. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis

- candidates: eligibility allows subset groupings for families that merge across groups; generation shows the roll-up RAQE sets; accuracy is read at G_r with KLL's fan-out merge count, and past the measured merge counts the guarantee applies, floored by the last measurement; pruning lists every compared cost; the MILP lives only in the cost doc; one roll-up section. - cost model: top-k output and query CPU scale with k; k_D sums parts; mergeable_across_groups listed; KLL footprint points at #191/#192; Hydra moved out. - v1 and TODO: phases include compaction and the query job; TODOs match cost by use. - rqe_optimizer_hydra.md (new): Hydra in sketch-bench, accuracy bounds per combination (CMS, CountSketch, HLL, UnivMon, KLL) checked against the paper's Theorem 2 and the library (correlated inner seeds, fixed column hash, few rows), cost model, candidates and MILP edges, sketch-bench requirements and open decisions. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis

The lib footprint is the sketch's own allocation; say that the cdf row's prepared CDF table is not counted. KLL_LIB_MIN_LEVEL is the m init_kll passes, 8, with a const assertion that LIB_K_MIN agrees, instead of being defined through it. hydra-kll's footprint notes it omits each cell's merge buffer. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis

- candidates: eligibility allows subset groupings for families that merge across groups; generation shows the roll-up RAQE sets; accuracy is read at G_r with KLL's fan-out merge count, and past the measured merge counts the guarantee applies, floored by the last measurement; pruning lists every compared cost; the MILP lives only in the cost doc; one roll-up section. - cost model: top-k output and query CPU scale with k; k_D sums parts; mergeable_across_groups listed; KLL footprint points at #191/#192; Hydra moved out. - v1 and TODO: phases include compaction and the query job; TODOs match cost by use. - rqe_optimizer_hydra.md (new): Hydra in sketch-bench, accuracy bounds per combination (CMS, CountSketch, HLL, UnivMon, KLL) checked against the paper's Theorem 2 and the library (correlated inner seeds, fixed column hash, few rows), cost model, candidates and MILP edges, sketch-bench requirements and open decisions. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis

Tests that fail without their rule: - a DDSketch roll-up sizes its accumulator at the coarse grouping; - validate_facts rejects a RAQE grouping with more groups than series; - pruning drops a heap whose larger answered k only adds output memory. Docs: the KLL footprint paragraph now that #192 is in; univmon-cardinality reads merge curves like KLL (and has no guarantee past them); the TODO's done list covers roll-ups and the pruning costs; "temporal pre-merging"; enumerate.rs names the MILP variables z and u as the docs do. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis

…QEs (#190) * rqe-optimizer: let mergeable finer-group deployments serve coarser RAQEs A deployment grouped by G_d now serves a RAQE grouped by any subset G_r when its family merges across groups (exact sum/min/max, HLL, UnivMon cardinality, KLL, DD). The read is priced at G_r: card(G_d)·L/x − card(G_r) merges, with card(G_r) accumulators, probes and output. Accuracy reads G_r's shape and items, and KLL's merge curve at the average fan-out times L/x. Fine candidates take windows from coarser RAQEs. validate_facts rejects a label set with more groups than a superset. Closes #187 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * docs: split RQE optimizer design by concern * docs: clarify RQE optimizer cost model status * docs: one cost table for every phase, with roll-ups Merge the cost-by-use CPU and memory tables into one table per phase (ingest, compaction, query job, storage) in terms of I_d and I_r, so roll-ups (G_r ⊂ G_d) are priced as the code prices them: ℓ = card(G_r)·c_qry + (I_d·n − I_r)·c_mrg, accumulators at G_r. The snapshot-AUC section now lists only how it differs (k = 1, query memory always held, per-RAQE serial latency) instead of repeating the table; the shape/size-law side-by-side gives each family's I_d, I_r and tracker; the proposed Hydra cost model is §3's table with I_d = I_r = 1, adding compaction and using the doc's symbols. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * docs: KLL's memory is sketch-bench's nominal footprint, not measured at 1e6 The cost table's KLL m is kll_footprint = 4·k·sizeof(T), independent of the item count (1600 B at k = 50 for 1e3, 1e6 and 1e8 items), so it is not "the size measured at 1e6 values" and does not grow with n. Say so, and how it compares with asap_sketchlib's preallocated capacity. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * rqe-optimizer: prune on every cost cost-by-use bills; fix stale comments Dominance pruning compared ingest, query latency, merge memory and storage. Cost by use also bills compaction (per window in every chain, per slide in CPU and byte-seconds) and query output memory, which differs between top-k configs answering different k. A direct RAQE never merges at query time, so a candidate with cheaper ingest but costlier merges could prune one that is cheaper by use. Compare those too; the new test fails without it. Also: workload_scenarios no longer says groupings never share; the exact-increase exclusion states its real (conservative) reason; merged_instance_count notes roll-ups; a circular doc comment; stale §N citations now name the docs. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * docs: consolidate the optimizer docs; add the Hydra admission design - candidates: eligibility allows subset groupings for families that merge across groups; generation shows the roll-up RAQE sets; accuracy is read at G_r with KLL's fan-out merge count, and past the measured merge counts the guarantee applies, floored by the last measurement; pruning lists every compared cost; the MILP lives only in the cost doc; one roll-up section. - cost model: top-k output and query CPU scale with k; k_D sums parts; mergeable_across_groups listed; KLL footprint points at #191/#192; Hydra moved out. - v1 and TODO: phases include compaction and the query job; TODOs match cost by use. - rqe_optimizer_hydra.md (new): Hydra in sketch-bench, accuracy bounds per combination (CMS, CountSketch, HLL, UnivMon, KLL) checked against the paper's Theorem 2 and the library (correlated inner seeds, fixed column hash, few rows), cost model, candidates and MILP edges, sketch-bench requirements and open decisions. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * rqe-optimizer: univmon-cardinality merges lossily; increase's real reason UnivMon's counters add exactly on merge, but each level's heavy-hitter heap is rebuilt from the union of the two heaps, so a key heavy only in the union is lost: a merged (or rolled-up) answer now reads merge curves, as KLL's does. exact-increase's accumulator merge joins two pieces of one counter in time, so merging two groups' counters is wrong; say so instead of "unconfirmed". Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * rqe-optimizer: check group counts only between label sets a plan uses validate_facts compared every pair of label sets in a metric's cardinality map, so an inconsistent entry no RAQE groups by rejected the whole workload. The roll-up merge count only reads RAQE groupings and the full label set, so check X ⊂ Y among those. Docs: why rate/increase, top-k and Hydra don't roll up. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * docs(hydra): hydra-kll is measured in the same Hydra rerun Every Hydra measurement predates the admission requirements, so they are all rerun; hydra-kll goes in that run with the other four variants. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * docs: move the Hydra admission design out of this PR Keep #190 to roll-ups. docs/rqe_optimizer_hydra.md and its links move to a separate PR stacked on this one; the docs here only say Hydra is future work and doesn't roll up. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * rqe-optimizer: tests for roll-up DD memory, facts and pruning; doc fixes Tests that fail without their rule: - a DDSketch roll-up sizes its accumulator at the coarse grouping; - validate_facts rejects a RAQE grouping with more groups than series; - pruning drops a heap whose larger answered k only adds output memory. Docs: the KLL footprint paragraph now that #192 is in; univmon-cardinality reads merge curves like KLL (and has no guarantee past them); the TODO's done list covers roll-ups and the pruning costs; "temporal pre-merging"; enumerate.rs names the MILP variables z and u as the docs do. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * rqe-optimizer: name every lossy non-heap sketch that needs merge curves Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: zz_y <zeyingz@umd.edu>

…n (§6.3) (#203) * polars: tag Hydra baseline subset keys by column (#74) The polars exact baselines keyed a label subset by its `;`-joined values, asap_sketchlib 0.2.2's subkey. 0.3.0's Hydra tags each value with its column (`label0:a;label1:b`), so on label columns that share a value domain the baseline merged a key1 group with the same key2 group and scored nonzero against itself. subset_key now builds the 0.3.0 key over hydra_shared's `label{i}` schema (escaping `\`, `:` and `;`), empty labels keep their column, and every lookup goes through it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * eval: multi-grouping anomaly-detection templates (http, flows) The generator gets a metric-level ordered label schema (labels with cardinality, child_of fan-out and skew; per-value data shape) and per-RQE `grouping` and `covers_share`. The 10 templates are now the "classic" set, byte-for-byte the committed tables; "all" adds templates 11, 12, 14, 15, 16, 17 on `http` and `flows`, whose streams each carry several groupings so roll-ups (and later Hydra) can share. The plan runs both sets over the grid. The runner builds each schema stream's MetricFacts from the schema (labels = schema + x; cardinality per used grouping and the full set), takes Raqe.grouping_labels from the RQE's grouping, maps the `cardinality` capability to HLL, and reports covers_share and roll-up deployment groupings in each choice. Classic output is unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * polars: write subset key tags without a temporary String; fix test comment Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * hydra: interleaved merge split; merge exactness and bound tests `--merge-split contiguous|interleaved` (default contiguous, unchanged) picks how merge runs split the stream into shards. Interleaved reorders the stream so the existing contiguous `partition` deals it round-robin by record, at the contiguous split's shard sizes, so every row's folds take it unchanged. Tests: hydra-cms, hydra-cs and hydra-hll grids folded from 4 and 7 shards, under both splits, answer every probed subpopulation exactly as the single stream; and every scored group on a fixed 20k-record stream lands inside its hydra doc §2.2 bound (CMS also never under). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * datagen: child_of columns for hierarchical labels A `child_of: <index>` column with `fan_out: f` nests under an earlier string column: its value is the parent row's value, a `.`, and a child index in [0, f) drawn from the column's own distribution. Bad indices, a fan_out that disagrees with the domain, and fields that would not mean what they say on a child are refused before drawing. Adds configs/datagen/hydra_hier.yaml (region 4 -> service 25 -> endpoint 25, status 4, user_id value) for the hydra rows, docs, and tests. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * hydra: score any target grouping; per-group error output Every hydra-* row (lib and polars, all five variants and hydra-univmon's four statistics) takes `--group-columns J,K,...` (default `0`, today's behavior) in place of the hardcoded SCORED_LABEL_COLUMN. A subpopulation probe is now positioned (`Vec<Option<String>>`, Some at the grouped columns), so `hydra_shared::query` asks any non-empty subset of the key columns and the polars baselines look up `subset_key` with that subset's mask (`group_key`, generalizing `prefix_key`). Each grouped comparator reports its per-group error in its own metric (relative error; mean rank error for hydra-kll) and summarises it as err_mean, err_p50, err_p90, err_max, groups_scored, plus schema_width, records and fanned_mass. `--per-group-out PATH` writes the rows as `group_key,n_q,error`. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * datagen: whole uniform bounds for child_of; 2- and 3-label hierarchy cuts A uniform draw is continuous, so `uniform{0.0, 2.9}` with fan_out 2 drew child indices 0, 1 and 2; child_of now requires whole uniform bounds. hydra_hier_d2.yaml and hydra_hier_d3.yaml cut hydra_hier.yaml to its first 2 and 3 labels for the Hydra study's schema-width sweep. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * eval: skewed http labels, per-RQE group coverage, fresh-checkout test Review of #196. http's labels are now Zipf-skewed (region 0.5, service 1.1, endpoint 1.1 within its service, status 2.0) and the flows DDoS burst targets a tail subnet, so covers_share selects a few groups. The generator computes each group's share (product of its labels' shares) and, per RQE, the covered groups (share >= tau, plus always the largest) and the smallest covered group's share and items per window. The runner reads a covered RQE's accuracy (ASAP and AutoSketch) at that group's items instead of the mean group's; covers_share = null is unchanged. The schema test reads an inline HLL curve fixture instead of the gitignored study curves. The synthetic plot keys runs by template set too, and both plot scripts title classic "mixed" and all "mixed + multi-grouping". README: the coverage rule, the N the runner uses, and which schema fields are descriptive. label_set.keys_per_window is commented as the universe K. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * eval: read null-coverage RQEs at their smallest group; template 17 at {service, endpoint} - covers_share = null covers every group, so the runner reads its accuracy at the smallest group (min_covered_share), like a tau RQE's smallest covered group. Classic RQEs (no covers_share) are unchanged. - Template 17 (negative control) groups by {service, endpoint}: its smallest group holds 4.5e4 items in 5m, on the curves. - The generator asserts every schema RQE's covered_min_N >= 1e3 (drops the vacuous `assert covered`); a test checks the old full-label-set template fails it. - Covered facts go through validate_facts. - README: null semantics, template 17, and the numbers' provenance (fresh `all` table, committed saturation inputs). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * datagen: the hydra_hier usage example passes --dtype Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * hydra: list every group in --per-group-out; escape keys; refuse no accuracy - Group keys escape `\`, `:` and `;` as polars_shared::subset_key does. - Groups that cannot be scored (zero true statistic, e.g. zero entropy) are written with an empty `error`, so n_q sums to the stream's records. - --per-group-out with no accuracy measurement is refused instead of writing an empty file. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * hydra: KLL CDF and UnivMon sum rows; KLL cell merge buffer - hydra-kll-cdf (lib, polars): asks the KLL grid HydraQuery::Cdf(x) at each group's own values at phi = 0.01..0.99, scored by absolute CDF error (SubpopCdfErrorGT, comparator subpop-cdf-error). The polars baseline is the exact share at or below x over the same sorted groups. - hydra-univmon-sum (lib, polars): each record inserted with count = value, so the UnivMon grid's L1 is the group's sum, scored by relative error against exact per-group sums (SubpopSumGT, comparator subpop-sum). Integer value dtypes only; f64/str and values outside [0, i32::MAX] are refused by name. - memory_hydra_kll counts each cell as kll_lib_bytes does, merge buffer of cell_k items included. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * hydra-kll: the footprint's merge buffer is the steady state; tie tests The grid clones its cells from one template, and a cloned Vec keeps no capacity, so a cell's merge buffer is empty at build and grows to about cell_k items at its first compaction. Say the footprint counts it at that bound. The CDF test now asks a group of ties (at or below) and a mixed-tie group on a grid wide enough that it shares no cell. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * hydra: record merge_split; review fixes - `merge_split` (contiguous|interleaved) is a record field next to `merge_shards` (BenchSection and MergeMetrics; schema regenerated). Merge records stamp it; a record without it split contiguously, so existing results parse, and flatten refuses two different splits in one row as it does two shard counts. - `MergeSplit` moves to aqpbm-core (it is a record type) and is parsed by clap directly instead of matching strings. - The interleaved copy of the stream is built only when a merge runs. - README states the round-robin exactly (contiguous shard sizes, so a short last shard shifts the pattern); MergeMetrics docs name the split. - hydra-cs bound test: note its margin (passes from beta ~4.8). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * study: --phase hydra on the eval's datasets; Hydra cost rows configs/datagen/hydra_http.yaml, hydra_http_latency.yaml and hydra_flows.yaml follow the §6.3 eval's http and flows schemas (labels in schema order, cardinalities, child_of, skews; user_id Zipf 0.8, latency Pareto a = 2, src_ip uniform). The flows DDoS burst is not generated: datagen draws each column independently. --phase hydra runs HYDRA_SWEEPS: the eval's variants on those datasets (every grouping, W 1024/4096/16384, N 1e5..1e7, 1/4/16 interleaved shards, 3 seeds) and every statistic on hydra_hier{_d2,_d3,}.yaml, one run per grouping with --per-group-out, into hydra_saturation.csv with the coverage columns err_max_cov_0.01/0.05 (share >= tau plus the largest group). --jobs under a memory budget, --resume, and --hydra-shard i/n to split it across nodes. --phase optimizer-cost adds Hydra rows (eval variants x W x dataset); MeasuredAt gains dataset and schema_width, the reducer reading the width off the row's scores. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * hydra-univmon-sum: the ground truth refuses negative values too The lib row inserts each value as a UnivMon count and refuses negatives; the truth (and so the polars baseline) accepted them, so a spec with negatives ran on polars and failed on lib. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * study: Hydra cost rows at 3 runs; their test; datagen path from the script Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * docs: Hydra cost rows' run count Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * study_saturation: atomic hydra table rewrite, stricter resume, schema checks --phase hydra --resume now writes the kept rows to hydra_saturation.csv.tmp and os.replace()s it over the table, and keeps a row only when it ends in a line end and every column parses (integers, numbers or "" for the error columns, non-empty text). The CSV format is unchanged, so tables written by 003eb31 resume as before. The eval-spec test also checks each label's cardinality, skew and child_of against PR #196's SCHEMAS. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * study_saturation: --hydra-variants/--hydra-ws narrow the Hydra cost rows too A hydra-univmon-cardinality grid at W=16384 merged from 16 shards outgrows a 251 GB node, and the cost phase can't resume; these flags let the cost phase leave a variant or width out, as they do the hydra phase. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * eval: hold covered RQEs to the worse of their smallest and largest group Per sketch-bench#189, KLL/DD/HLL error does not fall as N grows, so the largest covered group can be harder than the smallest. The generator also emits max_covered_share / covered_max_N (the largest group, always covered); the runner reads each covered RQE's accuracy at both the smallest and the largest covered group and keeps the worse (direction-aware), unknown if either read is. Classic RQEs and Hydra's lookup are unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * aqpbm-datagen: scale_by/scale_range; hydra_http_latency scales latency per service A f64 column with `scale_by: <label column>` and `scale_range: [lo, hi]` multiplies each row by a factor fixed per label value, log-uniform in [lo, hi] from a hash of the value and the column's seed. Validation: earlier string label column, f64 only, 0 < lo <= hi. Absent fields keep every existing spec's output byte-identical. hydra_http_latency.yaml scales latency by service in [1, 10], so groups differ in latency scale and a Hydra-KLL mixture is no longer the queried group's own CDF. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * eval: template 12 groups by (dst_subnet, proto), not (dst_subnet, dst_port) (dst_subnet, dst_port) has 1e6 groups; answering each every minute takes about 49 s on one core, which set every plan's batch latency above the largest SLA (10 s) and collapsed the multi-grouping frontier. (dst_subnet, proto) has 3e3 groups and keeps three groupings over one stream. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * rqe-optimizer: admit Hydra (HLL, UnivMon cardinality, KLL) as candidates, with measured accuracy by coverage - FamilyProperties for hydra-hll, hydra-univmon-cardinality and hydra-kll: one fixed-size grid, DeltaSet key tracker, no roll-up, and the new answers_any_subgrouping. Added to Capability::families, not to DEPLOYABLE_FAMILIES. - Candidates: on a metric with a hydra_dataset, one grid per schema Λ (each RAQE grouping and the union, where card(Λ) is known); eligible for any RAQE with ∅ ≠ G_r ⊆ Λ. The subpopulations ceiling is gone. - Accuracy: hydra_saturation.csv under the saturation dir (optional), keyed by variant, config, dataset and grouping; the column the RAQE's accuracy_covers_share selects; worst over records and seeds; lossy merges bracketed by merge_shards, unknown past the largest. - Runner: ASAP and PerQuery plan with undeployable families; RQEs carry covers_share; http/flows streams name their Hydra dataset; results report hydra_rqes for workloads Hydra may serve. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * rqe-optimizer: Hydra grids at the full schema from their dataset's rows Round-1 review of #201: - One Hydra grid per (stream, variant, config, window, slide), over the metric's full schema only (MetricFacts::hydra_dataset now carries the dataset and the schema), the width the study measures. - MeasuredAt gains optional `dataset` and `schema_width`; a Hydra row is a candidate on a metric only if measured on its dataset. - Runner: card(full schema) = product of label fan-outs, capped at the stream's series; AutoSketch also allows undeployable families (it skips Hydra itself, #159). - read_csv checks required columns and short rows; config params must be numbers. - Docs: Hydra held to the worst case, built only at the full schema. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * rqe-optimizer: validate the Hydra schema, skip empty Hydra cells, reject long CSV rows validate_facts reports a Hydra schema outside the metric's labels or without a nonzero cardinality, instead of a later panic. The Hydra accuracy fold skips empty cells (no covered group scored in that run) and is unknown only when every row left the column empty. read_csv rejects rows longer than the header too. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * eval: test fixtures carry max_covered_share and label cardinalities After rebasing onto #196's two-ended coverage read: #201's runner test RQEs need max_covered_share, and #196's coverage test schema needs the label cardinality the full-schema Hydra grid reads. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * eval: roll-up ablation, ASAP without roll-ups The runner adds the method `asap-norollup` ("ASAP (no roll-ups)"): ASAP's MILP, candidates, frontier sweep and SLA grid, with each RQE allowed only deployments at its own grouping or Hydra grids. Every result counts `rolled_up_rqes` (RQEs served by a finer non-Hydra deployment), and sanity checks asap <= asap-norollup <= perquery unbounded and at each SLA. The plot scripts draw it as a dashed violet line, add fig_rollup_ablation.png (unbounded cost with and without roll-ups, absolute, CPU and Fargate panels) and a roll-up section in both summaries; results without the method still plot. The run script passes --no-chosen. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * eval: AutoSketch vs. ASAP with Hydra candidates and a roll-up ablation Results of the §6.3 synthetic eval (ProjectASAP/ASAPQuery#777) rerun with roll-ups (#190) and their ablation, Hydra candidates measured on the eval's own data (#195–#202), the multi-grouping template set (#196), #192's KLL footprint and a re-measured cost table. 8 workloads, 0 RQEs dropped, 0 sanity violations. Roll-ups serve 40% of the multi-grouping RQEs from a finer deployment and halve ASAP's deployments, but save only 3.3–3.6% (CPU) / 4.4–4.8% (Fargate). Hydra is eligible for many RQEs but never chosen: its insert fans out to every label subset (1.9–5.1 µs per record vs. 3.85 ns for a per-group HLL). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * eval: cube template and a multi-grouping-only workload Template 16 is now p99 latency GROUP BY CUBE(region, service, endpoint), all 7 groupings, replacing {service}, {region, service}, {service, status}. A new `multigroup` template set runs the multi-grouping templates alone, so the roll-up ablation is not diluted by the classic set. Roll-ups now save 60% on multi-grouping (0.551 -> 0.221 vCPU, 26 -> 5 deployments) and 8% on mixed + multi-grouping. Classic is unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis * eval: each workload's chosen plans, gzipped Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FhcWJExZcmVqS6r6rEtjis --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

zzylol mentioned this pull request Oct 9, 2026

rqe-optimizer: let mergeable finer-group deployments serve coarser RAQEs #190

Merged

zzylol merged commit f9c4b68 into main Oct 10, 2026
2 checks passed

zzylol mentioned this pull request Oct 10, 2026

eval: commit the AutoSketch-vs-ASAP accuracy curves (#138) #194

Merged

Navigation Menu

Repository navigation

kll: report asap_sketchlib's allocation as the lib rows' footprint - #192

kll: report asap_sketchlib's allocation as the lib rows' footprint#192
zzylol merged 2 commits into
mainfrom
191-kll-lib-footprint

zzylol commented Oct 9, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

zzylol commented Oct 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

zzylol commented Oct 9, 2026 •

edited

Loading