feat(asap-tools): render sketch-bench e2e queries as ClickHouse SQL by zzylol · Pull Request #844 · ProjectASAP/ASAPQuery

zzylol · 2026-10-13T01:50:18Z

Builds on #837 (PromQL → ClickHouse SQL for generated2, merged); now based on main.

Summary

Renders a sketch-bench #777 workload table's e2e.queries as ClickHouse SQL, for the ClickHouse baseline of the end-to-end runs (#824). The new datasets/fake_metric/sketch_bench_workloads.py sits next to #837's generate_workloads.py and shares its extrapolatedRate (split out as rate_from_samples, whose generated text is unchanged).

Classic templates (PromQL): parsed into a Query record, adding the shapes feat(asap-tools): run generated2 PromQL workloads on ClickHouse #837 lacks: topk(k, sum by (x) (...)), sum by (x) (rate(...)), quantile by (x) (q, instant), the ratio of two quantile_over_time, and the per-series forms.
- Windows follow Prometheus 2.53, which the bare-metal experiments install (v2.53.2): range selectors are closed, [T−d, T], and instant vectors take each series' latest sample in [T−1s, T]. Measured on CloudLab against a real Prometheus 2.53: its [5s] sums match ClickHouse's ts >= T−5s AND ts <= T exactly, while (T−5s, T] misses the boundary samples. The paced feeder's timestamps and aligned eval times always put samples on that boundary.
- Note for feat(asap-tools): run generated2 PromQL workloads on ClickHouse #837: generated2 still reads (T−d, T] (left-open is Prometheus 3.0). With 1 s scrapes on whole seconds, a [1m] window reads 60 samples where Prometheus 2.53 reads 61. This PR leaves it unchanged; it's your call there.
Multi-grouping templates (structured entries, sketch-bench#209): uniqExact(value) for cardinality and quantileExactInclusive(q)(value) pooled over every sample of a group. Windows are ASAPQuery's left-open (E−W, E], since no Prometheus is involved and ASAPQuery's SQL frontend and precompute windows are (start, end].
- Also writes <table>/asapquery.sql for the ASAPQuery arm: each entry's sql from the table as written (sketch-bench#209 renders the ASAPQuery dialect once, QUANTILE(q, col), COUNT(DISTINCT col), BETWEEN DATEADD(s, -W, NOW()) AND NOW()), with only NOW() replaced by the eval time floored to T. All 936 distinct queries of the 12 tables parse in ASAPQuery's SQL frontend to (⌊E⌋−W, W).
Eval time: {{ end_time }} is floored to a multiple of T in the SQL, matching the aligned query client (feat(asap-tools): run sketch-bench workload tables end to end with the paced feeder #841).
Output: one experiment_run_clickhouse.py config per table, one query group per interval_secs (repetition = T), plus init.sql with the tables (gauge/counter: ts, labels, value; records: ts, labels, value fields). Unknown shapes, capabilities or labels raise.

Testing

Python reference: a separate implementation of the PromQL 2.53 semantics, and of the pooled multigroup semantics, over small datasets.
- Classic: 6 series at about 4 samples/s on ms timestamps, with a late-starting counter, a series that stops and a 3 s gap.
- Multigroup: a skewed http records table with endpoint a child of service.
Coverage: 18 classic shapes and template 16's full CUBE plus 7 cardinality groupings, each at several eval times, all run in one clickhouse-local process. Every SQL result matches the reference within 1e-9, including empty windows and every extrapolatedRate branch.
Mutation checks: a left-open window on classic, or a closed one on multigroup, makes 84–252 subtests fail.
Real tables: every statement for sketch-bench's classic (108), multigroup (104) and all-metrics8 (1696) tables executes in clickhouse-local.
ASAPQuery dialect: all 52 multigroup queries, rendered at aligned and unaligned end times, parse with ASAPQuery's SQLPatternParser/SQLPatternMatcher to (start, duration) = (⌊E⌋−W, W).
CLICKHOUSE_BINARY=<path> python3 -m pytest -q tests/test_sketch_bench_workloads.py passes (the ClickHouse tests are skipped without a binary). The experiments suite passes. black/flake8 clean.

🤖 Generated with Claude Code

https://claude.ai/code/session_01AnJ7LocWxguh3q4QWkCYae

…nerated2 workloads Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…e SQL configs Adds a client_options.simulated_start_time so templated SQL can be evaluated against bulk-loaded data with fixed timestamps. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

… configs sketch_bench_workloads.py reads a sketch-bench table's workloads[0].e2e block and writes, per table, the metric tables' init SQL, one SQL file per query interval and a baseline experiment_run_clickhouse.py config whose query groups repeat every interval. Queries keep generate_workloads.py's Prometheus 2.53 semantics (range vectors read (T - d, T], extrapolatedRate for rate/increase, quantileExactInclusive) and add the classic template shapes: topk over sum by, sum/count/quantile by over instant vectors (latest sample per series in (T - 1s, T]) and rates, and the ratio of two quantile_over_time. Eval time is {{ end_time }} floored to the interval. Unsupported shapes and structured (non-PromQL) entries raise. generate_workloads.extrapolated_rate now builds on rate_from_samples, which the new module reuses; its SQL output is unchanged. The test compares every generated statement in clickhouse-local against a Python implementation of the Prometheus semantics at several eval times, covering empty windows and every extrapolation branch. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AnJ7LocWxguh3q4QWkCYae

… SQL Experiments run Prometheus 2.53.2 (installation/prometheus/install.sh), whose range selectors and instant lookback are closed, [T - d, T]; left-open windows arrived in 3.0. Aligned eval times put a sample on the left boundary, so the SQL now reads ts >= T - d. The Python reference selects [T - d, T] as well. generate_workloads.py's generated2 SQL is unchanged; only the shared rate_from_samples docstring no longer names a window shape. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AnJ7LocWxguh3q4QWkCYae

Structured e2e entries {table, value, capability, grouping, quantile, window_secs, interval_secs} become pooled queries over a records table. Cardinality is uniqExact of the value field per group, and quantile is quantileExactInclusive over all of a group's samples. Both read the same closed [T - W, T] window, with T floored to the interval as the PromQL queries are. init.sql now creates records tables, with one Float64 column per value field. Each table's structured queries are also written in ASAPQuery's SQL frontend dialect (COUNT(DISTINCT col), QUANTILE(q, col), BETWEEN DATEADD(s, -W, NOW()) AND NOW()) to asapquery.sql. The clickhouse-local test adds a skewed records dataset in which endpoint is a child of service. It checks template 16's CUBE over (region, service, endpoint) and several cardinality groupings against a Python reference. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AnJ7LocWxguh3q4QWkCYae

Multigroup (structured) queries involve no Prometheus, so they now read ASAPQuery's left-open window (E - W, E] instead of the closed [E - W, E] that classic PromQL keeps to match Prometheus 2.53. ASAPQuery's SQL frontend and precompute windows are (start, end], so ClickHouse and ASAPQuery compute over the same samples. Query gets a `left_open` flag that structured entries set; the raw SQL uses `ts > E - W`. The ASAPQuery-dialect file ends its window at the floored eval time instead of NOW(). The frontend cannot floor a literal, so the client's {{ end_datetime }} is moved back by {{ end_time % T }} inside DATEADD: ts BETWEEN DATEADD(s, -{{ end_time % T + W }}, '{{ end_datetime }}') AND DATEADD(s, -{{ end_time % T }}, '{{ end_datetime }}') which the frontend parses to (start, duration) = (E - W, W). The Python reference for multigroup now pools records in (t - w, t]. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AnJ7LocWxguh3q4QWkCYae

…se configs The SQL floors {{ end_time }} to T, but the current query client sleeps T after each repetition's queries finish, so its eval times drift by the query latency and the arms answer different windows. Each query group now sets `align_to_interval: true` and `query_time_offset: 10` (PR #841's default). With #841's client, repetition k is evaluated at floor((start - O) / T) * T + k * T and sent O seconds later. The current client ignores `align_to_interval` (unknown client options become unused attributes) and shifts its eval time back by the offset. The module docstring records the drift until #841's client lands. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AnJ7LocWxguh3q4QWkCYae

…w SQL ClickHouse 26.10 turns `ts > <integer>` on a DateTime64 key column into `ts in [<integer> + 1, +Inf)` for index pruning, so the left-open multigroup SQL dropped the samples in the window's first second from the result (A6: raw vs MV relative differences of 1e-6 to 2.6e-3 on 49 of 52 queries). The raw SQL now compares ts with toDateTime64(<bound>, 3) on both ends, for classic and multigroup queries. A regression test runs a left-open and a closed query in clickhouse-local over a table with 8-row granules and samples every 7 ms around the window start. Without the cast it reproduces the loss (8435 of 8571 rows). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AnJ7LocWxguh3q4QWkCYae

Brings in #837 (this branch's base, squash-merged) and #828. The one conflict, generate_workloads.py (add/add), keeps this branch's version: main's copy is #837's original commit, and this branch later moved extrapolatedRate into rate_from_samples and closed its windows. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AnJ7LocWxguh3q4QWkCYae

sketch-bench's exporter now writes each structured e2e entry's ASAPQuery SQL (`sql`, e.g. `... WHERE ts BETWEEN DATEADD(s, -300, NOW()) AND NOW() ...`), so it is the single ASAPQuery SQL renderer. asapquery.sql now takes that SQL as written and only replaces NOW() with the floored eval time; this module's own dialect renderer is gone. The ClickHouse raw SQL still comes from the structured fields. The frontend's DATEADD takes only NOW(), a literal or a CAST as its base, so NOW() becomes a datetime literal floored in the client's Jinja template: {{ end_datetime }} (UTC) with its seconds floored for T dividing 60, or its minutes for 60 | T | 3600. Other intervals raise. All 936 distinct structured queries in sketch-bench cd51be9's tables (T = 60 s and 300 s), rendered through the client's QueryTemplate at three end times, parse in ASAPQuery's SQL frontend to (start, duration) = (floor(E) - W, W) and pass its matcher. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AnJ7LocWxguh3q4QWkCYae

milindsrivastava1997 · 2026-10-14T02:58:02Z

Code review

Note: eval-time alignment depends on #841; ClickHouse vs PromQL comparisons aren't valid until it lands. The rate_from_samples step order matches Prometheus 2.53.2 and the UTC flooring is correct.

Medium — sketch_bench_workloads.py:252 (render): empty grouping renders invalid SQL. grouping: [] produces SELECT , quantileExactInclusive(0.5)(latency) AS value FROM http WHERE ... GROUP BY . That's the CUBE's empty set, which query_from_entry accepts; it fails only at query time. The test's full CUBE skips the empty set. Support it (no GROUP BY) or raise. (Also affects feat(asap-tools): exact materialized-view arm for sketch-bench e2e queries #845's view DDL.)
Medium — :71 (INSTANT_WINDOW_SECONDS = SCRAPE_INTERVAL_SECONDS): instant vectors use a 1 s lookback, but the docstring claims Prometheus 2.53 semantics, whose default --query.lookback-delta is 5m (nothing in the repo overrides it). A series missing one sample vanishes from the ClickHouse output while Prometheus returns its last value. The test reference shares the 1 s assumption. Document as a deliberate approximation or widen.
Low-medium — :402 + describe() :225: the -- comment uses raw entry["promql"]; a multi-line PromQL only gets its first line commented, the rest is bare text before SELECT → syntax error after experiment_run_clickhouse.py:236 splits on ;. Comment the whitespace-normalized text.
Low — :121-133 / raw_agg: PromQL on a records metric is accepted, but the SQL reads value, which records tables lack → "Missing columns: value" at runtime. Raise at generation time.
Low — :345 (asapquery_sql): .replace("NOW()", ...) is case-sensitive and silently no-ops; also the literal is UTC while ASAPQuery reads naive literals as server-local. Assert a substitution happened and document the UTC requirement.
Low — lines 33-40, 66, 425: comments reference "PR feat(asap-tools): run sketch-bench workload tables end to end with the paced feeder #841" / narrate history, against AGENTS.md comment conventions.
Follow-up (not this diff): generate_workloads.py:71 (generated2) still uses ts > {END} - {window} against a bare integer — the pattern this PR's own comment says ClickHouse 26.10 prunes as ts >= int + 1, dropping samples. Likely the same bug; worth an issue.

Tests: strong (clickhouse-local reference, mutation checks, DateTime64 regression). Gaps: empty grouping, multi-line PromQL, PromQL on a records metric.

- Render an empty structured grouping as the grand total, no GROUP BY. - Comment PromQL on one line so multi-line PromQL stays a comment. - Reject PromQL over a records metric, and ASAPQuery sql without NOW(). - Document the 1 s instant lookback and the UTC server requirement. - Drop PR references from comments. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

milindsrivastava1997 · 2026-10-14T14:23:08Z

Addressed in 6418827:

Empty grouping: fixed. An empty grouping now renders the grand total with no GROUP BY (HAVING count() > 0, so an empty window returns no rows, the same as the grouped queries). The test's CUBE now includes the empty set and is checked against the reference in clickhouse-local; reverting the fix makes it fail. feat(asap-tools): exact materialized-view arm for sketch-bench e2e queries #845's view DDL needs the same change there.
Instant lookback: documented, not widened. The feeder writes every series at every scrape, so a 1 s lookback agrees with Prometheus unless a sample is missing. The docstring now says this is an approximation and describes what diverges.
Multi-line PromQL: fixed. describe collapses whitespace, so the comment is always one line. Added a test.
PromQL on a records metric: fixed. It now raises at generation time in query_from_entry. Added a test.
NOW(): fixed. It raises if the sql contains no NOW(). The docstring now says the ASAPQuery server must run in UTC.
PR references and history in comments: removed.
generated2 ts > {END} - w: not the same bug. generated2's table is ts DateTime (datasets/fake_metric/init.sql), not DateTime64, so ts > int is exact. The pruning rewrite only applies to the DateTime64 key here. No issue filed. The left-open vs closed window question for generated2 (PR body) is still open.

🤖 Generated with Claude Code

milindsrivastava1997

Automated code review (Claude Code). Findings are unverified candidates; ordered by severity — the first two affect correctness of baseline numbers on current main.

milindsrivastava1997 · 2026-10-14T15:29:08Z

+            f"-- {describe(entry)}\n{raw_sql(q, labels)}"
+        )
+        if "promql" not in entry:
+            asapquery.append(f"-- {describe(entry)}\n{asapquery_sql(entry)}")


align_to_interval is a no-op on main. QueryGroupConfig just copies it into __dict__; main_prometheus_client.py uses now() - query_time_offset each repetition and sleeps repetition_delay_ms after queries finish, so eval times drift by query latency. With intDiv(end_time, T)*T in the SQL, consecutive repetitions eventually floor to the same T (duplicate window) or skip one — breaks the docstring's "every run and arm answers the same windows" until #841 lands.

Agreed: this blocks results, not the merge. Alignment comes with #841's query client. No ClickHouse vs PromQL numbers will be collected from this config until #841 lands. The docstring no longer claims more than the config sets.

milindsrivastava1997 · 2026-10-14T15:29:08Z

+        "dataset": {
+            "name": "custom",
+            "local_data_file": data_file,
+            # The runner's loader fills one table; the paced feeder writes all.


All data lands in one table. dataset.table = next(iter(labels)) and the runner's custom loader streams the whole data_file into that one table; nothing on main ("the paced feeder") fills the others, and unlike generate_workloads.py no simulated_start_time is set. For multi-metric configs (e.g. all-metrics8) the first table gets polluted with other metrics' rows and the rest stay empty; with fixed-timestamp data files, wall-clock windows return empty results.

Real. The comment was wrong: on main the feeder's ClickHouse sink has a fixed schema, and nothing starts one feeder per metric for the ClickHouse runner (#841 feeds only Prometheus and ASAPQuery). In 18b9091 the comment now says the loader fills only dataset.table. I didn't restrict the generator to single-metric tables, since multigroup and all-metrics8 are the point of this PR. Wiring the feeder into the ClickHouse runner is tracked in #868.

milindsrivastava1997 · 2026-10-14T15:29:08Z

@@ -71,14 +71,23 @@ def extrapolated_rate(cols: str, table: str, window: int, is_rate: bool) -> str:
        f"FROM {table} WHERE ts > {END} - {window} AND ts <= {END} "


Two contradictory "Prometheus 2.53" window conventions now coexist. This file uses left-open ts > END - window; sketch_bench_workloads.py uses closed [T-d, T] for the same Prometheus version. With 1s whole-second scrapes a [1m] query reads 60 samples here vs 61 there, so sum/count/quantile and rate extrapolation (first_t/to_start) diverge. Suggest one shared window helper. (Anchored here; the select() at line ~60 has the same predicate.)

Prometheus 2.x range selectors are closed, [T−d, T]; left-open arrived in 3.0. So this PR is correct and generated2 is off by one sample. Changing generated2 (merged in #837) changes its results, so it gets its own PR rather than going into this one: #867. generated2's ts is DateTime, so the fix there is just > → >=.

milindsrivastava1997 · 2026-10-14T15:29:08Z

+        by_interval.setdefault(q.interval_secs, []).append(
+            f"-- {describe(entry)}\n{raw_sql(q, labels)}"
+        )
+        if "promql" not in entry:


Repetitions / starting delay unvalidated. repetitions = duration_secs // interval silently yields 0 when interval_secs > --duration-secs. And starting_delay < QUERY_TIME_OFFSET_SECS + max window (e.g. 1h window, 60s delay) makes early windows reach before ingestion started, biasing rate extrapolation and pooled quantiles. Raise on both.

Fixed in 18b9091. convert raises when an interval exceeds --duration-secs (0 repetitions). A short starting delay now warns instead of raising: every arm sees the same partial early windows, so the comparison holds, and raising would force a delay over 1 h for 1 h-window tables. Both are tested.

milindsrivastava1997 · 2026-10-14T15:29:08Z

+    text = " ".join(promql.split())
+    m = TOPK.match(text)
+    if m:
+        inner = parse_promql(m[2], interval, labels)


parse_promql only collapses whitespace, so valid formatting variants fail: multi-line sum by (job) (\n data\n) normalizes to sum by (job) ( data ) and SELECTOR ^(\w+)$ rejects ' data ' — the test's own describe output hits this. Same for sum by(job) (...), sum(...) by (job), topk(5,sum ...).

Deferring. These forms fail loudly at generation time (ValueError), and all 1908 statements of the real classic, multigroup and all-metrics8 tables parse. One correction: the describe test feeds multi-line PromQL to describe only, never to the parser, so it isn't evidence that real tables hit this. Will widen the regexes if sketch-bench starts emitting these forms.

milindsrivastava1997 · 2026-10-14T15:29:08Z

+    if t % 60 == 0 and 3600 % t == 0:
+        floored = f"end_time % 3600 // {t} * {t} // 60"
+        return f'{{{{ end_datetime[:14] }}}}{{{{ "%02d" | format({floored}) }}}}:00'
+    raise ValueError(f"cannot floor a datetime to {t}s: T must divide 60 or 3600")


No-label metrics produce invalid SQL: labels: [] → ORDER BY (, ts) in init.sql, and SELECT , argMax(value, ts) ... GROUP BY / SELECT , delta * ... in instant/rate paths. Fails at ClickHouse run time rather than at generation time. Either support it or raise here.

Fixed in 18b9091: init_sql raises on a metric with no labels. convert calls it first, so generation fails before any query is rendered. Tested.

milindsrivastava1997 · 2026-10-14T15:29:08Z

+        raise ValueError(f"unknown labels or value for {table!r}: {entry}")
+    return Query(
+        table,
+        entry["interval_secs"],


Unknown table in a structured entry raises bare KeyError from metric_labels(metrics)[table] (no entry context) instead of the module's ValueError. Also metric_labels(metrics) is rebuilt per entry (1696× for all-metrics8) — pass in the labels convert() already has.

Fixed in 18b9091: an unknown table raises ValueError with the entry, and query_from_entry takes the labels map convert already built. Tested.

milindsrivastava1997 · 2026-10-14T15:29:08Z

+    if 60 % t == 0:
+        floored = f"end_time % 60 // {t} * {t}"
+        return f'{{{{ end_datetime[:17] }}}}{{{{ "%02d" | format({floored}) }}}}'
+    if t % 60 == 0 and 3600 % t == 0:


Label / value-field names go into DDL unquoted and aren't checked against ts/value or each other. A gauge label named value, or a records value field sharing a label name, gives a duplicate-column CREATE TABLE; a keyword-named label breaks SELECT/GROUP BY.

Deferring: the current tables don't have these names, and a clash fails loudly at CREATE TABLE.

milindsrivastava1997 · 2026-10-14T15:29:09Z

+        q,
+        tuple(labels[q.table]),
+        q.table,
+        # DateTime64 bounds: ClickHouse 26.10 prunes `ts > <integer>` on the


asapquery_sql copies the table's sql verbatim then joins with ';\n\n'; a trailing ; in sketch-bench's SQL yields ;;. Splitting on ; (as _load_sql_queries does) then produces empty statements and misaligns ASAPQuery-arm query indices vs the ClickHouse arm. Strip trailing ;.

Fixed in 18b9091: trailing ; is stripped. Tested. For the record, it couldn't misalign indices: _load_sql_queries and the runner both drop empty statements (if s.strip()).

milindsrivastava1997 · 2026-10-14T15:29:09Z

+    }
+
+
+def write_statements(path: str, statements: List[str]) -> None:


(workload,) = json.load(f)['workloads'] requires exactly one workload, but the docstring says it reads workloads[0].e2e. More than one gives an opaque unpack error — either use [0] or raise a clear error.

Fixed in 18b9091: anything other than exactly one workload raises a ValueError with the count. Tested.

- Raise on an unknown structured table, a label-less metric, more than one workload, and an interval longer than --duration-secs. - Warn when --starting-delay-secs is shorter than offset + longest window. - Strip a trailing ';' from ASAPQuery sql; reuse convert()'s label map. - Correct the comment on which tables the runner fills. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

milindsrivastava1997 and others added 8 commits October 12, 2026 13:34

feat(experiments): fake_exporter-equivalent dataset for ClickHouse ge…

9d6727f

…nerated2 workloads Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

feat(experiments): translate generated2 PromQL workloads to ClickHous…

dfa8022

…e SQL configs Adds a client_options.simulated_start_time so templated SQL can be evaluated against bulk-loaded data with fixed timestamps. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

milindsrivastava1997 mentioned this pull request Oct 13, 2026

feat(planner): --planner saved for SQL (multigroup e2e tables) #856

Closed

zzylol and others added 2 commits October 13, 2026 18:33

zzylol changed the base branch from clickhouse-generated2-workloads to main October 13, 2026 18:38

milindsrivastava1997 mentioned this pull request Oct 14, 2026

feat(asap-tools): exact materialized-view arm for sketch-bench e2e queries #845

Merged

milindsrivastava1997 reviewed Oct 14, 2026

View reviewed changes

This was referenced Oct 14, 2026

asap-tools: generated2 ClickHouse SQL uses left-open windows; Prometheus 2.53 is closed #867

Open

asap-tools: feed every metric table of a sketch-bench workload into ClickHouse #868

Closed

milindsrivastava1997 merged commit dbca6b6 into main Oct 14, 2026
11 checks passed

milindsrivastava1997 deleted the feat/e2e-sql-from-table branch October 14, 2026 17:50

This was referenced Oct 14, 2026

ClickHouse runner: pass data_start_time to aligned query groups #873

Closed

feat(asap-tools): feed every metric table of a sketch-bench workload into ClickHouse #875

Merged

		@@ -71,14 +71,23 @@ def extrapolated_rate(cols: str, table: str, window: int, is_rate: bool) -> str:
		f"FROM {table} WHERE ts > {END} - {window} AND ts <= {END} "

		}


		def write_statements(path: str, statements: List[str]) -> None:

Conversation

zzylol commented Oct 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Testing

Uh oh!

milindsrivastava1997 commented Oct 14, 2026

Code review

Uh oh!

milindsrivastava1997 commented Oct 14, 2026

Uh oh!

milindsrivastava1997 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

zzylol commented Oct 13, 2026 •

edited

Loading