[austin-barrington]

• top(), bottom() and sample() work. They were rejected with "unsupported aggregate function".
  • They return up to N stored points per group (GROUP BY time() bucket and tags), each at its own time, in time order. top and bottom break ties by the earliest point.
  • top/bottom take tag arguments (at most one point per combination). Other fields or tags can be selected beside them, taken from the same points.
  • LIMIT/OFFSET apply to the points returned. Combining them with another function fails with InfluxDB's error.
  • Sharded measurements: each region returns its own N per group, and the coordinator ranks the union again, then applies LIMIT/OFFSET. The top N of a union of per-region top-N lists is the global top N, also with tag arguments.
  • The SQL keeps the usual SELECT * FROM (\n...\n) layout, so region and tombstone predicates still filter the points before ranking.
• Subqueries over aggregates work. The inner bucket was renamed to time inside its own SQL, so ClickHouse read time in its GROUP BY as the bucket (NOT_AN_AGGREGATE). An inner aggregate without buckets had no time column at all.
  • The inner query is now wrapped and exposes a real time: the bucket, a lone selector's point time, or the query's lower bound (as InfluxDB does).
• Sharded subqueries work. The gathered inner rows exposed only __time (a string, or the client's epoch unit), so every outer query on time failed. They now carry a real time column.
• A raw field with GROUP BY time() returns InfluxDB's "GROUP BY requires at least one aggregate function" instead of a chDB error.

Tests:

• Compat suite (real chDB): values checked by hand against the data. The suite covers top with and without tags, extra fields, bottom per bucket and host, sample counts, the error cases, subqueries over buckets, over unbucketed aggregates, and over and under top.

• Two-region sharded tests:

  • global top 2 and top per minute
  • bottom per host
  • LIMIT/OFFSET after ranking
  • sample
  • a subquery over buckets

  With the coordinator re-rank removed, the sharded test fails, returning each region's own top 2.

• Unit tests for the re-rank.

• CI-equivalent run (libchdb 26.9.0, -D warnings): passes.

Docs: the selector functions are added to the reference; the clustering deep-dive lists how they merge across regions.

[cursor]

PR Summary

Medium Risk
Touches core query translation, subquery wrapping, and sharded scatter-gather merge paths; behavior changes are broad but covered by compat and cluster integration tests.

Overview
Adds InfluxQL selector functions top(), bottom(), and sample(): up to N points per GROUP BY time() bucket/tag set, with optional tag arguments, LIMIT/OFFSET on returned points, and Influx-style validation (no mixing with other functions; GROUP BY time() without an aggregate errors clearly).

Query engine: New ClickHouse translation ranks points per group (hidden __bucket / __tns columns for coordinator merge). Sharded clusters: each region returns local top-N; the coordinator re-ranks the union (exact for top/bottom), then applies LIMIT/OFFSET—documented in clustering deep-dive.

Subqueries: Fixes aggregate/subquery time exposure (subquery_source, nanosecond inner results on shards) so nested GROUP BY time() and outer filters on time work instead of ClickHouse NOT_AN_AGGREGATE or missing time.

Compat and two-region integration tests cover selectors, subqueries, and multi-region behavior.

^{Reviewed by Cursor Bugbot for commit 3637d91. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 3637d91. Configure here.}

[cursor]

LIMIT applied twice on local fallback

Medium Severity

When shard_routing is set, top/bottom/sample always rerank, which applies LIMIT/OFFSET again. Region SQL strips those clauses first, but the no-space path still runs the original statement locally, so LIMIT/OFFSET is already in the SQL. A query such as top(v, 3) LIMIT 1 OFFSET 1 can return too few points or none on a sharded node that falls back to a local table.

Additional Locations (1)

• hyperbytedb/src/application/query_service.rs#L2751-L2758

 

^{Reviewed by Cursor Bugbot for commit 3637d91. Configure here.}

[austin-barrington]

Right, thanks. Fixed in e533adf: the sharded path now gets the statement without LIMIT/OFFSET on every route (regions and the local fallback), and only the re-rank applies them. Test: top_limit_offset_on_the_local_fallback_applies_once (drops the shard space so the node falls back to its table); it returns no points with the old code.

[cursor]

Stale comment

Left a non-blocking comment: Cursor Bugbot reported one unresolved medium-severity issue on this head (LIMIT/OFFSET can be applied twice on the local fallback path), so I am not approving. Human review is needed, and no reviewers were assigned.

  

_{Sent by Cursor Approval Agent: Pull Request Router and Approver}

[cursor]

Approved. Cursor Bugbot was not running on this head after the first check poll, and the earlier LIMIT/OFFSET finding no longer applies on the current sharded path. CI completed successfully.

  

_{Sent by Cursor Approval Agent: Pull Request Router and Approver}