DROP STATISTICS

The DROP STATISTICS statement discards statistics previously collected by ANALYZE TABLE. The table's data is untouched; only the optimizer's metadata about it is removed.

Syntax

DROP STATISTICS ON <workspace>.<collection>.<table_name>
  [ FOR COLUMNS <column> [, ...] ];

Note the ON — unlike ANALYZE TABLE, this statement names its target after ON, and any other form is rejected.

Parameters

<workspace>.<collection>.<table_name> — the table whose statistics should be dropped, fully qualified.
FOR COLUMNS <column> [, ...] — limit the drop to the named columns, comma-separated. Omitted, all statistics for the table are dropped.

Examples

Discard All Statistics for a Table

DROP STATISTICS ON workspace.collection.large_dataset;

Discard Statistics for Specific Columns

DROP STATISTICS ON workspace.collection.large_dataset FOR COLUMNS region, created_at;

Columns not named keep their statistics.

Why Drop Statistics

Statistics describe the data as it was when ANALYZE TABLE last ran. If a table has changed shape substantially and has not been re-analyzed, stale statistics can be worse than none — the optimizer will plan confidently from a distribution that no longer holds. Dropping them returns the optimizer to planning without that input until the table is analyzed again.

Notes

Requires the owner role on the table — the same tier as ANALYZE TABLE, since it destroys what that statement builds.
Local filesystem datasets only. Catalog-backed (workspace) datasets are rejected: their manifest entries carry statistics from the moment each file is written, so there is no "statistics absent" state to drop back to — removing them would mean deleting the dataset's record of the file itself. Other backends are rejected with ANALYZE / DROP STATISTICS is not supported for this dataset's storage backend.
Idempotent — dropping statistics that were never collected is not an error.
Dropping statistics never changes query results, only the plans chosen to produce them.
Re-run ANALYZE TABLE to collect them again.