mergeTreeCodecBlockCounts

Reports, per (part, column, substream) of a MergeTree table, how many compressed blocks use each codec. This is how you observe adaptive codec selection (enabled by allow_experimental_adaptive_codec_selection setting), which can pick a codec per block for default-codec columns. Selecting codec_block_counts reads .bin data files, not just metadata. The other columns are metadata-only. Parts that do not record their substreams in columns_substreams.txt are not listed. If a row policy applies to the table for the current user, reading codec_block_counts throws ACCESS_DENIED, because the counts would cover rows the policy hides. The other columns stay readable, system.parts_columns reports them regardless of row policies.

Syntax

mergeTreeCodecBlockCounts(database, table)

Arguments

Argument	Description
`database`	The database name of the table.
`table`	The MergeTree table name.

Returned value

A table object with one row per (active part, column, substream) of the source table:

part_name (String) — The active data part the column belongs to.
column (String) — The column name.
substream (String) — The physical stream of the column the counts are for. Matches system.parts_columns.substreams.
data_compressed_bytes (Nullable(UInt64)) — Size of compressed data in the substream, in bytes. NULL for Compact parts.
data_uncompressed_bytes (Nullable(UInt64)) — Size of uncompressed data in the substream, in bytes. NULL for Compact parts.
codec_block_counts (Map(String, UInt64)) — The number of compressed blocks of this substream grouped by codec. Empty for Compact parts, whose columns share one data file and so have no per-stream codec attribution.

Usage example

CREATE TABLE mt (a UInt64) ENGINE = MergeTree ORDER BY a
SETTINGS min_bytes_for_wide_part = 0;

INSERT INTO mt SELECT number FROM numbers(100000);

SELECT column, substream, codec_block_counts
FROM mergeTreeCodecBlockCounts(currentDatabase(), mt);

┌─column─┬─substream─┬─codec_block_counts─┐
│ a      │ a         │ {'LZ4':13}         │
└────────┴───────────┴────────────────────┘

Column-level totals are a GROUP BY column with sumMap:

SELECT column, sumMap(codec_block_counts)
FROM mergeTreeCodecBlockCounts(currentDatabase(), mt)
GROUP BY column;

A LowCardinality column reports its dictionary and indexes streams separately:

CREATE TABLE mt_lc (s LowCardinality(String)) ENGINE = MergeTree ORDER BY tuple()
SETTINGS min_bytes_for_wide_part = 0;

INSERT INTO mt_lc SELECT toString(number % 1000) FROM numbers(1000000);

SELECT substream, codec_block_counts
FROM mergeTreeCodecBlockCounts(currentDatabase(), mt_lc)
WHERE column = 's'
ORDER BY substream;

┌─substream─┬─codec_block_counts─┐
│ s         │ {'LZ4':31}         │
│ s.dict    │ {'LZ4':1}          │
└───────────┴────────────────────┘

SQL Reference

Data Types

Engines

Functions

Formats

Settings

System Tables

Data Lakes

Syntax

Arguments

Returned value

Usage example

​Syntax

​Arguments

​Returned value

​Usage example

Syntax

Arguments

Returned value

Usage example