Follow-up for compressed chunk collation #4236

Add a changelog message and add a check for broken chunks to the update
script.
This commit is contained in:
Alexander Kuzmenkov 2022-04-21 16:56:05 +03:00 committed by Alexander Kuzmenkov
parent 730a72880e
commit a1e76d2e84
3 changed files with 71 additions and 1 deletions

View File

@ -6,6 +6,8 @@ accidentally triggering the load of a previous DB version.**
## Unreleased
If you use compression with a non-default collation on a segmentby-column you might have to recompress the affected hypertable.
**Features**
* #4120 Add logging for retention policy
* #4169 Add support for chunk exclusion on DELETE to PG14
@ -15,6 +17,7 @@ accidentally triggering the load of a previous DB version.**
* #3899 Fix segfault in Continuous Aggregates
* #4225 Fix TRUNCATE error as non-owner on hypertable
* #4259 Fix logic bug in extension update script
* #4236 Fix potential wrong order of results for compressed hypertable with a non-default collation
**Thanks**
* @jsoref for fixing various misspellings in code, comments and documentation

View File

@ -88,7 +88,7 @@ docker_pgcmd() {
docker_pgscript() {
local database=${3:-single}
docker_exec $1 "psql -h localhost -U postgres -d $database $PGOPTS -v ON_ERROR_STOP=1 -f $2"
docker_exec $1 "psql --set VERBOSITY=verbose --set ECHO=all -h localhost -U postgres -d $database $PGOPTS -v ON_ERROR_STOP=1 -f $2"
}
docker_pgtest() {

View File

@ -5,3 +5,70 @@ AS '@MODULE_PATHNAME@', 'ts_relation_size' LANGUAGE C VOLATILE;
DROP VIEW IF EXISTS _timescaledb_internal.hypertable_chunk_local_size;
DROP INDEX IF EXISTS _timescaledb_catalog.chunk_constraint_chunk_id_dimension_slice_id_idx;
CREATE INDEX chunk_constraint_dimension_slice_id_idx ON _timescaledb_catalog.chunk_constraint (dimension_slice_id);
-- Report the compressed chunks that have a wrong collation. See https://github.com/timescale/timescaledb/pull/4236
DO $$
DECLARE
_hypertable regclass;
_column_name text;
_chunks regclass[];
BEGIN
FOR _hypertable,
_column_name,
_chunks IN
-- We materialize this CTE so that the filter on dropped chunks works
-- first, and we don't try to look up regclass for dropped chunks.
WITH chunk AS MATERIALIZED (
SELECT
format('%I.%I', compressed_chunk.schema_name, compressed_chunk.table_name) compressed_chunk,
format('%I.%I', normal_chunk.schema_name, normal_chunk.table_name) normal_chunk,
normal_chunk.hypertable_id hypertable_id
FROM
_timescaledb_catalog.chunk normal_chunk,
_timescaledb_catalog.chunk compressed_chunk
WHERE
normal_chunk.compressed_chunk_id = compressed_chunk.id
AND NOT normal_chunk.dropped
),
col AS (
SELECT
hypertable_id,
normal_chunk,
normal_column.attname column_name
FROM
chunk,
pg_attribute normal_column,
pg_attribute compressed_column
WHERE
normal_column.attrelid = normal_chunk::regclass
AND compressed_column.attrelid = compressed_chunk::regclass
AND normal_column.attname = compressed_column.attname
AND compressed_column.atttypid != '_timescaledb_internal.compressed_data'::regtype
AND normal_column.attcollation != compressed_column.attcollation
),
report_rows AS (
SELECT
format('%I.%I', schema_name, table_name)::regclass hypertable,
normal_chunk::regclass chunk,
column_name
FROM
col,
_timescaledb_catalog.hypertable
WHERE
hypertable.id = hypertable_id
)
SELECT
hypertable,
column_name,
array_agg(chunk) chunks
FROM
report_rows
GROUP BY
hypertable,
column_name LOOP
RAISE warning 'some compressed chunks for hypertable "%" use a wrong collation for the column "%"', _hypertable, _column_name
USING detail = 'This may lead to wrong order of results if you are using an index on this column of the compessed chunk.',
hint = format('If you experience this problem, disable compression on the table and enable it again. This will require decompressing and compressing all chunks of the table. The affected chunks are "%s".', _chunks);
END LOOP;
END
$$;