pg.ddx.io  pgpartman/pg_partman GitHub issues and pull requests  
help / color / mirror / Atom feed
From: princekumar1291 (@princekumar1291) <noreply+princekumar1291@github.com>
To: pgpartman/pg_partman <noreply+pgpartman-pg_partman@github.com>
Subject: [pgpartman/pg_partman] PR #881: Fix format argument references in partition_data_time lock-wait branch
Date: Fri, 09 Oct 2026 08:08:46 +0000
Message-ID: <gh-pgpartman-pg_partman-881@github.com> (raw)

## Summary

Fix the `format()` argument references in the `p_lock_wait > 0` branch of `partition_data_time()`.

The existing expression supplies five arguments but references `%6$L`, which raises `too few arguments for format()` when the source contains data. It also uses `%4$s` for the partition expression, although argument 4 is the lower timestamp bound. Use explicit positional placeholders throughout: schema `%1$I`, table `%2$I`, expression `%3$s`, lower bound `%4$L`, and upper bound `%5$L`.

This leaves the existing `FOR UPDATE NOWAIT` and retry logic unchanged.

## Regression coverage

Add a transactional pgTAP test covering:

- Nonempty DEFAULT migration with `p_lock_wait := 2`, including creation of missing monthly partitions.
- ASC and DESC processing and preservation of the next month's rows between batches.
- Mixed-case schema, table, and control-column identifiers.
- Migration from an explicit source table with a positive lock wait.
- Existing `p_lock_wait := 0` behavior and row/payload preservation.

## Validation

Tested against an isolated local PostgreSQL 18.6 instance with pgTAP 1.3.5:

- Unpatched development source: the new test fails at its first positive-lock-wait migration with `too few arguments for format()`.
- Patched source: new regression test passes all 14 assertions.
- Existing `test-time-monthly.sql`: 121 assertions pass.
- Existing `test-time-daily.sql`: 221 assertions pass.
- Existing `test-time-mixed-case.sql`: 222 assertions pass.
- `git diff --check` passes.

The full suite and PostgreSQL 14–17 were not run locally; the repository's existing CI matrix can provide that coverage. Concurrent row-lock retry behavior was not separately tested; this patch does not change that logic.

## Release integration

This PR changes the canonical function source and tests only. I have not modified an already-published upgrade script or chosen the next release version. Please include the corrected function definition in the next extension upgrade script so existing installations receive the fix too.


inline patch 1: sql/functions/partition_data_time.sql, test/test-time-lock-wait.sql (+63/-2 lines) download inline-1.patchdiff --git a/sql/functions/partition_data_time.sql b/sql/functions/partition_data_time.sql index fb441ab..f51b523 100644 --- a/sql/functions/partition_data_time.sql +++ b/sql/functions/partition_data_time.sql @@ -283,7 +283,7 @@ FOR i IN 1..p_batch_count LOOP WHILE v_lock_iter <= 5 LOOP v_lock_iter := v_lock_iter + 1; BEGIN - v_sql := format('SELECT * FROM ONLY %I.%I WHERE %s >= %L AND %4$s < %6$L FOR UPDATE NOWAIT' + v_sql := format('SELECT * FROM ONLY %1$I.%2$I WHERE %3$s >= %4$L AND %3$s < %5$L FOR UPDATE NOWAIT' , v_source_schemaname , v_source_tablename , v_partition_expression @@ -404,4 +404,3 @@ RETURN v_total_rows; END $$; - diff --git a/test/test-time-lock-wait.sql b/test/test-time-lock-wait.sql new file mode 100644 index 0000000..696622d --- /dev/null +++ b/test/test-time-lock-wait.sql @@ -0,0 +1,62 @@ +-- Exercise the positive lock-wait branch with nonempty DEFAULT and source tables. +\set ON_ERROR_ROLLBACK 1 +\set ON_ERROR_STOP true + +BEGIN; +SELECT set_config('search_path', 'partman, public', false); +SELECT set_config('TimeZone', 'UTC', true); +SELECT plan(14); + +CREATE SCHEMA "Partman Lock Test"; +CREATE TABLE "Partman Lock Test"."Monthly Data" ( + id integer NOT NULL, + payload text NOT NULL, + "Expires On" timestamptz NOT NULL +) PARTITION BY RANGE ("Expires On"); + +SELECT create_partition('Partman Lock Test.Monthly Data', 'Expires On', '1 month'); + +INSERT INTO "Partman Lock Test"."Monthly Data" VALUES + (1, 'first month', (date_trunc('month', CURRENT_TIMESTAMP) + interval '2 years')::date), + (2, 'second month', (date_trunc('month', CURRENT_TIMESTAMP) + interval '2 years 1 month')::date); + +SELECT is((SELECT count(*) FROM ONLY "Partman Lock Test"."Monthly Data_default"), + 2::bigint, 'Future rows initially route to DEFAULT'); +SELECT is(partition_data_time('Partman Lock Test.Monthly Data', p_lock_wait := 2), + 1::bigint, 'Positive lock wait migrates the first monthly batch'); +SELECT is((SELECT count(*) FROM ONLY "Partman Lock Test"."Monthly Data_default"), + 1::bigint, 'The upper bound leaves the second month in DEFAULT'); +SELECT is(partition_data_time('Partman Lock Test.Monthly Data', p_lock_wait := 2, p_order := 'DESC'), + 1::bigint, 'Positive lock wait also works in descending order'); +SELECT is((SELECT count(*) FROM ONLY "Partman Lock Test"."Monthly Data_default"), + 0::bigint, 'DEFAULT is empty after both batches'); +SELECT is((SELECT count(*) FROM "Partman Lock Test"."Monthly Data"), + 2::bigint, 'The parent still exposes both rows'); +SELECT results_eq('SELECT id, payload FROM "Partman Lock Test"."Monthly Data" ORDER BY id', + $$VALUES (1, 'first month'::text), (2, 'second month'::text)$$, + 'Migration preserves row values'); + +CREATE TABLE "Partman Lock Test"."Source Data" (LIKE "Partman Lock Test"."Monthly Data"); +INSERT INTO "Partman Lock Test"."Source Data" VALUES + (3, 'source row', (date_trunc('month', CURRENT_TIMESTAMP) + interval '2 years 2 months')::date); +SELECT is(partition_data_time('Partman Lock Test.Monthly Data', p_lock_wait := 2, + p_source_table := 'Partman Lock Test.Source Data'), + 1::bigint, 'Positive lock wait also migrates an explicit source table'); +SELECT is((SELECT count(*) FROM "Partman Lock Test"."Source Data"), + 0::bigint, 'The explicit source is drained'); +SELECT is((SELECT count(*) FROM "Partman Lock Test"."Monthly Data"), + 3::bigint, 'Source migration adds the row to the parent'); + +INSERT INTO "Partman Lock Test"."Monthly Data" VALUES + (4, 'zero wait row', (date_trunc('month', CURRENT_TIMESTAMP) + interval '2 years 3 months')::date); +SELECT is((SELECT count(*) FROM ONLY "Partman Lock Test"."Monthly Data_default"), + 1::bigint, 'A new future row routes to DEFAULT'); +SELECT is(partition_data_time('Partman Lock Test.Monthly Data', p_lock_wait := 0), + 1::bigint, 'Zero lock wait remains functional'); +SELECT is((SELECT count(*) FROM ONLY "Partman Lock Test"."Monthly Data_default"), + 0::bigint, 'Zero lock wait drains DEFAULT'); +SELECT is((SELECT count(*) FROM "Partman Lock Test"."Monthly Data"), + 4::bigint, 'All four rows remain visible through the parent'); + +SELECT * FROM finish(); +ROLLBACK;

Message-ID: <gh-pgpartman-pg_partman-881@github.com>
Permalink:  ../gh-pgpartman-pg_partman-881@github.com/

 ·  · 


This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox