pg_savior

Overview

-tuplestore_donestoring , breaks on pg18 @ el

Version

Build

You can build the RPM / DEB packages for pg_savior using pig build:

pig build pkg pg_savior         # build RPM / DEB packages

Install

You can install pg_savior directly. First, make sure the PGDG and PIGSTY repositories are added and enabled:

pig repo add pgsql -u          # Add repo and update cache

Install the extension using pig or apt/yum/dnf:

• Install
• pig
• dnf
• apt

pig install pg_savior;          # Install for current active PG version

pig ext install -y pg_savior -v 18  # PG 18
pig ext install -y pg_savior -v 17  # PG 17
pig ext install -y pg_savior -v 16  # PG 16
pig ext install -y pg_savior -v 15  # PG 15
pig ext install -y pg_savior -v 14  # PG 14

dnf install -y pg_savior_18       # PG 18
dnf install -y pg_savior_17       # PG 17
dnf install -y pg_savior_16       # PG 16
dnf install -y pg_savior_15       # PG 15
dnf install -y pg_savior_14       # PG 14

apt install -y postgresql-18-pg-savior   # PG 18
apt install -y postgresql-17-pg-savior   # PG 17
apt install -y postgresql-16-pg-savior   # PG 16
apt install -y postgresql-15-pg-savior   # PG 15
apt install -y postgresql-14-pg-savior   # PG 14

Create Extension:

CREATE EXTENSION pg_savior;

Usage

Sources: README, release 0.1.0, PGXN 0.1.0, SQL file, C source, pg_savior.control

pg_savior is a PostgreSQL safety extension for blocking specific high-risk DML and DDL statements before they run. Version 0.1.0 is a deliberate PGXN release and a major rewrite from 0.0.1; the README still labels it pre-1.0 and not production-ready.

Activation

CREATE EXTENSION alone does not activate the checks. The SQL file only documents that protection lives in the loaded shared library, so each backend must load pg_savior by one of the upstream-supported paths:

Cluster-wide activation uses shared_preload_libraries and requires a PostgreSQL restart:

shared_preload_libraries = 'pg_savior'

Per-session activation for new connections can use session_preload_libraries after a config reload:

session_preload_libraries = 'pg_savior'

For development or test sessions, load the library manually:

LOAD 'pg_savior';
CREATE EXTENSION pg_savior;

Once the library is loaded, _PG_init installs post_parse_analyze_hook, ExecutorStart_hook, and ProcessUtility_hook for that backend.

DML Guards

pg_savior blocks DELETE and UPDATE statements that have no WHERE clause. The parser hook checks the analyzed Query tree and raises ERROR, so the transaction aborts and the application sees the failure.

CREATE TABLE emp (id int);
INSERT INTO emp VALUES (1), (2), (3);

DELETE FROM emp;
-- ERROR: pg_savior: DELETE without WHERE clause is blocked

UPDATE emp SET id = id + 1;
-- ERROR: pg_savior: UPDATE without WHERE clause is blocked

DELETE FROM emp WHERE id = 1;
-- allowed

The optional row-count guard applies to DELETE and UPDATE statements whose planner estimate exceeds pg_savior.max_rows_affected. It runs from ExecutorStart_hook, after planning and before tuples are touched.

SET pg_savior.max_rows_affected = 100;

DELETE FROM emp WHERE id > 0;
-- blocked if the planner estimate is greater than 100 rows

DDL Guards

The ProcessUtility_hook guards only the DDL operations listed by upstream:

• CREATE INDEX without CONCURRENTLY is always blocked.
• DROP DATABASE is always blocked.
• ALTER TABLE ADD COLUMN ... DEFAULT is blocked when the target table is larger than pg_savior.large_table_threshold_rows.
• ALTER TABLE ALTER COLUMN TYPE is blocked for large tables.
• TRUNCATE is blocked when any target table is large.
• DROP TABLE is blocked when any target table is large.

Large-table checks use pg_class.reltuples > pg_savior.large_table_threshold_rows.

CREATE INDEX emp_idx ON emp (id);
-- ERROR: pg_savior: CREATE INDEX without CONCURRENTLY is blocked

CREATE INDEX CONCURRENTLY emp_idx ON emp (id);
-- allowed by this guard

ALTER TABLE big_emp ADD COLUMN status text DEFAULT 'active';
-- blocked when big_emp is over pg_savior.large_table_threshold_rows

TRUNCATE big_emp;
-- blocked when big_emp is over pg_savior.large_table_threshold_rows

Configuration

All documented GUCs are session-scoped USERSET variables:

Use SET LOCAL for a deliberate one-transaction bypass:

BEGIN;
SET LOCAL pg_savior.bypass = on;
DELETE FROM staging_table;
COMMIT;

Caveats

• The library must be loaded in the backend before protection exists; CREATE EXTENSION pg_savior only registers extension metadata.
• The row-count and large-table guards depend on planner/catalog estimates. Run ANALYZE when recent changes make estimates stale.
• UPDATE coverage is limited to unguarded UPDATE and the optional planner-estimate threshold; the README does not claim semantic review of every WHERE predicate.
• DDL coverage is limited to the listed ProcessUtility_hook cases. Do not assume other schema changes are blocked.
• The ADD COLUMN ... DEFAULT guard is conservative and blocks any default on a large table, including non-volatile defaults that newer PostgreSQL versions may handle without a full table rewrite.