1 # Incremental SQL Backup System Using PostgreSQL Logical Replication
3 **PostgreSQL Version**: This design is based on PostgreSQL 18.0 documentation. While most features (logical replication, event triggers, pg_recvlogical) are available in earlier versions (PostgreSQL 10+), verify specific parameter availability (e.g., `max_slot_wal_keep_size` requires PostgreSQL 13+) for your target version.
7 This document details the design for a PostgreSQL backup system that produces human-readable, plain SQL incremental backups using logical replication. The system creates backups that remain readable and restorable for 10+ years while supporting online operation and crash safety.
9 **Design Decision**: Use `pg_recvlogical` with the `wal2sql` plugin for DML capture, combined with event triggers using `pg_logical_emit_message()` for DDL tracking and periodic `pg_dumpall --globals-only` for shared objects.
12 - **Built-in tooling handles complexity**: `pg_recvlogical` provides streaming infrastructure, crash recovery, and position tracking
13 - **No transformation layer needed**: `wal2sql` produces production-ready SQL directly
14 - **Complete coverage**: Event triggers + `pg_logical_emit_message()` + `pg_dumpall --globals-only` captures all DDL at correct chronological positions
15 - **Long-term readability**: Plain SQL format that can be executed years later
16 - **Correct DDL/DML ordering**: DDL messages appear in replication stream at exact time of execution
18 **Note on Backup Strategy**:
20 This system implements **differential backups** using logical replication. Each incremental backup file contains all changes since the base backup, allowing simple restore: base + most recent differential. This differs from PostgreSQL 17+'s native incremental backups (`pg_basebackup --incremental`), which are true incremental backups where each backup depends on the previous backup, requiring all incremental backups in sequence to restore. See file-handling.md for detailed terminology and filesystem organization.
23 1. **DDL tracking** - Event triggers emit DDL via `pg_logical_emit_message()`; `pg_dumpall --globals-only` handles shared objects
24 2. **Replica identity configuration** - All tables need proper configuration for UPDATE/DELETE
25 3. **Aggressive monitoring** - Replication slots must be monitored to prevent operational issues
26 4. **wal2sql extension** - ✅ **COMPLETE** - Logical decoding plugin with full DML, DDL, and TRUNCATE support
28 ## Architecture Overview
33 ┌─────────────────────────────────────────────────────────────┐
34 │ PostgreSQL Database │
36 │ ┌────────────────┐ ┌──────────────────┐ │
37 │ │ Regular Tables│────────▶│ WAL (Write-Ahead│ │
38 │ │ (DML Changes) │ │ Log) │ │
39 │ └────────────────┘ └──────────────────┘ │
42 │ ┌─────────────────────────┐ │
43 │ │ Logical Decoding Process│ │
44 │ │ (wal2sql plugin) │ │
45 │ └─────────────────────────┘ │
47 └──────────────────────────────────────┼───────────────────────┘
50 ┌─────────────────────────────┐
52 │ (Tracks position, durable) │
53 └─────────────────────────────┘
56 ┌─────────────────────────────┐
57 │ pg_recvlogical Tool │
58 │ (Built-in PostgreSQL util) │
59 └─────────────────────────────┘
61 ┌──────────────┴──────────────┐
63 ┌─────────────────────┐ ┌─────────────────────┐
64 │ Differential Files │ │ New Chain Creation │
65 │ (active.sql, │ │ (Base Backup) │
66 │ diff-*.sql) │ │ via --new-chain │
67 └─────────────────────┘ └─────────────────────┘
69 └──────────────┬───────────────┘
71 ┌──────────────────────────┐
72 │ Chain-Based Storage │
78 └──────────────────────────┘
83 1. **Logical Replication Slot**: Durable position tracker in PostgreSQL
84 2. **wal2sql Plugin**: ✅ **IMPLEMENTED** - Transforms binary WAL to executable SQL
85 3. **pg_recvlogical**: Built-in PostgreSQL tool that streams logical decoding output
86 4. **Base Backup System**: ⏳ **PLANNED** - Regular full `pg_dump` backups with `--snapshot` for consistency
87 5. **Schema Tracking System**: ⏳ **PLANNED** - Event triggers + `pg_dumpall --globals-only` for DDL changes
91 ### DML Capture via Logical Replication
93 PostgreSQL's logical replication decodes the Write-Ahead Log (WAL) into logical changes. The `wal2sql` plugin outputs these directly as executable SQL:
97 INSERT INTO public.users (id, name, email) VALUES (1, 'Alice', 'alice@example.com');
98 UPDATE public.users SET name = 'Alice Smith' WHERE id = 1;
99 DELETE FROM public.orders WHERE id = 42;
104 - **Crash-safe**: Replication slots persist position across crashes (positions persisted at checkpoint intervals; after crash, slot may return to earlier LSN causing recent changes to be replayed)
105 - **Consistent**: Transaction boundaries are preserved
106 - **Online**: Runs without blocking database operations
107 - **Idempotent positioning**: Can restart from last known position (clients responsible for handling duplicate messages)
109 ### DDL Capture via Event Triggers and Logical Messages
111 Logical replication does **not** capture DDL (CREATE TABLE, ALTER TABLE, etc.). We solve this by emitting DDL commands directly into the logical replication stream using PostgreSQL's `pg_logical_emit_message()` function.
113 The wal2sql extension (installed via `CREATE EXTENSION wal2sql`) provides an event trigger that automatically captures DDL and emits it into the replication stream with the 'ddl' prefix. See `wal2sql/wal2sql--0.1.sql` for the implementation.
116 1. Event trigger fires on all DDL commands
117 2. `pg_logical_emit_message()` writes DDL into WAL as a logical decoding message
118 3. Message appears in replication stream at exact chronological position relative to DML
119 4. Output plugin's `message_cb` callback outputs DDL as executable SQL
120 5. Restore is simple: just execute the incremental backup file sequentially
122 **Example incremental backup output**:
125 INSERT INTO public.users (id, name, email) VALUES (1, 'Alice', 'alice@example.com');
128 -- DDL message appears at exact time of execution
129 ALTER TABLE public.users DROP COLUMN email;
132 -- Subsequent DML only references remaining columns
133 INSERT INTO public.users (id, name) VALUES (2, 'Bob');
138 - ✅ **Perfect chronological ordering**: DDL appears exactly when it was executed
139 - ✅ **Transactional integrity**: DDL message commits with its transaction
140 - ✅ **Simple restore**: Execute backup file sequentially with `psql -f`
141 - ✅ **PostgreSQL built-in**: `pg_logical_emit_message()` available since PostgreSQL 9.6
144 - Event triggers don't fire for shared objects: databases, roles, tablespaces, parameter privileges, and ALTER SYSTEM commands
145 - Solution: Use periodic `pg_dumpall --globals-only` to capture shared objects
147 ## Database Schema Prerequisites
149 This section summarizes the **mandatory database requirements** for this backup system. The setup script MUST validate these and fail with clear errors if not met.
151 ### Critical Requirements
153 **1. Replica Identity** (see "Key Challenges" section #1 for details):
154 - Every table MUST have: PRIMARY KEY, or `REPLICA IDENTITY USING INDEX`, or `REPLICA IDENTITY FULL`
155 - Without this: UPDATE/DELETE operations **fail on the source database**
156 - Validation: `SELECT n.nspname, c.relname FROM pg_class c JOIN pg_namespace n ON n.oid = c.relnamespace WHERE c.relkind = 'r' AND c.relreplident = 'd' AND NOT EXISTS (SELECT 1 FROM pg_index i WHERE i.indrelid = c.oid AND i.indisprimary) AND n.nspname NOT IN ('pg_catalog', 'information_schema');`
158 **2. PostgreSQL Configuration**:
159 - `wal_level = logical` (required)
160 - `max_replication_slots >= 1` (required)
161 - `max_wal_senders >= 1` (required)
163 **3. Event Trigger Installation**:
164 - Run `CREATE EXTENSION wal2sql;` in every database being backed up
165 - Without this: DDL changes NOT captured
167 ### Objects NOT Captured (plan alternatives)
169 - **Unlogged/temporary tables**: Silently excluded from logical replication
170 - **Sequence state**: Not replicated incrementally (restore script automatically synchronizes via `setval()`)
171 - **Large objects**: Use `BYTEA` instead or accept loss (see "Key Challenges" #4)
172 - **Shared objects**: Captured via `pg_dumpall --globals-only`
173 - **System columns**: `ctid`, `xmin`, `xmax` will differ after restore
175 ### Additional Constraints
177 - **REPLICA IDENTITY FULL** + geometric types (`point`, `box`, `circle`, etc.): Use PRIMARY KEY instead
178 - **Foreign keys**: TRUNCATE CASCADE fails if related tables split across different publications
179 - **Row-level security**: Disable RLS on target tables during restore
180 - **Triggers**: Disable triggers during restore to avoid duplicate side effects
181 - **Extensions**: Must be available on restore target (document in metadata)
182 - **Collation/encoding**: Must be compatible between source and target
184 ### Pre-Deployment Validation
187 -- CRITICAL: Find tables without adequate replica identity
188 SELECT n.nspname, c.relname, c.relreplident
189 FROM pg_class c JOIN pg_namespace n ON n.oid = c.relnamespace
190 WHERE c.relkind = 'r' AND n.nspname NOT IN ('pg_catalog', 'information_schema')
191 AND c.relreplident IN ('d', 'n') -- DEFAULT or NOTHING
192 AND NOT EXISTS (SELECT 1 FROM pg_index i WHERE i.indrelid = c.oid AND i.indisprimary);
193 -- Must return 0 rows
195 -- CRITICAL: Verify configuration
196 SHOW wal_level; -- Must be 'logical'
197 SELECT count(*) FROM pg_event_trigger WHERE evtname = 'wal2sql_ddl_trigger'; -- Must be 1
199 -- WARNING: Check for unlogged tables (won't be backed up)
200 SELECT n.nspname, c.relname FROM pg_class c JOIN pg_namespace n ON n.oid = c.relnamespace
201 WHERE c.relkind = 'r' AND c.relpersistence = 'u'
202 AND n.nspname NOT IN ('pg_catalog', 'information_schema');
204 -- WARNING: Check for large objects (not incrementally backed up)
205 SELECT count(*) FROM pg_largeobject_metadata;
208 See "Key Challenges and Solutions" section below for detailed explanations and solutions.
210 ## Implementation Components
212 ### 1. Initial Setup Script
214 **Purpose**: Bootstrap the backup system
215 **Status**: ⏳ **PLANNED** - Not yet implemented
217 **Philosophy**: The setup script validates prerequisites and **fails fast** with clear error messages. It does NOT automatically fix schema issues - human oversight required for replica identity decisions due to performance tradeoffs.
220 1. **Validate Prerequisites** (print report to stdout):
221 - **CRITICAL** (must fix before proceeding):
222 - Verify `wal_level = logical`, `max_replication_slots >= 1`, `max_wal_senders >= 1`
223 - Check all tables have adequate replica identity (PRIMARY KEY, USING INDEX, or FULL)
224 - **Coverage Warnings**:
225 - List unlogged tables (will not be backed up)
226 - Check for large objects (not incrementally backed up)
227 - **If validation fails**: Print detailed report and EXIT with error
228 - **Human must fix CRITICAL issues** before re-running setup
230 2. **Setup Replication Infrastructure** (after validation passes):
231 - Create logical replication slot with snapshot export
232 - Capture the exported snapshot identifier
233 - Set up event triggers via `CREATE EXTENSION wal2sql`
235 3. **Initial Backup**:
236 - Take synchronized base backup using exported snapshot (`pg_dump`)
237 - Create initial `pg_dumpall --globals-only` backup
238 - Write `pg_scribe_version.txt` to backup directory (PostgreSQL version, pg_scribe version, setup timestamp)
240 **Note**: Most metadata (sequences, extensions, replica identity, encoding) is already captured in the `pg_dump` and `pg_dumpall` files. The restore script queries this information directly from the backup files or restored database as needed.
242 **Critical Detail - Initial Snapshot Consistency**:
244 From PostgreSQL documentation (Section 47.2.5):
245 > When a new replication slot is created using the streaming replication interface, a snapshot is exported which will show exactly the state of the database after which all changes will be included in the change stream.
247 This ensures the base backup and incremental stream are perfectly aligned with no gaps or overlaps.
249 **Step-by-Step Setup Procedure**:
251 1. **Verify prerequisites**:
252 - Check backup directory is empty (or create if doesn't exist)
253 - Verify replication slot doesn't already exist (fail if it does)
254 - Validate PostgreSQL configuration and table replica identity
256 2. **Create replication slot**:
257 - Use `pg_create_logical_replication_slot()` SQL function
258 - Capture returned LSN position for base backup coordination
259 - If creation fails, exit with error (no cleanup needed yet)
261 3. **Take synchronized base backup**:
262 - Use `pg_dump --file=backup.sql` immediately after slot creation
263 - The slot preserves WAL from its creation LSN forward, ensuring no changes are lost
264 - Track created files for cleanup if subsequent steps fail
266 4. **Capture globals and metadata**:
267 - Run `pg_dumpall --globals-only` for roles, tablespaces
268 - Document PostgreSQL version, extensions, encoding in metadata file
269 - Track all created files for cleanup if any step fails
271 **Critical Considerations**:
272 - **Cleanup on failure**: If any step fails, automatically remove replication slot and partial backup files
273 - **Non-idempotent**: Following PostgreSQL conventions (`initdb`, `pg_basebackup`), the operation requires a clean state and does not support re-running
274 - **Empty directory required**: Like `pg_basebackup`, the backup directory must be empty to prevent accidental overwrites
276 ### 2. Incremental Backup Collection
278 **Status**: ✅ **IMPLEMENTED** - `pg_scribe --start` command available
279 **Tool**: Built-in `pg_recvlogical` utility with `wal2sql` plugin
281 **Implementation Approach**:
283 The `pg_scribe --start` command is a thin wrapper that:
284 1. Validates the database connection and replication slot
285 2. Displays configuration and user information
286 3. Uses `exec` to **replace itself** with `pg_recvlogical`
288 This design has several advantages over traditional parent-child process management:
289 - **No orphaned processes**: The pg_scribe process becomes pg_recvlogical (same PID)
290 - **Direct signal handling**: Signals (SIGTERM, SIGINT, SIGHUP) go directly to pg_recvlogical
291 - **Simpler implementation**: No need for signal forwarding or child process tracking
292 - **Reliable cleanup**: Test suites and process management tools interact with a single process
294 **Key Configuration**:
296 The wrapper invokes `pg_recvlogical --start` with the following important parameters:
298 - **Status interval** (`--status-interval`): Controls how often client reports position back to server (default: 10 seconds)
299 - Lower values allow server to advance slot position and free WAL more quickly
300 - Balance between slot health and network overhead
302 - **Fsync interval** (`--fsync-interval`): Controls disk write safety (default: 10 seconds)
303 - Frequency of forced disk synchronization for crash safety on backup client
304 - 0 disables fsync (faster but risks data loss if client crashes)
305 - Higher values improve performance but increase window of potential loss
307 - **Plugin options** (`--option`): Pass `include_transaction=on` to wal2sql
308 - Includes BEGIN/COMMIT statements in output
309 - Essential for maintaining transaction boundaries during restore
311 **What pg_recvlogical provides**:
312 - Streams decoded changes continuously from the replication slot
313 - Handles connection failures and automatic reconnection
314 - Tracks LSN positions with status updates to the server
315 - Supports file rotation via SIGHUP signal (for log rotation without stopping stream)
317 **What wal2sql provides** (✅ **ALL IMPLEMENTED**):
318 - ✅ Schema-qualified table names: `INSERT INTO public.users (...)`
319 - ✅ Proper column name quoting with `quote_identifier()`
320 - ✅ Transaction boundaries via `include_transaction=on` option (BEGIN/COMMIT)
321 - ✅ Intelligent replica identity handling (DEFAULT, INDEX, FULL, NOTHING)
322 - ✅ Comprehensive data type support (booleans, NaN, Infinity, NULL, strings, bit strings)
323 - ✅ TOAST optimization (skips unchanged TOAST columns in UPDATEs)
324 - ✅ Production-quality memory management
325 - ✅ **`message_cb` callback** to handle DDL messages from `pg_logical_emit_message()`
326 - ✅ **`truncate_cb` callback** to handle TRUNCATE operations
328 ### 3. Periodic Full Backup Script
330 **Status**: ⏳ **PLANNED** - Not yet implemented
331 **Purpose**: Take regular full backups as restore points
334 - Execute `pg_dump --file=backup.sql` to create full database backup in plain SQL format
335 - Execute `pg_dumpall --globals-only` to capture shared objects (databases, roles, tablespaces)
336 - Compress backups to save space
337 - Implement retention policy (delete old backups)
340 - Use `pg_dump --file=backup.sql` for plain SQL output
341 - Maintains human-readable format consistent with design goals
342 - Single connection to database
343 - Universal compatibility across PostgreSQL versions
346 - Use `pg_dump --compress=<method>` for built-in compression (e.g., `--compress=zstd:9`)
347 - Or compress after creation with external tools (gzip, zstd)
348 - Compression significantly reduces storage requirements while maintaining recoverability
350 ### 4. Restore Script
352 **Status**: ⏳ **PLANNED** - Not yet implemented
353 **Purpose**: Restore database to latest captured state from base + incremental backups
356 1. Locate most recent full backup and all incremental backups since then
357 2. Create new target database
358 3. Restore `pg_dumpall --globals-only` (shared objects: roles, tablespaces)
359 4. Restore full `pg_dump` backup using `psql -f backup.sql`
360 5. Disable triggers to prevent duplicate side effects during replay
361 6. Apply all incremental SQL backups in chronological order using `psql -f incremental-*.sql`
362 - DDL and DML are already interleaved in correct chronological order
363 7. Automatically synchronize all sequence values to match table max values
364 8. Re-enable triggers
365 9. Verify data integrity (row counts, application smoke tests)
368 - **Sequences**: Script automatically runs `setval()` for all sequences after restore
369 - **Triggers**: Script automatically disables/re-enables to prevent side effects
370 - **RLS**: Script handles row-level security if present (disable during restore or use superuser)
372 **Handling Duplicate Transactions**:
374 After PostgreSQL crash, replication slot may return to earlier LSN, causing some transactions to be streamed again. The restore process handles this naturally through idempotent operations:
376 - Most SQL operations are idempotent or fail safely:
377 - INSERT fails on duplicate primary key (acceptable during restore)
378 - UPDATE reapplies same values (idempotent)
379 - DELETE succeeds or reports row not found (acceptable)
380 - Transaction boundaries (BEGIN/COMMIT from `include_transaction=on`) preserve consistency
381 - Simply apply all incremental files in order; duplicates will be handled correctly
382 - No additional LSN tracking infrastructure required
384 ### 5. Monitoring and Health Check Script
386 **Status**: ⏳ **PLANNED** - Not yet implemented
387 **Purpose**: Prevent operational issues from inactive replication slots
389 **Critical Metrics**:
391 -- Check replication slot health
397 pg_size_pretty(pg_wal_lsn_diff(pg_current_wal_lsn(), restart_lsn)) as replication_lag,
398 pg_size_pretty(pg_wal_lsn_diff(pg_current_wal_lsn(), confirmed_flush_lsn)) as confirmed_lag
399 FROM pg_replication_slots
400 WHERE slot_type = 'logical';
404 - **Critical alert** when `restart_lsn` falls more than 1GB behind
405 - **Emergency alert** when slot lag exceeds 10GB or age exceeds 24 hours
406 - **Emergency procedure** documented to drop slot if it threatens database availability
409 - **Prometheus + postgres_exporter + Grafana**: Open-source monitoring stack
410 - **pgDash** (https://pgdash.io/): Commercial PostgreSQL monitoring
411 - **check_postgres**: Nagios/Icinga/Zabbix integration
412 - **Built-in views**: `pg_replication_slots`, `pg_stat_replication_slots`
414 ## wal2sql Plugin Details
416 **Status**: ✅ **COMPLETE** - All required functionality implemented and tested
417 **Source**: Forked from decoder_raw (https://github.com/michaelpq/pg_plugins/tree/main/decoder_raw)
418 **Location**: `wal2sql/` directory
419 **Compatibility**: PostgreSQL 18.0+
425 PGUSER=postgres make installcheck # Run tests
428 **Implemented Features**:
429 - ✅ INSERT/UPDATE/DELETE statements with proper escaping
430 - ✅ Transaction boundaries (BEGIN/COMMIT)
431 - ✅ Replica identity handling (DEFAULT, INDEX, FULL, NOTHING)
432 - ✅ **`message_cb`** - DDL capture via `pg_logical_emit_message()`
433 - ✅ **`truncate_cb`** - TRUNCATE statement capture
434 - ✅ Comprehensive data type support (booleans, NaN, Infinity, NULL, strings)
435 - ✅ TOAST optimization
436 - ✅ Full test coverage
438 ## Key Challenges and Solutions
440 ### 1. Replica Identity Required for UPDATE/DELETE
442 **Problem**: Tables need replica identity for UPDATE/DELETE operations.
444 From PostgreSQL documentation (Section 29.1.1):
445 > Tables with a replica identity defined as `NOTHING`, `DEFAULT` without a primary key, or `USING INDEX` with a dropped index, **cannot support UPDATE or DELETE operations**. **Attempting such operations will result in an error on the publisher.**
447 This means UPDATE/DELETE will **fail on the source database**, not just during restore!
449 **Solution**: Ensure all tables have one of:
450 - A primary key (automatic replica identity)
451 - A unique index configured via `REPLICA IDENTITY USING INDEX index_name`
452 - Explicit `REPLICA IDENTITY FULL` setting (inefficient, last resort)
456 -- Table without primary key will error on UPDATE/DELETE
457 CREATE TABLE logs (timestamp TIMESTAMPTZ, message TEXT);
459 -- Fix: Set replica identity to FULL
460 ALTER TABLE logs REPLICA IDENTITY FULL;
463 ### 2. Replication Slots Prevent WAL Cleanup
465 **Problem**: Inactive replication slots prevent WAL cleanup, leading to:
466 1. Disk fills up (WAL files not cleaned)
467 2. Table bloat (VACUUM cannot clean old row versions)
468 3. **Database shutdown** (transaction ID wraparound)
470 From PostgreSQL documentation (Section 47.2.2):
471 > In extreme cases this could cause the database to shut down to prevent transaction ID wraparound.
474 - **Monitor slot lag aggressively** (see monitoring section)
475 - Set `max_slot_wal_keep_size` parameter (PostgreSQL 13+) to limit WAL retention
476 - Have documented emergency procedure to drop slot if needed
477 - Consider `pg_replication_slot_advance()` to skip ahead (loses backup coverage)
479 ### 3. Sequences Are Not Replicated
481 **Problem**: Sequence values are not captured in logical replication.
484 - Use `pg_dump --sequence-data` (enabled by default) in periodic full dumps
485 - After restore, synchronize sequences:
487 SELECT setval('users_id_seq', (SELECT MAX(id) FROM users));
490 ### 4. Large Objects Are Not Replicated
492 **Problem**: PostgreSQL large objects are not captured in logical replication.
495 - **Preferred**: Use `BYTEA` columns instead (these ARE replicated)
496 - **Alternative**: Use `pg_dump --large-objects` in periodic full backups
497 - Note: Incremental changes to large objects NOT captured between full backups
499 ### 5. Crash Recovery and Duplicate Handling
501 **Problem**: After database crash, slot position may roll back, causing duplicate changes.
503 From PostgreSQL documentation (Section 47.2.2):
504 > The current position of each slot is persisted only at checkpoint, so in the case of a crash the slot might return to an earlier LSN, which will then cause recent changes to be sent again when the server restarts.
506 **Solution**: The restore process handles duplicates naturally through idempotent operations. Per PostgreSQL documentation (Section 47.2.2): "Logical decoding clients are responsible for avoiding ill effects from handling the same message more than once."
509 - Most SQL operations in backup files are naturally idempotent:
510 - INSERT will fail on duplicate primary key (acceptable during restore)
511 - UPDATE will reapply same values (idempotent)
512 - DELETE will succeed or report row not found (acceptable)
513 - Transaction boundaries (BEGIN/COMMIT from `include_transaction=on`) ensure consistency
514 - Simply apply all incremental files in chronological order
515 - No additional LSN tracking infrastructure required
516 - See Restore Script section (Section 4) for implementation details
519 - Test crash scenarios with `pg_ctl stop -m immediate` to verify duplicate handling
520 - Monitor `confirmed_flush_lsn` lag during normal operations (see Monitoring section)
522 ### 6. Long-Term Readability
525 - PostgreSQL syntax may change between major versions (rare)
526 - Extension dependencies may not exist in future systems
527 - Encoding/collation definitions may change
529 **Solution**: Include metadata file with each backup:
530 - PostgreSQL version (full version string)
531 - All installed extension names and versions
533 - Locale and collation settings
534 - Custom data types and enums
536 Periodically test restoring old backups on current PostgreSQL versions.
538 ## Prerequisites and Configuration
540 ### PostgreSQL Configuration
545 # Required: Set WAL level to logical
548 # Required: Allow at least one replication slot
549 max_replication_slots = 10
551 # Recommended: Allow replication connections
554 # Recommended: Keep more WAL for safety
557 # Recommended: Limit WAL retention for safety (PostgreSQL 13+)
558 max_slot_wal_keep_size = 10GB
560 # Optional: Tune checkpoint frequency to persist slot positions more often
561 checkpoint_timeout = 5min
564 ### Client Requirements
566 - PostgreSQL client utilities installed (`pg_recvlogical`, `pg_dump`, `pg_dumpall`)
567 - Superuser or role with `REPLICATION` privilege
568 - Permission to create replication slots
569 - wal2sql plugin compiled and installed
571 ## Operational Procedures
576 - **Continuous streaming**: Run `pg_scribe --start` as a daemon (writes to `active.sql` in latest chain)
577 - **Differential rotation**: Daily via `pg_scribe --rotate-diff` (seals daily differential)
578 - **New chain creation**: Monthly via `pg_scribe --new-chain` (fresh base backup + globals + metadata)
579 - **Monitoring**: Periodic slot health checks via `pg_scribe --status`
583 - **Differential rotation**: Rotate differentials daily (via `--rotate-diff`)
584 - **Chain rotation**: Create new chains monthly (via `--new-chain`)
585 - **Chain retention**: Keep recent chains for 30 days, then one per month for 1 year
586 - **Cleanup**: Manual removal of old chains using standard Unix tools (`rm -rf chain-{CHAIN_ID}`)
587 - **Monitor disk space**: Alert if backup directory exceeds 80% capacity
589 ### Disaster Recovery Runbook
591 1. **Stop application** to prevent new writes during restore
592 2. **Create new database** (don't overwrite production)
593 3. **Use `--restore` command**: `pg_scribe --restore --backup-dir /backups/production -d restored_db --create`
594 - Automatically restores globals (roles, tablespaces)
595 - Applies base backup from selected chain
596 - Applies all sealed differentials in chronological order
597 - DDL and DML are already interleaved in correct chronological order
598 4. **Sync sequences**: Automatically synchronized via `setval()` (unless `--no-sync-sequences` specified)
599 5. **Verify data integrity**: Check row counts, run application smoke tests
600 6. **Test application** against restored database
601 7. **Switch over** application to restored database
603 **Manual restore (alternative)**:
605 CHAIN_ID=20231215T120000Z # Or use latest
606 CHAIN_DIR=/backups/production/chain-$CHAIN_ID
608 # Restore globals and base
609 psql -f $CHAIN_DIR/globals.sql
610 psql dbname < $CHAIN_DIR/base.sql
612 # Apply all sealed differentials
613 for f in $CHAIN_DIR/diff-*.sql; do
618 pg_scribe --restore --backup-dir /backups/production --chain-id $CHAIN_ID \
619 -d dbname # (handles sequence sync automatically)
622 **See Section 4 (Restore Script)** for detailed procedures including sequence synchronization and duplicate transaction handling. See `doc/cli.md` for `--restore` command options.
626 ### 1. Basic Functionality Test
629 -- Create test database and setup
630 CREATE DATABASE backup_test;
634 CREATE TABLE test_users (id SERIAL PRIMARY KEY, name TEXT, created_at TIMESTAMP DEFAULT now());
636 -- Generate data and schema changes to test DDL/DML ordering
637 INSERT INTO test_users (name) VALUES ('Alice'), ('Bob'), ('Charlie');
638 UPDATE test_users SET name = 'Alice Smith' WHERE id = 1;
639 DELETE FROM test_users WHERE id = 3;
641 -- Add column - DDL message should appear in stream here
642 ALTER TABLE test_users ADD COLUMN email TEXT;
644 -- Use the new column - should work because DDL already executed
645 UPDATE test_users SET email = 'alice@example.com' WHERE id = 1;
647 -- Drop column - DDL message should appear in stream here
648 ALTER TABLE test_users DROP COLUMN created_at;
650 -- Subsequent inserts should work without the dropped column
651 INSERT INTO test_users (name, email) VALUES ('David', 'david@example.com');
653 -- Restore and verify:
654 -- 1. All operations should replay successfully
655 -- 2. DML before column add should not reference email column
656 -- 3. DML after column add should reference email column
657 -- 4. DML after column drop should not reference created_at column
660 ### 2. Crash Recovery Test
663 # Start collecting incrementals
664 # Generate load with pgbench
665 # Simulate crash: pg_ctl stop -m immediate
667 # Verify no data loss and duplicates handled correctly
671 ### 3. Long-Term Storage Test
676 # Wait (or simulate) years passing
677 # Restore on modern PostgreSQL version
678 # Verify SQL is still readable and executable
681 ### 4. Replica Identity Test
684 -- Create table without primary key
685 CREATE TABLE test_no_pk (col1 TEXT, col2 INT);
687 -- Attempt UPDATE (should fail with replica identity error)
688 UPDATE test_no_pk SET col2 = 5 WHERE col1 = 'test';
690 -- Fix with REPLICA IDENTITY FULL
691 ALTER TABLE test_no_pk REPLICA IDENTITY FULL;
693 -- Retry UPDATE (should succeed)
696 ### 5. TRUNCATE Handling Test
700 CREATE TABLE test_truncate (id INT);
701 INSERT INTO test_truncate VALUES (1), (2), (3);
704 TRUNCATE test_truncate;
706 -- Verify: Check if wal2sql incremental backup captured TRUNCATE
707 -- Expected: ✅ CAPTURED by wal2sql with truncate_cb
708 -- Look for: TRUNCATE TABLE public.test_truncate;
709 -- Note: Event triggers do NOT capture TRUNCATE (it's DML, not DDL)
711 -- Test TRUNCATE with multiple tables (foreign key cascade)
712 CREATE TABLE parent_table (id INT PRIMARY KEY);
713 CREATE TABLE child_table (parent_id INT REFERENCES parent_table(id));
714 INSERT INTO parent_table VALUES (1), (2);
715 INSERT INTO child_table VALUES (1), (2);
717 -- TRUNCATE CASCADE should capture both tables
718 TRUNCATE parent_table, child_table;
719 -- Expected output: TRUNCATE TABLE public.parent_table, public.child_table;
722 ## Performance Considerations
724 **Write Amplification**:
725 - WAL must be written (normal)
726 - WAL must be decoded into logical format (additional CPU)
727 - Event triggers fire on every DDL operation (minimal overhead)
730 - Additional WAL volume retained by replication slots
731 - More frequent checkpoint I/O if checkpoint_timeout is tuned
734 - Benchmark overhead on test system with production-like workload
735 - Monitor CPU usage of WAL sender processes
736 - Monitor disk usage for WAL and backup directories
738 ## Implementation Status & Next Steps
742 1. **wal2sql Plugin** - Fully implemented and tested
743 - ✅ Core DML operations (INSERT/UPDATE/DELETE)
744 - ✅ `message_cb` for DDL capture
745 - ✅ `truncate_cb` for TRUNCATE support
746 - ✅ Comprehensive test suite passing
750 1. **Setup Script** - Create initial backup environment
751 - Create replication slot with `wal2sql`
752 - Set up event triggers for DDL capture using `pg_logical_emit_message()`
753 - Take synchronized base backup with `pg_dump --snapshot`
754 - Configure `REPLICA IDENTITY` on tables
756 2. **Backup Collection Script** - Wrapper around `pg_recvlogical`
757 - File rotation and timestamping
758 - Metadata generation (PostgreSQL version, extensions, encoding)
759 - Error handling and restart logic
761 3. **Restore Script** - Apply base + incremental backups
762 - Locate and order backup files
763 - Apply in sequence with error handling
764 - Sequence synchronization with `setval()`
766 4. **Monitoring Script** - Replication slot health checks
767 - Track slot lag and WAL retention
768 - Alert on dangerous conditions
769 - Emergency procedures documentation
771 5. **Integration Testing**
772 - End-to-end backup and restore
773 - DDL/DML interleaving scenarios
774 - Crash recovery with `pg_ctl stop -m immediate`
775 - Performance benchmarking
779 ### PostgreSQL Documentation
780 - PostgreSQL Documentation: Chapter 25 - Backup and Restore
781 - PostgreSQL Documentation: Chapter 29 - Logical Replication
782 - PostgreSQL Documentation: Chapter 47 - Logical Decoding
783 - PostgreSQL Documentation: Section 29.8 - Logical Replication Restrictions
784 - PostgreSQL Documentation: Section 47.6.4.8 - Generic Message Callback (message_cb)
785 - PostgreSQL Documentation: Section 9.28.6 - `pg_logical_emit_message()` function
786 - PostgreSQL Documentation: `pg_recvlogical` man page
787 - PostgreSQL Documentation: `pg_dump` man page
788 - PostgreSQL Documentation: `pg_dumpall` man page
791 - **wal2sql**: ✅ **IMPLEMENTED** - SQL output plugin for logical decoding
792 - Location: `wal2sql/` directory in this repository
793 - Forked from: https://github.com/michaelpq/pg_plugins/tree/main/decoder_raw
794 - License: PostgreSQL License (production-ready)
795 - Compatibility: PostgreSQL 18.0+
798 - **Prometheus + postgres_exporter + Grafana**: Open-source monitoring stack
799 - **pgDash**: PostgreSQL monitoring - https://pgdash.io/
800 - **check_postgres**: Nagios/Icinga/Zabbix integration
801 - **pg_stat_replication_slots**: Built-in PostgreSQL monitoring view