Building Multi Master Workflows on PostgreSQL Logical Replication
Presented by:
Dr. Ibrar Ahmed
Ibrar Ahmed, Principal Engineer at pgEdge, brings 25 years of experience in software design and open-source development, particularly PostgreSQL. With a strong background in system-level embedded development, Ibrar has made impactful contributions during his tenure at companies like EnterpriseDB, Percona, and Bitnine. Since 2006, he's been instrumental in enhancing PostgreSQL's core engine, driving performance improvements, and refining essential modules.
His expertise spans MySQL, Oracle, and NoSQL solutions like MongoDB and Hadoop, alongside tools like Hive, HBase, and Spark. A prolific author and blogger, Ibrar shares deep insights into PostgreSQL with several authoritative books. Over the past year, he’s delivered over fifteen talks worldwide at PostgreSQL conferences, further cementing his reputation. His dedication to advancing data management technology continues to shape the PostgreSQL landscape.
No video of the event yet, sorry!
PostgreSQL logical replication provides stable one-way change delivery, supports live upgrades, and keeps standby nodes ready, yet it remains tied to a single source of truth. Building a multi-master workflow demands explicit control over data flow, identity rules, and conflict resolution. This session presents a structured method for extending logical replication into a bidirectional or mesh arrangement. The discussion begins with publication and subscription design for each node, including the mapping rules that avoid redundant loops. Sequence handling follows, covering allocation ranges, generation patterns, and safe collision prevention across writers. Conflict resolution uses commit timestamp ordering and replica identity selection to control row-level merges without silent loss. DDL propagation methods address schema updates through controlled queues or metadata channels to avoid drift. Latency effects and split-brain risk are examined to show how write-anywhere designs behave under network delays. The session documents frequent failure cases such as missing replica identities, heavy write bursts, and incremental schema changes that diverge. The goal is a straightforward procedure for building, operating, and evaluating multi-master logical replication in environments that require distributed write capacity.
- Date:
- Duration:
- 50 min
- Room:
- Conference:
- Postgres Conference: 2026
- Language:
- Track:
- Ops
- Difficulty:
- Medium