TechPrufer
Get started
PlanetScale
PlanetScale logoPlanetScale · Feature trail

Postgres High Availability with CDC

6 posts · Oct 2020 – Sep 2026

What this trail covers

PlanetScale offers robust database replication solutions for MySQL and Postgres, focusing on high availability and data consistency. For MySQL, best practices include active/passive configurations, using GTIDs for transaction tracking, semi-synchronous replication for enhanced consistency with configurable timeouts, and optimizing log storage. The platform emphasizes proactive monitoring and a clear failover strategy. For cross-datacenter replication, asynchronous mode is recommended due to late This post details how PlanetScale's custom Kubernetes operator enhances Postgres high availability by actively managing logical replication slots during cutovers. It explains the importance of `failover = true` for logical slots to ensure they are synchronized with replicas, preventing data loss for downstream consumers. The operator detects misconfigurations, alerts users, and blocks planned cutovers that would compromise replication slot integrity. It also manages `synchronized_standby_slots` to prevent a single dead replica from blocking all logical replication. The post also clarifies the trade-offs of enabling `hot_standby_feedback` and `sync_replication_slots`.

All PlanetScale trails
  1. Blocking cutovers to save replication slotsThis post details how PlanetScale's custom Kubernetes operator enhances Postgres high availability by actively managing logical replication slots during cutovers. It explains the importance of `failover = true` for logical slots to ensure they are synchronized with replicas, preventing data loss for downstream consumers. The operator detects misconfigurations, alerts users, and blocks planned cutovers that would compromise replication slot integrity. It also manages `synchronized_standby_slots` to prevent a single dead replica from blocking all logical replication. The post also clarifies the trade-offs of enabling `hot_standby_feedback` and `sync_replication_slots`.
  2. Postgres High Availability with CDCThis post details the challenges of Postgres logical replication slots for CDC and HA, contrasting it with MySQL's GTID-based approach. It highlights how Postgres's slot progress dependency can hinder failover readiness and introduce coupling with external systems, while MySQL's binlog retention offers greater flexibility. The post quantifies the failure scenarios in Postgres, such as temporary slots on standbys during CDC quiet periods and issues with replacing replicas, and explains how MySQL's binlog architecture avoids these HA constraints.
  3. MySQL replication: Best practices and considerationsThis post details best practices for MySQL replication, emphasizing active/passive configurations over active/active due to conflict resolution challenges. It strongly advocates for the use of Global Transaction Identifiers (GTIDs) for robust transaction tracking and discusses the benefits and trade-offs of asynchronous versus semi-synchronous replication, noting PlanetScale's use of semi-sync within regions and high timeouts for consistency. The post also covers optimizing log storage by separating binary logs from database disks, the importance of monitoring replication lag, and outlines a step-by-step strategy for unplanned failovers. Finally, it addresses considerations for cross-datacenter replication, recommending asynchronous mode due to increased latency and highlighting the importance of understanding network latency between regions.
  4. What is MySQL replication and when should you use it?This post details the use cases for MySQL replication, including read-only connections, backups, analytical workloads, high availability, and planned upgrades. It explains how replication can be used to balance read load, offload backup processes, isolate analytical queries, facilitate high availability through failover, and enable rolling upgrades with minimal downtime. The post also touches on the potential for replication lag in read-only scenarios.
  5. Building a multi-region Rails application with PlanetScaleThis post details how to build a multi-region Rails application by co-locating database replicas with application servers globally. It covers setting up read-only replicas in different regions, configuring Rails' Active Record to use multiple databases (primary for writes, replicas for reads), and addressing replication lag with a 2-second delay for reading own writes. The post also explains how to dynamically connect to the nearest database replica based on application deployment region, significantly reducing latency for read-heavy workloads.
  7. MySQL semi-sync replication: durability consistency and split brainsThis post introduces and explains MySQL semi-synchronous replication, detailing its configuration parameters (`rpl_semi_sync_master_wait_for_slave_count`, `rpl_semi_sync_master_timeout`). It elaborates on how semi-sync guarantees durability by persisting changelogs on replicas before acknowledging commits and discusses its implications for consistency, particularly in the context of eventual consistency and avoiding split-brain scenarios. The post analyzes various topologies and failure scenarios, including colocated primaries and replicas, network partitions, and the challenges of determining replica status when multiple semi-sync replicas are lost. It highlights the trade-offs between durability, consistency, availability, and the cost of achieving them.