gh-ost

gh-ost
Name	gh-ost
Author	GitHub
Released	2016
Programming language	Go
Platform	Cross-platform
License	MIT

gh-ost is an online schema migration tool for MySQL and MariaDB that performs nonblocking table alterations using logical replication and controlled cutovers. It was developed to allow large-scale schema changes in production environments used by cloud providers, content platforms, social networks, and enterprise services. The project is notable within database operations, site reliability engineering, and platform engineering communities for enabling zero-downtime migrations across distributed systems and high-traffic applications.

Overview

gh-ost originated as an engineering solution at GitHub to address live schema changes for large repositories and high-throughput services, following challenges encountered by teams working alongside projects like MySQL, Percona Server, and Amazon Aurora. It fills a niche alongside tools such as pt-online-schema-change from Percona and features discussed in literature from O'Reilly Media and conference presentations at events like Strata Data Conference and Velocity Conference. Operators from organizations including Facebook, Twitter, Netflix, Spotify, Airbnb, and LinkedIn have cited similar migration challenges in talks at KubeCon, AWS re:Invent, and Google Cloud Next.

Design and Architecture

gh-ost is implemented in the Go language and leverages the MySQL binary log (binlog) replication stream used by MySQL and MariaDB to capture data-changing events. Its architecture centers on a lightweight replication daemon that reads events, applies changes to a shadow table, and coordinates a controlled switchover. Components and concepts referenced in system designs include binary log, replication, packet protocol, and orchestration patterns similar to those in Kubernetes and Docker-based deployments. The tool’s interaction models are comparable to change-data-capture systems discussed in literature from Confluent and Debezium and integrate with operational stacks from Prometheus and Grafana for observability.

Features and Use Cases

gh-ost supports online ALTER TABLE operations such as adding columns, dropping indexes, and changing column types for large tables used by platforms like GitHub and cloud services from Amazon Web Services, Google Cloud Platform, and Microsoft Azure. Use cases include migrations for multi-tenant databases in companies like Salesforce and Shopify, analytics platforms similar to Apache Kafka pipelines, and high-traffic web properties comparable to Reddit and Stack Overflow. Features highlighted in talks at Percona Live and FOSDEM include minimal locking, throttling and rate control, safe cutover with hooks for orchestration by Ansible or Terraform, and detailed event logging compatible with monitoring systems such as Datadog.

Operation and Workflow

Operationally, gh-ost creates a shadow table, copies existing rows, streams binlog events to replay concurrent writes, and performs an atomic switch of table names when ready—an approach resonant with techniques described in operational guides from Google SRE and The Linux Foundation. Administrators typically run gh-ost from bastion hosts or CI/CD runners similar to Jenkins or GitLab CI/CD with credentials managed through systems like HashiCorp Vault. It offers hooks for pre- and post-migration scripts, integrates with change management processes at organizations following frameworks such as ITIL and COBIT, and fits into incident response workflows employed by teams at Facebook and Dropbox.

Performance and Limitations

gh-ost reduces tenant-visible downtime but has limitations: it depends on stable binlog availability as found in MySQL 5.6 and later, requires sufficient disk space and I/O throughput akin to provisioning considerations for Amazon RDS or Google Cloud SQL, and may be constrained in environments using complex features like Galera Cluster or certain MariaDB plugins. Performance tuning often references benchmarking techniques used by SPEC and sizing guidance similar to materials from Intel and AMD for storage and CPU. Edge cases include interactions with unusual locking semantics, foreign key constraints, or triggers, topics explored in presentations at Percona Live and papers from ACM SIGMOD.

Adoption and History

After its introduction by GitHub engineers, gh-ost gained adoption among engineering teams at startups and enterprises, mentioned alongside migration stories from Pinterest, Zynga, Dropbox, and Medium in conference talks and blog posts. The project has been discussed in open-source communities including GitHub Actions workflows, package repositories, and issue trackers where contributors from projects like Prometheus and Go ecosystem libraries participate. Its evolution has paralleled shifts in cloud database services from Amazon Aurora and the growing emphasis on continuous delivery popularized by proponents like Jez Humble and Martin Fowler.

Category:Database administration Category:MySQL