GitHub Availability Report: March 2026

In March 2026, GitHub experienced a total of 4 incidents, resulting in degraded performance and request failures across multiple services.

On March 3, github.com, GitHub API, GitHub Actions, Git operations, and GitHub Copilot were affected for 1 hour and 10 minutes. At peak, the github.com request failure rate was about 40%, GitHub API was about 43%, HTTP-based Git operations were about 6%, and GitHub Copilot was about 21%, while SSH was not affected.

The cause was a surge in write volume to a user settings caching mechanism. As cache expiration, recalculation, and rewriting occurred simultaneously across all users, replication lag spread, and GitHub responded by immediately rolling back the problematic deployment.

Immediate follow-up actions included:

• Adding a killswitch to the caching mechanism and strengthening monitoring
• Moving the relevant cache to dedicated hosts to limit the scope of impact

On March 5, GitHub Actions was delayed for 2 hours and 55 minutes. 95% of workflow runs failed to start within 5 minutes, with an average delay of 30 minutes, and 10% failed due to infrastructure errors.

The cause was an incorrect configuration loaded into a load balancer during a Redis infrastructure update, which sent internal traffic to the wrong hosts. GitHub recovered by fixing the load balancer, then reverted all related changes and froze changes in this area.

GitHub also announced further automation improvements, blocking the propagation of incorrect configurations, strengthening proactive detection alerts for load balancers, and adjusting Redis client settings for Actions.

On March 19 and March 20, the Copilot Coding Agent service was unstable, blocking the start of new sessions and lookups of existing sessions. The first incident had an average error rate of about 53%, peaking at 93%, while the second averaged 99%, peaking at 100%, with severe retry amplification.

Both incidents stemmed from a system authentication issue that prevented connections to the backup datastore. Recovery was achieved by rotating the affected credentials, and the second incident recurred because the first fix was incomplete.

On March 24, Microsoft Teams Integration and Teams Copilot Integration failed to deliver GitHub event notifications to Teams. The average error rate was 37.4%, peaking at 90.1%, and during this period about 19% of all integration installations did not receive GitHub-to-Teams notifications.

The cause was HTTP 500 errors and connection resets resulting from an upstream dependency outage. GitHub worked with the relevant service teams and resolved the issue at 19:51 UTC, when the upstream outage was mitigated, and is now pursuing observability and runbook improvements.