March Availability Report: Four Incidents Across GitHub Services
GitHub experienced four distinct incidents in March 2026, each causing degraded performance across different areas of the platform. The affected services ranged from core infrastructure like github.com and the API to specific offerings like GitHub Actions and the Teams integrations.
User Settings Cache Bug Hits Core Services
On March 3, between 18:46 and 20:09 UTC, a widespread incident impacted github.com, the GitHub API, GitHub Actions, Git operations, GitHub Copilot, and dependent services. At its peak, approximately 40% of github.com requests failed, and around 43% of API requests errored. HTTP Git operations saw roughly a 6% error rate, SSH remained unaffected, and Copilot requests failed at about 21%. Actions impact stayed under 1%.
The root cause originated from the same issue that triggered a February incident: heavy writes to the user settings caching mechanism. During a deployment meant to reduce that write load, a bug forced every user's cache to expire, recalculate, and rewrite simultaneously. The resulting load created replication delays that propagated across all affected services. Rolling back the faulty deployment resolved the issue.
GitHub has taken immediate steps to prevent recurrence. A killswitch and improved monitoring have been added to the caching mechanism to ensure earlier notification and faster response. Additionally, the cache is being migrated to a dedicated host so that any future issue will be isolated to services that depend on it.
Redis Misconfiguration Disrupts Actions
GitHub Actions was degraded on March 5 from 16:24 to 19:30 UTC. During this window, 95% of workflow runs failed to start within five minutes, with an average start delay of 30 minutes, and 10% of runs failed with an infrastructure error.
The cause was traced to a set of Redis infrastructure updates being rolled out to production. An incorrect configuration change to the Redis load balancer routed internal traffic to an errant host, which triggered two related incidents. The misconfiguration was corrected, and jobs began running successfully by 17:24 UTC. The remaining outage time was spent processing the backlog of queued jobs.
The contributing updates were rolled back, and all changes in that area are frozen pending follow-up work. GitHub is enhancing automation to prevent incorrect configuration propagation, improving alerting to catch load balancer misconfigurations earlier, and updating the Redis client configuration in Actions to better tolerate brief cache interruptions.
Credential Rotation Issues Strike Copilot Agent
The Copilot Coding Agent service was unavailable on March 19 from 01:05 to 02:52 UTC and again on March 20 from 00:42 to 01:58 UTC. Users could not start new Copilot Agent sessions or view existing ones during these periods.
During the first incident, the average error rate was approximately 53%, peaking at 93%. The second occurrence was more severe, with an average error rate of about 99% and a peak of effectively 100%, compounded by significant retry amplification. Both incidents stemmed from the same system authentication problem that prevented the service from connecting to its backing datastore.
In each case, rotating the affected credentials restored connectivity and returned error rates to normal. The first mitigation took 01:24. The second incident occurred because the remediation for the first was incomplete. GitHub has since implemented automated monitoring for credential lifecycle events and is refining operational processes to speed up detection and mitigation.
Upstream Outage Takes Down Teams Integration
From March 24, 15:57 to 19:51 UTC, the Microsoft Teams Integration and Teams Copilot Integration services were degraded, unable to deliver GitHub event notifications to Microsoft Teams. The average error rate was 37.4%, peaking at 90.1% of requests. Roughly 19% of all integration installs failed to receive GitHub-to-Teams notifications during this period.
An outage at an upstream dependency produced HTTP 500 errors and connection resets for the Teams integration. GitHub coordinated with the responsible service teams, and the issue resolved at 19:51 UTC once the upstream incident was mitigated. Observability and runbooks are being updated to cut time to mitigation for similar issues moving forward.
