GitHub’s monthly reliability summary for July, published on August 12, 2026, counts eight separate incidents that degraded the platform, and it quantifies how much of the company’s own traffic now runs on Azure rather than in its datacenters (GitHub availability report: July 2026).
What broke in July
The month’s worst failure started on July 8, when a change to configuration metadata scrambled service discovery inside data-resident Enterprise Cloud environments and pushed the Web UI, the REST and GraphQL APIs, Actions, Packages, Copilot, and Git operations into 5xx errors for seven hours and four minutes; at peak roughly 84% of active tenants in those environments saw a majority of their requests fail, and the 5xx rate in the hardest-hit region reached about 96%. The remaining events were shorter but hit familiar workflows: an Actions fault on July 9 delayed or broke job starts for nine hours and 18 minutes; the GitHub MCP Server’s web_search tool ran at a 42% average error rate on July 16, peaking at 82%; July 19 delivered both a DNS reconfiguration that degraded github.com for two hours and 11 minutes and an expired certificate that cut larger-runner connectivity for five hours and 10 minutes; SSH authentication failed for 12.2% of requests at peak on July 21; and a July 24 database backfill dropped a backing table, blocking 113,930 pull-request creation attempts from 50,904 users over 57 minutes (the July incident breakdown).
Azure now carries more than half of read traffic
The same report frames those failures against an infrastructure move already past its midpoint. On July 28, Azure Central US served 52.75% of monolith read traffic, the first time that share held above half, while Git traffic on Azure climbed to 47% from 43% in June and a second Central US replica began backing 29% of repositories to speed regional failover. GitHub’s dedicated pull request service reached 99.87% parity with the monolith for authenticated reads, and the company is aiming for 70% of read traffic and 30% of write traffic in Central US this quarter, with a stated line of sight to move dotcom production traffic out of its own datacenters by the end of 2026. The report also calls the August 6 Actions incident “unacceptable” and says GitHub is accelerating the migration of Actions to Azure to buy headroom (GitHub’s migration figures).
Read together, the two halves of the report explain each other: a platform rewiring its request paths region by region will expose configuration, DNS, and certificate edges that a static estate never touches. The summary is signed by Jakub Oleksy, listed with the company’s staff badge on his profile (Jakub Oleksy on GitHub).
For teams that treat GitHub as production infrastructure, the useful response is not alarm at any single outage but pipelines that survive a bad afternoon: retries and idempotent job triggers around Actions, mirrors for clones that must not stall, and a documented fallback for code review when pull request creation refuses to complete. Administrators auditing their own guardrails alongside these reliability notes can also review our coverage of GitHub’s org-wide repository rule insights.
