Incident with Pull Requests
This summary is created by Generative AI and may differ from the actual content.
Overview
On May 6, 2026, between 15:12 UTC and 19:02 UTC, GitHub’s Pull Request service was unable to create new review threads (line‑ and file‑level comments). Existing threads and top‑level comments continued to work, and other PR operations (open, merge, actions) were unaffected. The failure was caused by a 32‑bit integer key in a Vitess lookup table reaching its maximum value after the primary table had already been migrated to a 64‑bit key. Updating the lookup table schema to 64‑bit across all shards restored normal operation. The team will add monitoring for Vitess lookup tables to warn of future column‑size limits before they impact customers.
Impact
The incident prevented users from creating any new review threads on pull requests, resulting in a near‑100 % failure rate for that functionality. Top‑level comments, opening/merging PRs, and actions remained operational. The outage lasted roughly four hours, affecting developers’ ability to conduct code reviews during that window.
Trigger
A 32‑bit integer column in a Vitess lookup table overflowed when its value exceeded the maximum representable number, after the primary table had already been migrated to a 64‑bit key space.
Detection
Degraded availability was observed by internal monitoring systems and corroborated by user reports of failures when attempting to create new PR comment threads. The on‑call team was paged once the error rate spiked.
Resolution
Engineers altered the schema of the affected Vitess lookup table on every shard, changing the key column from 32‑bit to 64‑bit. After the schema change propagated globally, new thread creation succeeded and the service returned to normal.
Root Cause
The primary PR thread table had been migrated to a 64‑bit identifier, but the associated Vitess lookup table was left with a 32‑bit identifier. This mismatch allowed the lookup table’s key to overflow, causing the creation‑thread failures.