DownResolved 1
Bluesky goes down intermittently for about half its users for 8 hours after a missing concurrency limit
An engineer's postmortem traced the worst outage in his time at the company to one unbounded lookup endpoint.
Bluesky's engineering postmortem says the service went down intermittently for roughly half of users for about eight hours on Monday, April 6, after degraded service over the preceding weekend. The engineer who wrote it called it the worst outage in his time at the company.1
The cause was a lookup endpoint with no concurrency limit. A newly deployed internal service sent batches of 15,000 to 20,000 post URIs, which exhausted network ports on the memcached connections. Heavy error logging and out-of-memory restarts then fed a death spiral.1
The author also corrected the status page, which had blamed a third-party provider; the postmortem said that was wrong. Fixes planned include per-client observability.1
Sources
- 1April 2026 Outage Post-Mortem
pckt (Bluesky engineer blog) · Official statement · Apr 10, 2026
Related
- Judged magnitude 4.0
- Judged magnitude 4.0
- Judged magnitude 4.0
- Judged magnitude 5.0