Linear Says Its Alerts Did Not Reliably Page the On-Call Engineer During an EU Sync Outage

From September 26 to 27, 2026, live updates were unavailable to users of Linear's EU service for approximately 10 hours. Linear's post mortem says service has been restored, and changes made while offline should now sync. It apologizes for the disruption.

Customers encountered connection failures, the post mortem says. Linear also says its alerts "did not reliably page the on-call engineer despite sustained customer impact." Parts of the post mortem follow, not the whole.

How Sync Works, and What Broke

In its overview, the post mortem explains the setup. A sync source reads a database change stream. Sync servers subscribe to that stream, apply authorization filters for each client connection, and send updates to clients.

Linear says database maintenance delayed the rollout of its sync source. It did so by blocking the creation of a replication slot in the database. Its timeline says the database maintenance was started by a cron to repack tables in the EU cell, at 09:00 UTC on September 26. It also says that, because of earlier sync server issues, the rollout was delayed until late Saturday.

When the sync source became available, the sync servers hit a timing bug. Linear describes it as a race between stream closure and decoding. That race made the servers drop a reset message. Replacement servers could not become ready after the previous ones stopped serving traffic.

What the Alarms Did

Near the start of the outage, a monitor fired and kept reporting the failure. But it notified an infrastructure alerts channel, not the primary on-call engineer.

At 01:05 UTC a separate alert, that "linear-sync has too few available replicas," was acknowledged. It was then resolved without corrective action. Linear says that was because its message was unclear.

One dashboard showed the sync server pods as "running and healthy." Other dashboards, not linked in the monitor, would have shown that sync was not sending sync packets to clients.

Throughout the outage, sync traffic telemetry stopped. Yet there was no reliable page for a service that was not processing any traffic. Customers encountered 503 connection failures, and there was no load-balancer alert for them. The persistent source-unavailable condition did not repeatedly page an on-call responder either.

Recovery and the Fixes

At 10:41 UTC, an engineer raised an incident internally and paged the team on call. At 10:57 the team restarted the affected sync servers, which restored their ability to connect. At 11:15 the EU service was reported operational, and the team kept checking service health and processing queued changes. The public status page incident was marked resolved at 12:02.

Linear says it has deployed a fix for the way its sync servers handle the reset. It has improved alert routing, alert wording, coverage of missing data, and repeat notifications. It is exploring "volume-based paging alerts for weekend support volume." And it is following up on the safety of its sync source rollouts.