Blacksmith Outage History

We've rolled out a fix and are currently monitoring the results. We are seeing major recovery in the request latencies in our dashboard.

Our mitigation has taken effect and job start times and cache operations are improving. Customers may still see some delayed job starts and cache failures while recovery completes. We will provide an update within the next 30 minutes.

Caching has recovered, and job queue times across EU regions have recovered. Queue times for larger jobs (16 and 32 vcpu jobs) in US west and US east continue to remain elevated. We are continuing to monitor for any regressions.

This incident is resolved. Job starts and cache operations returned to normal in all regions, and we are continuing to monitor. Jobs that failed during the incident can be re-run.

We have restored the cache infrastructure in us-west, and cache failures have decreased but are not yet back to normal. Jobs in us-west may still see some cache restores and saves fail and fall back to a full install, while the earlier delayed job starts have cleared and other regions are unaffected. We are bringing additional cache capacity online to fully restore service.

Cache errors for jobs in us-west have largely subsided. As we complete recovery, some jobs may see a one-time cache miss on their next run. We are monitoring and will provide an update within the next hour.

The affected rack remains offline while our upstream provider works to restore its power, and jobs in us-west are running normally on the rest of the fleet. Any jobs that failed with runner communication errors can be safely re-run.

After rerouting traffic, congestion on the upstream network link has subsided, and git checkout performance in us-west has returned to normal. We are monitoring and will resolve this incident once performance has continued to stay stable.

We have identified a small subset of runners that are exhibiting this network stalling and have isolated them. We're continuing to monitor recovery.

Since 06:30 UTC Canonical has had three outages affecting [archive.ubuntu.com](http://archive.ubuntu.com), [us.archive.ubuntu.com](http://us.archive.ubuntu.com) and [security.ubuntu.com](http://security.ubuntu.com). Although they have marked these as resolved, we are still seeing intermittent connection failures to those mirrors, which can cause apt-get steps to fail or hang. In the meantime, pointing apt at [azure.archive.ubuntu.com](http://azure.archive.ubuntu.com) instead of [archive.ubuntu.com](http://archive.ubuntu.com) and [security.ubuntu.com](http://security.ubuntu.com) will unblock affected jobs.

We are deploying a mitigation on our side that routes apt around the affected Canonical mirrors. For the time being pointing apt at [azure.archive.ubuntu.com](http://azure.archive.ubuntu.com) in place of [archive.ubuntu.com](http://archive.ubuntu.com) and [security.ubuntu.com](http://security.ubuntu.com) will unblock affected jobs.

We are seeing recovery of our caching infrastructure and our dashboard. We will continue to monitor this situation closely.

Between approximately 12:20 and 15:15 UTC, some jobs in our eu-west region failed with the GitHub error "The self-hosted runner lost communication with the server"; other regions were unaffected. Job failures in eu-west returned to normal levels as of 15:15 UTC, and any affected jobs can be safely re-run while we continue to investigate the underlying cause.

Our us-west region storage cluster's latency and error rates have returned to baseline as of approximately 17:15 UTC, following mitigations that reduce load on the affected storage. Git-cached checkouts, incremental Docker builds, and container caching in us-west are back to normal, and we are monitoring to confirm the recovery holds while we continue to investigate the underlying cause. We will provide a final update within the next hour.

Connections to [github.com](http://github.com) and [ghcr.io](http://ghcr.io) from our eu-west, eu-central, and us-east regions have returned to normal as of approximately 15:00 UTC, so checkouts are completing at normal speed and login failures to GitHub Container Registry have dropped to baseline levels. We are continuing to monitor, and re-running any jobs that failed during this period should succeed.

Jobs failure rate at an elevated rate caused by an upstream GitHub being rejected by 429 rate-limit errors. We are seeing a single-digit percentage increase in job failure rate across all organizations and are seeing similar failures on non-Blacksmith infrastructure.

We're seeing signs of GitHub recovery. The dashboard is now loading and jobs should be running again. We are monitoring the recovery.

We are still seeing intermitting GitHub API errors at a low rate.

We are no longer seeing upstream errors, and are continuing to monitor impact of the upstream outage.

We implemented a fix and are seeing improvements. We are continuing to monitor the result.

Metrics and log ingestion in our us-west region is recovering, and job performance in the region has returned to normal. We are monitoring to confirm the recovery holds and are continuing to investigate the underlying cause. We will provide an update shortly.

Github has reported degraded performance for Actions, jobs may take a moment to be adopted. We are monitoring this incident.

Jobs adoption has recovered and are operational.

There is still a delay in job adoption times as we continue to recover.

Job adoption is now fully operational.

We implemented a fix and are currently monitoring the result.

We are still monitoring recovery.