Two public cloud servers were affected by a known race-condition related to the virtual filesystem. If a file path is resolved to a vnode inside of an NFS mount at the same time the NFS mount is unmounted, then the file is deleted. This causes a race condition where the Directory Name Lookup Cache (DNLC) is being cleared at the same time that a vnode is being removed from the cache, which results in a host machine reboot.
This was a known race condition that had already been patched in an updated server image, but required a server reboot to apply. This race condition was and is extremely rare. Because of its rarity, it was decided to forgo purposely rebooting servers in order to apply the new image. Each of the affected servers was booted using the updated image and are no longer susceptible to the race condition.
The first server was affected by the race condition at approximately 16:30 UTC and was fully recovered at approximately 17:10 UTC. In the process of the first server coming back online, the same race condition was triggered in a second server housing a service dependent on an NFS service on the first affected server. The second server rebooted at approximately 17:05 UTC and was fully recovered at approximately 17:45 UTC.