CycleCloud would sometimes get into a mode where connections failed with __getattribute__ not found on type HttpsURLConnectionImpl until it was restarted
When editing accounts, the credential's default region was erroneously reset to WestUS.
In some cases, CycleCloud requested Azure metrics for an interval that was too large and got errors
Spot VM evictions that occurred while VMs were booting were not handled properly
Certain VM errors were incorrectly classified as a VM error rather than a no-spot-capacity indicator
Catching up on metrics could cause an out-of-memory error if CycleCloud had been off for an extended time