Date of Incident: March 24, 2017
Time of Incident: 12:30pm - 2:20pm PDT
Wait times for VMs exceeded 30 seconds, which can cause some tests to error or not start.
Why it Happened:
We have determined that the service responsible for creating and halting VMs was recovering slower than usual after a restart, which happens when Sauce Labs deploys changes.
What we did to fix it:
We increased the overall amount of CPU and RAM we allocate to the service responsible for creating and halting VMs. We also significantly refactored how that service reads and writes to its data store, providing a significant boost in speed.
We thoroughly tested these changes under production conditions and we are confident we have addressed the issue.
What we are doing to prevent this from happening again:
We added a considerable amount of instrumentation and analytics to that service so we can quickly and proactively diagnose similar issues in the future.