EC2 InsufficientInstanceCapacity on one VM of a parallel benchmark fleet: relaunch in another AZ, and record it

Launching 5 identical c6id.2xlarge on-demand instances for a parallel benchmark, one failed with InsufficientInstanceCapacity in its availability zone while the other four started. No instance is created on that error, so nothing is billed. Relaunch in another AZ of the same region (same instance type, so per-lane timings stay comparable) and record the AZ per lane in the run's pins. Lanes in different AZs can sit on different hardware generations, so cross-VM numbers are only as comparable as your calibration step makes them. Write launch scripts to take the subnet/AZ as a parameter, so a retry is a one-line change rather than an edited copy. Stagger parallel aws CLI calls by a few seconds, since concurrent calls can also race on the local credential cache.