Plan

General Outline

On a cron schedule
Need environment variables indicating account ID, region, cluster
Checking for whether re-balancing is required may be based on:
- Number of containers per instance (count)
- Remaining unreserved memory on each instance (mem-unreserved)
- Remaining available memory on each instance (mem-available)
- Remaining unreserved CPU on each instance (cpu-unreserved)
- Remaining available CPU on each instance (cpu-available)
Calculating whether to re-balance could be achieved with the following approach:
- Fetch all container instances
- Calculate check parameter for each, potentially relative to the size of the instance. This is easy enough for memory and CPU but more tricky for count. Need a proxy for total size of an instance. Could use memory + CPU credits but that's a little strange if there a mix of instance classes, e.g., where ratios of mem:cpu are different.
- Determine mean check parameter value.
- Sort by check parameter, worst first
- For each instance, if check parameter outside allowable variance execute re-balance repeatedly until check parameter within allowable variance
Performing re-balancing may be achieved through:
- Redeploying services on overburdened instances through force redeploy (redeploy-services)
- Draining overburdened instances (drain-instances)