Amazon ECS integrated with Ocean uses a draining process to safely remove container instances while allowing running tasks to complete. This article explains how ECS draining operates, the default timeout settings, and what occurs when the timeout is reached.

Draining timeout

When no draining timeout is defined, Ocean uses a default value of 900 seconds (15 minutes). Ocean enforces a minimum draining timeout of 900 seconds. If a user sets a value less than 900, Ocean still applies 900 seconds.

If a user specifies a draining timeout greater than 900 seconds, Ocean will apply the user-defined value.

Draining process

When an ECS-integrated instance is detached:

1. Ocean retrieves these instances and applies the following steps:

   • Instances without container instances that passed their grace period: Marked for termination.

   • Container instances not drained: Initiate draining.

   • Container instances with running tasks: Stop the tasks.

   • Instances in DRAINING status with zero tasks: Marked for termination, logging The following instances have finished draining.

2. If an instance approaches its termination time (within 2 minutes of the drainingTimeout limit), it is marked for termination, logging The following instances have passed their draining grace period and will be deleted:

3. Finally, Ocean terminates all instances marked for termination.

Important behavior

Even if the termination grace period has not fully elapsed, if an instance is in DRAINING status and all its tasks have successfully stopped, Ocean will terminate it immediately.

Outcome

By understanding ECS draining behavior and configuring an appropriate draining timeout, workloads maintain high availability during instance termination without task interruption.