Update readme for unstable reserved memory value reported when ECS_PO…

…LL_METRICS is enabled and ECS_POLLING_METRICS_WAIT_DURATION is set to a high value

README.md

@@ -190,7 +190,7 @@ additional details on each available environment variable.
 | `ECS_UPDATES_ENABLED` | <true | false> | Whether to exit for an updater to apply updates when requested. | false | false |
 | `ECS_DISABLE_METRICS`     | <true | false>  | Whether to disable metrics gathering for tasks. | false | false |
 | `ECS_POLL_METRICS`     | <true | false>  | Whether to poll or stream when gathering metrics for tasks. Setting this value to `true` can help reduce the CPU usage of dockerd and containerd on the ECS container instance. See also ECS_POLL_METRICS_WAIT_DURATION for setting the poll interval. | `false` | `false` |
-| `ECS_POLLING_METRICS_WAIT_DURATION` | 10s | Time to wait between polling for metrics for a task. Not used when ECS_POLL_METRICS is false. Maximum value is 20s and minimum value is 5s. If user sets above maximum it will be set to max, and if below minimum it will be set to min. | 10s | 10s |
+| `ECS_POLLING_METRICS_WAIT_DURATION` | 10s | Time to wait between polling for metrics for a task. Not used when ECS_POLL_METRICS is false. Maximum value is 20s and minimum value is 5s. If user sets above maximum it will be set to max, and if below minimum it will be set to min. As the number of tasks/containers increase, a higher `ECS_POLLING_METRICS_WAIT_DURATION` value can potentially cause a problem where memory reservation value of ECS cluster reported in metrics becomes unstable due to missing metrics sample at metric collection time. It is recommended to keep this value smaller than 18s. This behavior is only observed on certain OS and platforms. | 10s | 10s |
 | `ECS_PULL_DEPENDENT_CONTAINERS_UPFRONT` | <true | false> | Whether to pull images for containers with dependencies before the dependsOn condition has been satisfied. | false | false |
 | `ECS_RESERVED_MEMORY` | 32 | Reduction, in MiB, of the memory capacity of the instance that is reported to Amazon ECS. Used by Amazon ECS when placing tasks on container instances. This doesn't reserve memory usage on the instance. | 0 | 0 |
 | `ECS_AVAILABLE_LOGGING_DRIVERS` | `["awslogs","fluentd","gelf","json-file","journald","logentries","splunk","syslog"]` | Which logging drivers are available on the container instance. | `["json-file","none"]` | `["json-file","none"]` |