Merge pull request segmentio#104 from segmentio/drain-instances-for-rollover

achille-roussel · web-flow · commit 32b234f368af · 2017-03-28T18:52:50.000-07:00
[WIP] [roll-ami] drain old instances before terminating:
diff --git a/tools/roll-ami b/tools/roll-ami
@@ -6,9 +6,12 @@ import queue
 import sys
 import threading
 import time
+import copy
 
 asg = boto3.client('autoscaling')
 ec2 = boto3.client('ec2')
+ecs = boto3.client('ecs')
+
 
 def main(argv):
     options = parse_arguments(argv)
@@ -108,6 +111,28 @@ def run(options, step_size=1, launch_config=None, force=False, ignore_instances=
             info('%s - running', instance['InstanceId'])
             tmp_instances.remove(instance['InstanceId'])
 
+    container_instance_id_map = build_container_instance_map(options.cluster)
+
+    # Drain old instances
+    info('draining ECS container instances...')
+    for instance in old_instances:
+        info('%s - draining', instance)
+    ecs.update_container_instances_state(
+        cluster=options.cluster,
+        containerInstances=[container_instance_id_map[instance] for instance in old_instances],
+        status='DRAINING')
+
+    draining_instances = copy.copy(old_instances)
+    while draining_instances:
+        time.sleep(5)
+        ci_info = ecs.describe_container_instances(
+            cluster=options.cluster,
+            containerInstances=[container_instance_id_map[instance] for instance in draining_instances])
+        for container_instance in ci_info['containerInstances']:
+            if container_instance['runningTasksCount'] == 0:
+                draining_instances.remove(container_instance['ec2InstanceId'])
+                info('%s - drained', container_instance['ec2InstanceId'])
+
     # Terminates the old instances that aren't necessary anymore (the ones
     # that were picked by the iterator).
     for instance in old_instances:
@@ -209,6 +234,16 @@ def filter_new_instances(instances, group_instances):
     new_instances.sort()
     return new_instances
 
+def build_container_instance_map(cluster):
+    container_instances = ecs.list_container_instances(cluster=cluster)
+    full_info = ecs.describe_container_instances(cluster=cluster,
+                                                 containerInstances=container_instances['containerInstanceArns'])
+
+    container_instance_map = {}
+    for ci in full_info['containerInstances']:
+        container_instance_map[ci['ec2InstanceId']] = ci['containerInstanceArn']
+    return container_instance_map
+
 def iter_instance_groups(instances, group_size):
     while instances:
         n = min(group_size, len(instances))
