jonny/pod-evict
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Add main.py

Browse Source 
Initial commit of code
This commit is contained in:
Jonny Ervine 2020-12-15 02:57:02 +00:00
parent d0ca17aa1f
commit 96cedf88f3
1 changed files with 46 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

46
main.py Normal file
View File

@ -0,0 +1,46 @@
import os, sys, json, base64, re;
from kubernetes import client, config
def handler(event, context):
if 'data' in event:
message = json.loads(base64.b64decode(event['data']).decode('utf-8'))
print('Data found - for debugging purposes, message contents printed in next line')
print(message)
else:
print('Data not found - this is an unexpected error, and the function has been called without a Pub/Sub message. This should be checked')
return f'Data not found'
if 'pod_name' in message['incident']['resource']['labels']:
namespace = message["incident"]['resource']['labels']['namespace_name']
pod = message["incident"]['resource']['labels']['pod_name']
cluster = message['incident']['resource']['labels']['cluster_name']
print("Deleting " + pod + " in namespace " + namespace + " in cluster " + cluster + " because it has utilised more than 90% of the configured memory limit")
elif "deadlock_detected" in message['incident']['metric']['type']:
namespace = namespace = message["incident"]['resource']['labels']['namespace_name']
sum = message['incident']['summary']
extract = sum.split("pod_name=", 1)[-1]
pod = re.findall("[\w-]*", extract)[0]
cluster = message['incident']['resource']['labels']['cluster_name']
print("Deleting " + pod + " in namespace " + namespace + " in cluster " + cluster + " because it has encountered a deadlock condition")
elif "request-timeout" in message['incident']['metric']['type']:
namespace = namespace = message["incident"]['resource']['labels']['namespace_name']
sum = message['incident']['summary']
extract = sum.split("pod_name=", 1)[-1]
pod = re.findall("[\w-]*", extract)[0]
cluster = message['incident']['resource']['labels']['cluster_name']
print("Deleting " + pod + " in namespace " + namespace + " in cluster " + cluster + " because it has encountered a frequent request-timeout condition")
else:
print('From unknown source')
if 'KUBERNETES_LOAD_KUBE_CONFIG' in os.environ:
if cluster == 'ggv-k8s-prod':
config.load_kube_config(context='gke_api-project-1004752207173_us-east4_ggv-k8s-prod', persist_config=False)
if cluster == 'ggv-k8s-cluster':
config.load_kube_config(context='gke_api-project-1004752207173_asia-southeast1_ggv-k8s-cluster', persist_config=False)
if cluster == 'ggv-k8s-production-sg':
config.load_kube_config(context='gke_api-project-1004752207173_asia-southeast1_ggv-k8s-production-sg', persist_config=False)
else:
config.load_incluster_config()
v1 = client.CoreV1Api()
body = client.V1beta1Eviction(metadata=client.V1ObjectMeta(name=pod, namespace=namespace))
api_response = v1.create_namespaced_pod_eviction(name=pod, namespace=namespace, body=body)
print('Evicted pod?')