You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: ADOPTERS.md
+1Lines changed: 1 addition & 0 deletions
Original file line number
Diff line number
Diff line change
@@ -59,5 +59,6 @@ If you are open to others contacting you about your use of Karpenter on Slack, a
59
59
| Whoosh | Using Karpenter to scale the EKS clusters for many purposes |`@vainkop`|[Whoosh](https://whoosh.bike)|
60
60
| Next Insurance | Using Karpenter to manage the nodes in all our EKS clusters, including dev and prod, on demand and spots |`@moshebs`|[Homepage](https://www.nextinsurance.com)|
61
61
| Grover Group GmbH | We use Karpenter for efficient and cost effective scaling of our nodes in all of our EKS clusters |`@suraj2410`|[Homepage](https://www.grover.com/de-en) & [Engineering Techblog](https://engineering.grover.com)|
62
+
| Legit Security | We run Karpenter across all our EKS clusters to ensure efficient and cost-effective scaling across our infrastructure |`@Tal Balash`, `@Matan Ryngler`|[Homepage](https://www.legitsecurity.com)|
62
63
| Logz.io | Using Karpenter in all of our EKS clusters for efficient and cost effective scaling of all our K8s workloads |`@pincher95`, `@Samplify`|[Homepage](https://logz.io/)|
63
64
| X3M ads | We have been using Karpenter for (almost) all our workloads since 2023 |`@mreparaz`, `@fmansilla`, `@mrmartinez95`|[Homepage](https://x3mads.com)|
EventBridge is an Event Bus service within AWS that allows users to set rules on events to capture and then target destinations for those events. Relevant targets for Spot interruption notifications include SQS, Lambda, and EC2-Terminate-Instance.
48
50
49
51
```
50
-
`# Example spot interruption notification EventBridge rule`
@@ -113,17 +115,17 @@ SQS exposes a VPC Endpoint which will fulfill the isolated VPC use-case.
113
115
Dynamically creating the SQS infrastructure and EventBridge rules means that Karpenter’s IAM role would need permissions to SQS and EventBridge:
114
116
115
117
```
116
-
`"sqs:GetQueueUrl",`
117
-
`"sqs:ListQueues"``,`
118
-
`"sqs:ReceiveMessage"``,`
119
-
`"sqs:CreateQueue"``,`
120
-
`"sqs:DeleteMessage"``,`
121
-
`"events:ListRules",`
122
-
"`events:DescribeRule`",
123
-
"events:PutRule",
118
+
"sqs:GetQueueUrl",
119
+
"sqs:ListQueues",
120
+
"sqs:ReceiveMessage",
121
+
"sqs:CreateQueue",
122
+
"sqs:DeleteMessage",
123
+
"events:ListRules",
124
+
"events:DescribeRule",
125
+
"events:PutRule",
124
126
"events:PutTargets",
125
-
"`events:DeleteRule`",
126
-
`"events:RemoveTargets"`
127
+
"events:DeleteRule",
128
+
"events:RemoveTargets"
127
129
```
128
130
129
131
The policy can be setup with a predefined name based on the cluster name. For example, `karpenter-events-${CLUSTER_NAME}` which would allow for a more constrained resource policy.
@@ -144,7 +146,7 @@ The simplest option is to include [NTH IMDS mode](https://quip-amazon.com/EUgPAQ
144
146
145
147
**3B: Build a System Daemon (nthd)**
146
148
147
-
An option to transparently handle spot interruption notifications is to build a system daemon in a separate repo that performs the IMDS monitoring and triggers an instance shutdown when an interruption is observed. This would rely on K8s’ new [graceful shutdown](https://kubernetes.io/docs/concepts/architecture/nodes/#graceful-node-shutdown) feature which went beta in K8s 1.21.
149
+
An option to transparently handle spot interruption notifications is to build a system daemon in a separate repo that performs the IMDS monitoring and triggers an instance shutdown when an interruption is observed. This would rely on K8s’ new [graceful shutdown](https://kubernetes.io/docs/concepts/cluster-administration/node-shutdown/#graceful-node-shutdown) feature which went beta in K8s 1.21.
148
150
149
151
With graceful shutdown, the kubelet registers [systemd-inhibitor-locks](https://www.freedesktop.org/wiki/Software/systemd/inhibit/) to stop the shutdown flow until locks are relinquished, which in this case would be when the kubelet has drained pods off of the node. Two parameters were added to the kubelet to tune the drain timeouts: `shutdownGracePeriod` & `shutdownGracePeriodCriticalPods`
0 commit comments