I'm using Kubernetes to orchestrate my micro-services.
In my K8S cluster, I have CPU-Only instances and other instances with GPU.
I would like to know how could I force specific PODS to run on the instances without GPU?
Thank you
As explained here you can use taints and tolerations to ensure that some pods will not be scheduled on nodes with GPUs.
All nodes with GPU can be tainted like this:
kubectl taint nodes <nodename> hasgpu=true:NoSchedule
Now add the following to specs of pods - which need GPU. This will ensure that any pod which does not have this toleration will not go to an instance with GPU attached.
tolerations:
- key: "hasgpu"
operator: "Equal"
value: "true"
effect: "NoSchedule"
You can check out a detailed explanation and examples of taints and toleration in this blog
Though adding the toleration in YAML file is not as much clean and you can use an admission controller to dynamically add tolerations using an admission controller. This will add tolerations to pods which request specific resources such as GPU. You can find more details here, this solution is elegant but relatively more work.
Related
I am looking to find out if there is a way I can assign a specific Deployment to a specific node pool.
I am planning to deploy a big-size application using kubernetes. I am wondering if there is a way we can assign deployments to specific node pools. In other words, we have 3 types of services:
General services, low performance and low replica count
Monitor services, high I/O and high performance servers needed
Module services, most demanding services, we are aiming to allocate the biggest part of our budget for this.
So obviously we would like to best allocate nodes to specific deployments so no resources go wasted, for example low tier servers node pool X would be only utilized by General service deployments, high tier servers node pool Y would be only utilized by the monitor services, and the highest tier servers would only be utilized by the Module services.
I understand that there is a huge number of articles that talks about pod affinity and other related things, but what I seem to not be able to find anything that matches the following:
How to assign Deployment to specific node pool
Thanks in advance!
Another way (in addition to what Yayotrón proposed) would be to work with NodeAffinity and AntiAffinity. For more information check the official documentation here: https://kubernetes.io/docs/concepts/scheduling-eviction/assign-pod-node/
Taints and tolerations very strict and scheduling on other nodes would not be possible at all.
With Affinity and Antiaffinity you can specify wheter you want it to be strict (RequiredDuringSchedulingIgnoredDuringExecution) or a soft restriction (PreferredDuring....)
This can be achieved using Taints and Tolerations. A quick summary of what they are (from their documentation):
Node affinity, is a property of Pods that attracts them to a set of nodes (either as a preference or a hard requirement). Taints are the opposite -- they allow a node to repel a set of pods.
Tolerations are applied to pods, and allow (but do not require) the
pods to schedule onto nodes with matching taints.
Taints and tolerations work together to ensure that pods are not
scheduled onto inappropriate nodes. One or more taints are applied to
a node; this marks that the node should not accept any pods that do
not tolerate the taints.
Or simply by using NodeSelector
When you register a node to join the kubernetes cluster you can specify the taints and labels using kubelet --register-with-taints label=value --node-labels=label2=value2.
Or you can use kubectl taint for already registered nodes.
Then when you're going to deploy a pod/deployment/statefulset you can specify its nodeSelector and Tolerations
spec:
nodeSelector:
label2: value2
tolerations:
- key: "label"
operator: "Equal"
value: "value"
effect: "NoSchedule"
Here is what I am working with.
I have 3 nodepools on GKE
n1s1 (3.75GB)
n1s2 (7.5GB)
n1s4 (15GB)
I have pods that will require any of the following memory requests. Assume limits are very close to requests.
1GB, 2GB, 4GB, 6GB, 8GB, 10GB, 12GB, 14GB
How best can I associate a pod to a nodepool for max efficiency?
So far I have 3 strategies.
For each pod config, determine the “rightful nodepool”. This is the smallest nodepool that can accommodate the pod config in an ideal world.
So for 2GB pod it's n1s1 but for 4GB pod it'd be n1s2.
Schedule a pod only on its rightful nodepool.
Schedule a pod only on its rightful nodepool or one nodepool higher than that.
Schedule a pod only on any nodepool where it can currently go.
Which of these or any other strategies will minimize wasting resources?
=======
Why would you have 3 pools like that in the first place? You generally want to use the largest instance type you can that gets you under 110 pods per node (which is the default hard cap). The job of the scheduler is to optimize the packing for you, and it's pretty good at that with the default settings.
I would use a mix of Taints and Tolerations and Node affinity.
Taints and tolerations work together to ensure that pods are not scheduled onto inappropriate nodes. One or more taints are applied to a node; this marks that the node should not accept any pods that do not tolerate the taints. Tolerations are applied to pods, and allow (but do not require) the pods to schedule onto nodes with matching taints.
You can set a taint on a node kubectl taint nodes node1 key=value:NoSchedule
The taint has key key, value value, and taint effect NoSchedule. This means that no pod will be able to schedule onto node1 unless it has a matching toleration.
While you are writing a pod yaml you can specify PodSpec and add toleration which will match the taint created on node1 which will allow pod with either toleration to be scheduled onto node1
tolerations:
- key: "key"
operator: "Equal"
value: "value"
effect: "NoSchedule"
or
tolerations:
- key: "key"
operator: "Exists"
effect: "NoSchedule"
Taints and tolerations are a flexible way to steer pods away from nodes or evict pods that shouldn’t be running. A few of the use cases are
Dedicated Nodes: If you want to dedicate a set of nodes for exclusive use by a particular set of users, you can add a taint to those nodes (say, kubectl taint nodes nodename dedicated=groupName:NoSchedule) and then add a corresponding toleration to their pods (this would be done most easily by writing a custom admission controller). The pods with the tolerations will then be allowed to use the tainted (dedicated) nodes as well as any other nodes in the cluster. If you want to dedicate the nodes to them and ensure they only use the dedicated nodes, then you should additionally add a label similar to the taint to the same set of nodes (e.g. dedicated=groupName), and the admission controller should additionally add a node affinity to require that the pods can only schedule onto nodes labeled with dedicated=groupName.
Nodes with Special Hardware: In a cluster where a small subset of nodes have specialized hardware (for example GPUs), it is desirable to keep pods that don’t need the specialized hardware off of those nodes, thus leaving room for later-arriving pods that do need the specialized hardware. This can be done by tainting the nodes that have the specialized hardware (e.g. kubectl taint nodes nodename special=true:NoSchedule or kubectl taint nodes nodename special=true:PreferNoSchedule) and adding a corresponding toleration to pods that use the special hardware. As in the dedicated nodes use case, it is probably easiest to apply the tolerations using a custom admission controller. For example, it is recommended to use Extended Resources to represent the special hardware, taint your special hardware nodes with the extended resource name and run the ExtendedResourceToleration admission controller. Now, because the nodes are tainted, no pods without the toleration will schedule on them. But when you submit a pod that requests the extended resource, the ExtendedResourceToleration admission controller will automatically add the correct toleration to the pod and that pod will schedule on the special hardware nodes. This will make sure that these special hardware nodes are dedicated for pods requesting such hardware and you don’t have to manually add tolerations to your pods.
Taint based Evictions: A per-pod-configurable eviction behavior when there are node problems, which is described in the next section.
As for node affinity:
is conceptually similar to nodeSelector – it allows you to constrain which nodes your pod is eligible to be scheduled on, based on labels on the node.
There are currently two types of node affinity, called requiredDuringSchedulingIgnoredDuringExecution and preferredDuringSchedulingIgnoredDuringExecution. You can think of them as “hard” and “soft” respectively, in the sense that the former specifies rules that must be met for a pod to be scheduled onto a node (just like nodeSelector but using a more expressive syntax), while the latter specifies preferences that the scheduler will try to enforce but will not guarantee. The “IgnoredDuringExecution” part of the names means that, similar to how nodeSelector works, if labels on a node change at runtime such that the affinity rules on a pod are no longer met, the pod will still continue to run on the node. In the future we plan to offer requiredDuringSchedulingRequiredDuringExecution which will be just like requiredDuringSchedulingIgnoredDuringExecution except that it will evict pods from nodes that cease to satisfy the pods’ node affinity requirements.
Thus an example of requiredDuringSchedulingIgnoredDuringExecution would be “only run the pod on nodes with Intel CPUs” and an example preferredDuringSchedulingIgnoredDuringExecution would be “try to run this set of pods in failure zone XYZ, but if it’s not possible, then allow some to run elsewhere”.
Node affinity is specified as field nodeAffinity of field affinity in the PodSpec.
...
The new node affinity syntax supports the following operators: In, NotIn, Exists, DoesNotExist, Gt, Lt. You can use NotIn and DoesNotExist to achieve node anti-affinity behavior, or use node taints to repel pods from specific nodes.
If you specify both nodeSelector and nodeAffinity, both must be satisfied for the pod to be scheduled onto a candidate node.
If you specify multiple nodeSelectorTerms associated with nodeAffinity types, then the pod can be scheduled onto a node only if all nodeSelectorTerms can be satisfied.
If you specify multiple matchExpressions associated with nodeSelectorTerms, then the pod can be scheduled onto a node if one of the matchExpressions is satisfied.
We are having problem with several deployments in our cluster that do not seem to be working. But I am a bit apprehensive in touching these, since they are part of the kube-system namespace. I am also unsure as what the correct approach to getting them into an OK state is.
I currently have two daemonsets that have warnings with the message
DaemonSet has no nodes selected
See images below. Does anyone have any idea what the correct approach is?
A DaemonSet is creating a pod in each node of your Kubernetes cluster.
If the Kubernetes scheduler cannot schedule any pod, there are several possibilities:
Pod spec has a too high memory requests resource for the memory node capacity, look at the value of spec.containers[].resources.requests.memory
The nodes may have a taint, so DaemonSet declaration must have a toleration (kubernetes documentation about taint and toleration)
The pod spec may have a nodeSelector field (kubernetes documentation about node selector)
The pod spec may have an enforced node affinity or anti-affinity (kubernetes documentation about node affinity)
If Pod Security Policies are enabled on the cluster, a security policy may be blocking access to a resource that the pod needs to run
There are not the only solutions possible. More generally, a good start would be to look at the events associated to the daemon set:
> kubectl describe daemonsets NAME_OF_YOUR_DAEMON_SET
I have added podAntiAffinity to my DeploymentConfig template.
However, pods are being scheduled on nodes that I expected would be excluded by the rules.
How can I view logs of the kubernetes scheduler to understand why it chose the node it did for a given pod?
PodAntiAffinity has more to do with other pods than nodes specifically. That is, PodAntiAffinity specifies which nodes to exclude based on what pods are already scheduled on that node. And even here you can make it a requirement vs. just a preference. To directly pick the node on which a pod is/is not scheduled, you want to use NodeAffinity. The guide.
DaemonSet is a Kubernetes beta resource that can ensure that exactly one pod is scheduled to a group of nodes. The group of nodes is all nodes by default, but can be limited to a subset using nodeSelector or the Alpha feature of node affinity/anti-affinity.
It seems that DaemonSet functionality can be achieved with replication controllers/replica sets with proper node affinity and anti-affinity.
Am I missing something? If that's correct should DaemonSet be deprecated before it even leaves Beta?
As you said, DaemonSet guarantees one pod per node for a subset of the nodes in the cluster. If you use ReplicaSet instead, you need to
use the node affinity/anti-affinity and/or node selector to control the set of nodes to run on (similar to how DaemonSet does it).
use inter-pod anti-affinity to spread the pods across the nodes.
make sure the number of pods > number of node in the set, so that every node has one pod scheduled.
However, ensuring (3) is a chore as the set of nodes can change over time. With DaemonSet, you don't have to worry about that, nor would you need to create extra, unschedulable pods. On top of that, DaemonSet does not rely on the scheduler to assign its pods, which makes it useful for cluster bootstrap (see How Daemon Pods are scheduled).
See the "Alternative to DaemonSet" section in the DaemonSet doc for more comparisons. DaemonSet is still the easiest way to run a per-node daemon without external tools.