Scalability on Digi Hunch

Kubernetes Platform as a Service and Red Hat OpenShift

Sun, 25 Jun 2023 11:10:15 -0400

The Three-layer model

Kubernetes is so complex that it becomes a buzz word itself. I categorize the related work into three layers: a cluster layer, a platform layer and an application layer, by their purposes. The three layers are illustrated as below:

Let’s examine each layer in this model and where the Kubernetes Platform as a Service fits in.

The Kubernetes Cluster Layer

At the bottom, the Kubernetes Cluster layer is the foundational layer. It focus on using self-hosted VMs or cloud resources to build a functional Kubernetes cluster and worker node groups. A functional cluster includes a highly available control plane, as well as scalable node groups that all communicate with the control plane. Cloud Service Providers like AWS and Azure provides managed Kubernetes service, which takes away the complexity (and flexibility as well) of managing control plane components such as etcd store and API server. The managed services also automatically provisions computing nodes and join them into the cluster. The cluster layer may also involve integration with of CNI and CSI, to ensure Pod-to-Pod communication and available storage classes.

Professionals working at this layer are infrastructure experts who understand networking, storage, as well as how to manage cloud resources or VMs, infrastructure as code. On a daily basis, they deal with VPCs/V-Nets, subnets, EBS/Azure Disk, File storage, EC2/Azure VMs, etc. When the team is doing a bad job at this layer, you might see symptoms like unresponsive cluster API, orphaned worker nodes, or kubectl failing to connect to cluster endpoint.

The tenants (applications) of the Kubernetes platform does not directly interact with this layer. If you decide to switch CSP vendor, this layer requires 100% re-engineering because the managed Kubernetes service by each CSP is different.

The Kubernetes Platform Layer

The Platform layer sits in the middle. When organization decides to adopt Kubernetes, they often underestimate the efforts required in this layer. This layer works on a functional cluster, without directly interacting with the underlying cloud resources. This layer involves any Kubernetes abstractions that do not creates tangible business value. Rather, this layer is an enabler. It allows the applications to deploy smoothly, evolve quickly, and more importantly, focus on the business.

Teams working on this layer needs to be Kubernetes experts. On a daily basis, they play with common CNCF toolings, such as Prometheus, ArgoCD, Istio, Cilium, Tekton, Open Policy Agent, etc. They are comfortable with Operators, Helm Charts, Ingress, etc. Inside of the Kubernetes cluster, they also manage the foundational services such as Event streaming (e.g. Kafka), PostgreSQL database (e.g. PostgreSQL), software-defined storage (e.g. Ceph), service mesh (e.g. Istio), Authentication (e.g. Keykloak) , etc. These services act as the infrastructure layer to the business workload. If the team is doing a bad job, you would see data loss with database, observability service not populating data, ingress does not process request, etc.

The tenants (application) share services in this layer. If you decide to switch CSP vendor. I estimate 80% of the work at this layer is portable, and 20% requires re-engineering. That is because each CSP offers different external resources, therefor the low level Kubernetes objects in this layer, such as storage classes, load balancers, supported CNIs are different. High level objects such as Kafka remains portable across platforms.

The Application Layer

The next layer at the top is application layer. Workloads in this layer are directly linked to the business value. The applications are very diverse. Most of the time, the release team is the main player at this layer. If the organization develops its own application, the software development team also work at this layer.

In terms of knowledge, the members of development team are experts in software engineering, and Software Development Life Cycle (SDLC), etc. On a daily basis, they deal with programming languages, product development, build and release. If they screw up their work, expect business errors, such as orders sent to wrong client, incorrect balance sheet, etc. This team has high visibility in the organization due to its direct link to business value.

This layer of work involves multiple tenants. Each tenant is isolated within their own namespace. When you switch CSP vendor, this layer should be readily portable with minimal effort.

It is also worth noting that, with solid platform and cluster layers, the team working at this layer do not write bespoke code for networking, observability, authentication and authorization, encryption and many other aspects not relevant to the core business. Once deployed, the application services are resilient, scale to demands, and cost efficient. This layer reaps the benefits of Kubernetes.

Kubernetes Platform as a Service

As the Kubernetes dust is still settling, a builder’s title may not always reflect which layer she or he focuses on. Today it is pretty common for infrastructure engineers to expand their role into the platform layer, or likewise, a software engineer to drill down to the platform layer.

The boundary between platform layer and cluster layer is clear. The cluster layer deals with underlying infrastructure, either in the cloud or on premise. They abstract away the complex infrastructure world from those working with the platform layer.

The boundary between platform layer and application layer is a little tricky to articulate. The application layer focuses on implementing the business logics. The platform layer takes care of the functions that are not part of business logic but essential to the business application. Take an HTTP request for example, application developer should not have to write code to terminate TLS (not part of business logic). They should only write the code to process the HTTP request (business logic). TLS termination is delegated to an Ingress, to be configured by platform builders.

The folks working at the Platform layer needs to interface with both sides. They provide Platform as a Service to the Application teams. However, their work appears mostly invisible in an organization. Their effort is oftentimes underestimated. There are several reasons for that. First, the platform layer does not directly create tangible business value. They are just someone else’s enabler. Second, their building blocks involve a lot of abstractions by Kubernetes API. Third, the idea of platform engineering is newly emerged. There hasn’t been a populous recognition of its value.

Red Hat OpenShift Container Platform

The platform team builds the platform with their choice of open-source tools. For clusters using OpenShift Kubernetes Engine, Red Hat introduces Open Shift container platform consisting of Red Hat’s opinionated (but validated) choice of toolings, for example:

OpenShift Service Mesh: Istio
OpenShift Streams: Apache Kafka
OpenShift GitOps: ArgoCD
OpenShift Container Platform Pipelines: Tekton
OpenShift Serverless: Knative
OpenShift Data Foundation: Ceph

Clients building their clusters with OpenShift Kubernetes Engine may build their own platform with the toolings in the OpenShift enterprise Kubernetes container platform. For more services, check out the documentation for OpenShift Container Platform. For customers with OpenShift Kubernetes Engine, their options to DIY platform are:

Entry-Level: Red Hat OpenShift Kubernetes Engine: Enterprise Kubernetes distribution on RHEL CoreOS
Mid-Level: Red Hat OpenShift Container Platform (RHOCP):
Plus-Level: Red Hat OpenShift Platform Plus: RHOCP + advanced cluster management, security, data management essentials, enterprise container registry

OpenShift runs the business model of Kubernetes PaaS.This is a unique business model that I do not find a matching competitor. Even if you choose to DIY your own platform, the Red Hat’s choices are still a great reference. The OpenShift enterprise Kubernetes container platform maps perfectly to the platform layer of the three-layer model, aiming to simplify the work in the platform layer.

Managed RedHat OpenShift

At first, the OpenShift container platform started as a value add-on to the Kubernetes Engine. Now it’s a separate product line in their business model. In the mean time, OpenShift partners with major CSPs, to develop the cloud service editions, including:

Red Hat OpenShift on AWS (ROSA)
Microsoft Azure Red Hat OpenShift (ARO)
Red Hat OpenShift Dedicated – on AWS and GCP
Red Hat OpenShift on IBM Cloud

These offerings are managed Kubernetes Platform as a Service in the cloud. Since RedHat is the only player in this model, we can refer to them as managed OpenShift services. In addition to an already-confusing world of Kubernetes platform portfolios, these offerings gives consumers even more options. On AWS for example, users have the following options:

Managed Platform: OpenShift Dedicated, managed by Red Hat
Managed Platform: Red Hat OpenShift Service on AWS (ROSA), managed by Red Hat and AWS
Self-built cluster: OpenShift Container Platform

This video discussed more details about these options, such as support model. It is also worth noting that these options tend to be much pricier than managed clusters such as EKS and AKS.

Since a Managed RedHat Platform makes it easy to deploy, let’s take ROSA as an example and create a cluster. To enable ROSA in AWS console, click on “Getting Started”. The next page ensures ROSA is enabled and checks other prerequisite such as meeting service quotas and creating ELB service-linked role, as show below:

Now, with an AWS account (and ROSA enabled), a RedHat account, and the rosa-cli utility, we can create a cluster with just a few commands. As a note, be wary of the cost and do not forget to delete the cluster afterwards.

Create a ROSA cluster

With the following set of commands, we can kick off cluster creation, using STS. We can bring our own VPC, so long as it meets certain prerequisites. I use the Terraform template in the vpc-base project, to create the underlying VPC. We’ll need the followings from this template:

The CIDR range of the VPC: as input with a default
The subnet Ids of the private subnet to place, printed in the output

The subnets are private subnets, because we want to provision the cluster with private node and private endpoint. When we use rosa CLI, we provide the CIDR and subnet IDs.

# start with AWS cli configured to the correct profile
rosa login # with redhat account and past token
rosa create account-roles --mode auto -y # this command creates the IAM roles ManagedOpenShift-*-Role, with RedHat account as trust entity
rosa verify permissions # optional
rosa verify quota # optional

export ROSA_CLUSTER_NAME="dhc" \
OPENSHIFT_VERSION=4.13.4 \
AWS_ACCOUNT_ID=$(aws sts get-caller-identity --query Account --output text) \
AWS_DEFAULT_REGION=us-east-1

rosa create cluster --sts --private \
 --cluster-name $ROSA_CLUSTER_NAME \
 --multi-az \
 --private-link \
 --region $AWS_DEFAULT_REGION \
 --version $OPENSHIFT_VERSION \
 --enable-autoscaling \
 --min-replicas 3 \
 --max-replicas 3 \
 --compute-machine-type m5.xlarge \
 --machine-cidr 147.206.0.0/16 \
 --subnet-ids subnet-052852a1fb4d7d2ad,subnet-06d8d40ae39d55c47,subnet-0f67ce08bc588012c

The CLI will pick up the correct VPC by CIDR, and prompt you to confirm creation of private cluster. After the command kicks off, it will wait for OIDC provider creation, and role creation. Then it uses a Terraform template to create the related resources including VPC. Use this command to check status:

rosa list clusters
rosa describe cluster -c dhc

When the second command displays the state of waiting (Waiting for OIDC configuration), we can create OIDC provider:

rosa create operator-roles -c $ROSA_CLUSTER_NAME --mode auto --yes
rosa create oidc-provider -c $ROSA_CLUSTER_NAME --mode auto --yes

Throughout the process, we can monitor the install log (terraform output) with:

rosa logs install -c dhc --watch

In the log, you might see errors with terminals connecting to the terraform backend, which doesn’t necessarily indicate a cluster creation error. Always check the cluster state until it reports success.

Kick the tires

Eventually the describe cluster command will show ready state. We can now create an admin user:

rosa create admin -c $ROSA_CLUSTER_NAME

The command above prints an oc command (OpenShift CLI, equivalent to kubectl) with password to log in. Let’s examine the cluster with oc. Because it is a private cluster, the endpoint is not available publicly. However, it is accessible from the Bastion host. Use the SSM Session Manager technique from my previous post to SSH to the Bastion Host, which should have oc installed. To install oc yourself, use HomeBrew on Mac. On Linux or Windows, log on to OpenShift console, go to Downloads on the left pannel and find it out under CLI tools.

The oc command may report insecure TLS on the login URL. Wait for a few minutes for the certificate to come off as safe. Once you run the oc command with password, it should return “Login successful” and then we can connect to the cluster:

$ oc get node # or kubectl get node
NAME STATUS ROLES AGE VERSION
ip-147-206-135-41.ec2.internal Ready,SchedulingDisabled infra,worker 3m5s v1.26.5+7d22122
ip-147-206-155-141.ec2.internal Ready control-plane,master 25m v1.26.5+7d22122
ip-147-206-156-81.ec2.internal Ready worker 19m v1.26.5+7d22122
ip-147-206-164-21.ec2.internal Ready infra,worker 3m3s v1.26.5+7d22122
ip-147-206-179-90.ec2.internal Ready worker 19m v1.26.5+7d22122
ip-147-206-191-118.ec2.internal Ready,SchedulingDisabled control-plane,master 26m v1.26.5+7d22122
ip-147-206-192-232.ec2.internal Ready worker 19m v1.26.5+7d22122
ip-147-206-193-198.ec2.internal Ready infra,worker 3m20s v1.26.5+7d22122
ip-147-206-218-114.ec2.internal Ready control-plane,master 26m v1.26.5+7d22122

You can use oc the same way you’d use kubectl. Both works through SOCK5 proxy. In the meantime, log in to the RedHat console with your Red Hat credential, you can see the cluster in Ready state as well:

The rosa create admin command creates a htpasswd type (username-password) of identity provider (IdP) with a user named cluster-admin and a preset password. In real life however, we often configure third party IdP with OIDC integration. I’ll have to leave this to the next blog post.

We shall see the nodes as EC2 instances from AWS console as well. Note that there are three roles: control-plane, worker and infra. The infra nodes are for infrastructure services. These services (Ingress Controller, GitOps, Pipeliens) are the ones in the platform player as we discussed above.

There are many customizations you can make in this installation process and I’d have to defer to the ROSA documentation. To clean up, use the following ROSA command:

rosa remove cluster -c dhc

The output also gives you the command to delete operator roles and OIDC provider, for example:

rosa delete operator-roles -c 23o4u3j98tqmlbtjo612opb7a4bbim5f --mode auto --yes
rosa delete oidc-provider -c 23o4u3j98tqmlbtjo612opb7a4bbim5f --mode auto --yes

Then we can destroy the VPCs using terraform.

ROSA with HCP

Update Oct 2023:

The deployment above provisioned a few nodes for control plane, which add to the overall time to provision a cluster. In Aug 2023, there is a new option Hosted Control Plane (HCP) that came to allow users to provision a hosted control plane. This results in cost savings and shorter time to provision a cluster. Here is a table of comparison between the ROSA with HCP and ROSA classic.

Final words

In this post, I discussed the three-layer model and pointed out that platform layer isn’t as visible as the other two. I also experimented ROSA as a turn-key Kubernetes platform with its opinionated stack of services.

Some misinformed organizations even skip the entire platform layer in their estimate of effort. They build a cluster, ran a hello-world service and assumes they can start putting applications on the Kubernetes cluster. There are also customers who purchased the entire Managed OpenShift platform but only use it as a cluster. Yikes!

The concept of Kubernetes platform, or generally platform engineering is still spreading. The consulting team that I worked in full-time last year re-branded itself as platform engineering. Marketings are pushing it. Builders are doing it. We’ll keep an eye, on whether customers are buying it.

Computing services: from PaaS to Serverless

Fri, 21 Oct 2022 19:31:00 -0400

Silicon Valley startups in mid-2000s likely do not run their own IT operations (i.e. renting their own data centre spaces, purchasing their own rack-mounted servers). Since the launch of EC2, AWS has been renting extra computing capacity to those startups, in the IaaS model. The leased infrastructure requires maintenance work, and AWS realized that many of these customers cannot afford specialized database admins, network admins, storage admins, or even server admins. As a result, they created a handful of managed services aiming to cut out admin overhead and let their customer focus on coding. This is how Platform-as-a-service (PaaS) came about.

Let’s take a look at what are exactly operation activities.

Ops activities

IT operation team manages server provisioning, installation of operating system, tuning performance, configuring auto scaling and load balancing, configure networking and storage systems, etc. Networking can be so complex that many infrastructure teams have a dedicated Network Operation Center (NOC), who along with security team, manages key aspects of networking, such as segmentation, router configuration, load balancing, firewall configuration.

For client-server application, the client-side code will make outgoing connections, utilizing the TCP/IP stack on the host through an ephemeral port. The server-side code has to be wrapped as a service. A daemon ensures the process running this service stays up and listens to a TCP port in order to respond to request by invoking the functions. Application team usually assumes these activities.

If database is involved, then the patching, upgrade, replication, data protection are all Ops problems. If storage is involved, then Ops has to manage mass data accumulated over years, the integration between storage and database and applications, performance, replication, etc. Some larger organizations have full-time database administrator and storage administrators.

Then comes container. Containers have their benefits but it increases the operation overhead by an order of magnitude. Running container application at scale warrants its own platform, most likely a Kubernetes platform, to address all of the problems above again at the cluster level. Some organization created platform team to manage container and VM platforms.

It is the Ops, that turns functional code into a running business. It is also the Ops, that becomes a pain point as a startup scales. With IaaS and PaaS models, AWS managed to convince many small businesses to delegate their IT operations to AWS. This is the humble start of cloud computing.

Elastic Beanstalk

At first, I wasn’t too impressed with Elastic Beanstalk, since it abstracts away too many details. However, I later realized that it has been surprisingly popular in the developer community, especially with individual developers and SMBs. It simplifies deployment to the point that their users don’t need to know other AWS services, allowing them to focus on coding application logic.

You configure Applications and Environments (one application may have multiple environments). In the Environment layer, you can specify code platform (e.g. Python 3.8 on 64bit Amazon Linux 2, Java, Go, PHP, Ruby) and even container platform (Docker on EC2 or ECS). Behind the scene, Elastic Beanstalk configures EC2 instances, Elastic Load Balancers, etc on the selected VPC and integrate with logging and monitoring services. In the console, Elastic Beanstalk exposes a list of configurations options (e.g. AMI, instance type). This centralized configuration page is dummied down for those who don’t want to deal with Ops.

The downside of Elastic Beanstalk is it takes away a lot of flexibility. Many developers find Elastic Beanstalk limit their choices of deployment, as their applications scale. Elastic Beanstalk does not suit for applications that demand extensive operation efforts. Its niche market is individual developers and SMB. Few enterprise applications run on Elastic Beanstalk.

Containerization with ECS and EKS

When we containerize an application, we build container images. Then we run these images with container runtimes, is a core feature of container platform. Container platform also provides orchestration engine since we frequently take containers up and down. In addition, container platform provides mechanisms for container networking and storage.

AWS has a couple options for container platform. ECS (Elastic Container Service) came out earlier. It organizes a group of EC2 instances as a cluster. You can manage autoscaling, networking, and persistent storage (EFS, FSx etc) on ECS. EKS (Elastic Kubernetes Service) is the managed Kubernetes service by AWS. Just like AKS, it provides a managed control plane along with computing nodes.

I see ECS as a proprietary and simplified container platform, and Kubernetes as an open-source standard for full-fledged container platform with an entire ecosystem. EKS includes an upstream-certified Kubernetes distribution with a set of tools specific to AWS. Since Kubernetes is the de-facto standard container platform, I prefer EKS by default, unless I can justify the use of ECS. In fact, ECS and Kubernetes have many concepts in common. For example, a “Task” in ECS is equivalent to a Pod in Kubernetes.

Whether it is ECS or EKS, right-sizing the computing node group is always challenging especially when the application traffic load is irregular. AWS Fargate is a technology that provides on-demand, right-sized compute capacities. It works with ECS and EKS. When integrated with EKS, we delegate the node management (e.g. scaling) to Fargate and forget about sizing the node pool.

Using ECS and Fargate involves quite a bit of configurations. To simplify that, we can use App Runner, which builds ECS cluster and uses Fargate to execute the container behind the scenes. App Runner helps client in a way similar to Elastic Beanstalk, but concentrate on Container workload.

Serverless with Lambda and API Gateway

The services above have their limitations when it comes to scaling capability. First, they cannot scale to zero. You still pay for idling resources. Also, it is not easy to find the optimal autoscaling setting. Lambda and API Gateway together solves these challenges. AWS refers to it as serverless, which has since become a buzzword. To understand what it is, let’s examine two concepts:

Function as a Service: service with the ability to execute code on demand. Users only pay for code execution time and do not care where the underlying runtime is
Backend as a Service: service with the ability to listen to a port and respond to web request

Lambda itself is a function as a service. Triggered by events, it only incurs a charge when it’s invoked. It does not stay up and listening to a TCP port for incoming web request, as does a backend service. In order to act as a backend, Lambda needs to pair up with API gateway. In this configuration, API gateway listens to a web request, and it fires an event to trigger the execution of Lambda function. Lambda and API gateway together makes a backend as a service. In AWS, the coupling of API gateway and Lambda function ensures an idle service does not incur computing cost.

Since Lambda supports many types of events as trigger, it is also used in event-driven architecture, either standalone or from a VPC. Under the hood, Lambda runs code in a container (with a quick startup time relative to a VM).

Developers can release Lambda code by uploading zip package to S3 bucket, or just packaging code into container image. For deployment, apart from AWS console and CLI, one can leverage CloudFormation, SAM (serverless application model), or CDK.

Lambda vs Fargate

Both Lambda and Fargate are serverless capabilities, at least from a marketing perspective. Both can be used to back web service but there are differences. They provision computing resource at different granularity.

In a web service, the execution duration of a Lambda function is the response duration to an API request, in terms of seconds. While the server is waiting for a request, there is no usage of the computing resource so you’re not paying for waiting for a request. However, this also creates the delay of cold-start, especially when the code size is large. There are several ways to optimize the cold start (e.g. SnapStart for Java), but none of those can completely get rid of the cold-start delay with a once-after-a-while request. A light GET call could take 5 seconds with cold start.

On the Fargate side, the resource provisioning is based on container lifecycle, instead of request lifecycle. As a result, you’re still paying for wait time, and it is not per-request billing. Since the container remains up, your request is not going to experience the cold-start if it’s been idle for a while. Although, Fargate saves you from the effort to right-sizing the computing nodes for container execution, it is not quite the idea of “scale-to-zero when idle” by itself.

Serverless Architecture

In the white paper AWS Serverless Multi-Tier Architectures with Amazon API Gateway and AWS Lambda, AWS advocates the serverless architecture as a modern alternative to the traditional widely adopted three-tier architecture (presentation, logic and data tiers). In the three tier architecture, the scalability of three tier are managed separately. The modern serverless architecture that AWS whitepaper proposes uses API Gateway and Lambda function in place of Load Balancer and EC2 instances (e.g. in an Auto Scaling Group), as illustrated below:

Both API Gateway and Lambda scale automatically to support the need of application workload. It assumes the role of logic tier in three-tier architecture but requires minimal maintenance work. For presentation tier, AWS has serverless alternatives such as CloudFront, S3. For data tier, AWS has serverless alternatives such as Amazon Aurora for relational database and DynamoDB for NoSQL. However, the “no request, no pay” model for Lambda does not apply to the data tier in serverless architecture.

Whilst this paradigm benefits small shop IT who wants to minimize infrastructure cost, it has downsides. There is no ability for infrastructure optimization. Since you do not manage where the code runs, client may have concerns over security (e.g. multi-tenant runtime). As business grows, keep using Lambda can result in technology lock-in. Also, a less used application usually requires warm-up time. A code start (downloading the code and preparing the environment behind the scene) can take 100ms to over a second.

Conclusion

PaaS attempts to help startups simplify the “grunt work” of IT operation. Serverless takes it even further. The semantics of serverless computing is confusing and the Wikipedia page acknowledges it as a misnomer. The nature of serverless model, is the cloud users delegate server capacity management to cloud platforms. The users don’t need to manage servers, VMs, instances, containers, etc on their own. In a previous post, I discussed the ability to scale to zero, which is just one of the many enabling technologies of serverless. Also, “no request, no pay” is neither an inherent nature of serverless model. Serverless service may involve storage (e.g. data service, S3, Aurora serverless) which incurs storage cost. There is a whitepaper on choosing the right AWS service to deploy your website or web application, with a decision tree.

AWS pioneered serverless with Lambda release in 2014 but competitor follows. In the Azure landscape, there is an entire suite of computing services from virtual machine to serverless (also with a decision tree in documentation). Azure’s counterpart for serverless architecture is Azure Function (released in 2016 for GA) with API Management. As for GCP, the serverless suite includes the event-driven Cloud Function (introduced in 2017) and Knative-based FaaS Cloud Run (introduced in 2019).

There are voices in advocacy of standardization of serverless model, and CNCF had since made minuscule efforts such as CloudEvents. The status quo, unfortunately, is anything but standardized.

Lastly, here is a table that summarizes the pros and cons of each computing service model.

Computing Service Model	Pro	Con
EC2	– Most straightforward and widespread legacy model – Legacy	– Ops tasks can be heavy (e.g. patch and vulnerability management of OS) – Utilization can be low
ECS	– Container orchestration is managed – Convenient to scale – Well integrated with other AWS services	– Limited advanced features – Vendor lock-in
EKS	– Highly scalable and flexible – Advanced, platform-neutral deployment tools available (e.g. Helm, ArgoCD, etc) – Custom configurations (e.g. operators)	– Significant operation overhead – Steep learning curve (especially for teams)
Lambda	– automatically scale – low ops overhead – pay per use	– limited choices of runtime – subject to latency due to cold start; yet warm start incurs cost – not suitable for long running tasks (batch processing jobs etc)

In summary, the Lambda-based serverless model is good for stateless server-side workload with short response time (<15s), tolerance of cold-start, no need for portability across platforms, and no complex package dependency.

Kubernetes Storage on Azure 3 of 3 – Ceph by Rook

Fri, 26 Aug 2022 19:43:00 -0400

In the last two posts, I covered the native storage options on Azure Kubernetes Service, as well as Portworx as an example of a proprietary Software Defined Storage (SDS) solution. There are also a number of open-source alternative SDS solutions. Ceph has nearly a decade of history from prior to containerization, and is the most widely adopted storage platform. In this post, we continue to explore Ceph as an open-source storage solution on Azure Kubernetes.

Ceph by Rook

Ceph is an open-source SDS platform for distributed storage on a cluster and provides object, block and file storage. Installation of Ceph SDS can be complex, especially on Kubernetes platform. Rook is a graduated CNCF project to orchestrate storage platform. Rook by itself is not SDS and it supports:

Ceph: configure a Ceph cluster. Think of this as the equivalent of cephadm on Kubernetes platform.
NFS: configure an NFS server. Think of this as the equivalent of nfsd daemon on Kubernetes platform.
Cassandra: an operator to configure a Cassandra database cluster. It is now deprecated.

We play with Rook Ceph. I also refer to it as Ceph by Rook. The contribution of Rook project is it simplifies the installation as a matter of declaring custom resources using CRDs. Here are some high-level CRDs to know:

CephCluster: creates a Ceph storage cluster
CephBlockPool: represents a block pool
CephFilesystem: represents a file system
CephObjectStore: represents an object store
CephNFS: spins up a NFS Ganesha server to export NFS shares of a CephFilesystem or CephObjectStore.

As with typical Kubernetes resources in controller pattern, Ceph by Rook needs an operator along with custom resources. We can use YAML manifest for both of them, and the manifests are usually very tediously long. We can also use Helm to install both of them, by providing a value file. Now we will install Ceph on AKS.

Install Ceph Operator on AKS

The steps are influenced by two relevant posts (here and here). However, I’ve incorporated the cluster configuration in the Azure directory of the cloudkube project, a modular Terraform template to configure AKS cluster and facilitate storage configuration. The node group and instance sizes are selected to be just enough to run a ceph POC cluster with minimum cost. One of the node groups is tainted with storage-node, as if the following command were run:

kubectl taint nodes my-node-pool-node-name storage-node=true:NoSchedule

You will only need to taint the nodes with the command above if you choose not to use the cloudkube template. The taint ensures that only Pods with corresponding toleration and effect can be scheduled to those nodes.

We use Helm to install Rook Operator. We need a value file (e.g. rook-ceph-operator-values.yaml) with content as below:

# https://github.com/rook/rook/blob/master/Documentation/Helm-Charts/operator-chart.md
crds:
 enabled: true
csi:
 provisionerTolerations:
 - effect: NoSchedule
 key: storage-node
 operator: Exists
 pluginTolerations:
 - effect: NoSchedule
 key: storage-node
 operator: Exists
agent:
 # AKS: https://rook.github.io/docs/rook/v1.7/flexvolume.html#azure-aks
 flexVolumeDirPath: "/etc/kubernetes/volumeplugins"

Then we install the operator with Helm:

helm install rook-ceph-operator rook-ceph --namespace rook-ceph --create-namespace --version v1.9.6 --repo https://charts.rook.io/release/ --values rook-ceph-operator-values.yaml
kubectl -n rook-ceph get po -l app=rook-ceph-operator

After installing the operator, we check the Pod status to make sure it is running. Then we can install the actual Ceph Cluster in one of the two ways. We can declare a CephClusterCRD ourself, or we can use Helm again to declare the CRD. Helm Chart gives us a lot of useful default values and saves us from editing a large body of YAML manifest.

Install Ceph CR on AKS

We use Helm to install CephCluster CRD. We create a value file (e.g. rook-ceph-cluster-values.yaml) with content as below:

# https://github.com/rook/rook/blob/master/Documentation/Helm-Charts/ceph-cluster-chart.md
operatorNamespace: rook-ceph
toolbox:
 enabled: true
cephObjectStores: [] # by default a cephObjectStore will be created. Setting this to null disables it
#cephBlockPools: # by default a cephBlockPool will also be created with default values
#cephFileSystems: # by default a cephFileSystem will also be created with default values

cephClusterSpec:
 mon:
 count: 3
 volumeClaimTemplate:
 spec:
 storageClassName: managed-premium
 resources:
 requests:
 storage: 10Gi
 resources:
 limits:
 cpu: "500m"
 memory: "1Gi"
 requests:
 cpu: "100m"
 memory: "500Mi"
 dashboard:
 enabled: true
 storage:
 storageClassDeviceSets:
 - name: set1
 # The number of OSDs to create from this device set
 count: 3
 # IMPORTANT: If volumes specified by the storageClassName are not portable across nodes
 # this needs to be set to false. For example, if using the local storage provisioner
 # this should be false.
 portable: false
 # Since the OSDs could end up on any node, an effort needs to be made to spread the OSDs
 # across nodes as much as possible. Unfortunately the pod anti-affinity breaks down
 # as soon as you have more than one OSD per node. The topology spread constraints will
 # give us an even spread on K8s 1.18 or newer.
 placement:
 topologySpreadConstraints:
 - maxSkew: 1
 topologyKey: kubernetes.io/hostname
 whenUnsatisfiable: ScheduleAnyway
 labelSelector:
 matchExpressions:
 - key: app
 operator: In
 values:
 - rook-ceph-osd
 tolerations:
 - key: storage-node
 operator: Exists
 preparePlacement:
 tolerations:
 - key: storage-node
 operator: Exists
 nodeAffinity:
 requiredDuringSchedulingIgnoredDuringExecution:
 nodeSelectorTerms:
 - matchExpressions:
 - key: agentpool
 operator: In
 values:
 - storagenp
 topologySpreadConstraints:
 - maxSkew: 1
 # IMPORTANT: If you don't have zone labels, change this to another key such as kubernetes.io/hostname
 topologyKey: topology.kubernetes.io/zone
 whenUnsatisfiable: DoNotSchedule
 labelSelector:
 matchExpressions:
 - key: app
 operator: In
 values:
 - rook-ceph-osd-prepare
 resources:
 limits:
 cpu: "500m"
 memory: "4Gi"
 requests:
 cpu: "500m"
 memory: "2Gi"
 volumeClaimTemplates:
 - metadata:
 name: data
 spec:
 resources:
 requests:
 storage: 100Gi
 storageClassName: managed-premium
 volumeMode: Block
 accessModes:
 - ReadWriteOnce

During the cluster provisioning, there will be a number of preparing Pods. We want those Pods to run on nodes with label agentpool=storagenp. In real life, we need to orchestrate where to run each workload, by restricting the nodes to schedule certain types of workload.

Then we can install the cluster using Helm:

helm install rook-ceph-cluster rook-ceph-cluster --namespace rook-ceph --create-namespace --version v1.9.6 --repo https://charts.rook.io/release/ --values rook-ceph-cluster-values.yaml

After running the Helm install, it may take as long as 15 minutes for all resources to settle. Watch the Pod status in rook-ceph namespace. At the end, make sure that the cluster is created successfully:

kubeadmin@pro-sturgeon-bastion-host:~$ kubectl -n rook-ceph get CephCluster
NAME DATADIRHOSTPATH MONCOUNT AGE PHASE MESSAGE HEALTH EXTERNAL
rook-ceph /var/lib/rook 3 15m Ready Cluster created successfully HEALTH_OK
kubeadmin@pro-sturgeon-bastion-host:~$ kubectl -n rook-ceph get cephBlockPools
NAME PHASE
ceph-blockpool Ready
kubeadmin@pro-sturgeon-bastion-host:~$ kubectl -n rook-ceph get cephFileSystems
NAME ACTIVEMDS AGE PHASE
ceph-filesystem 1 20m Ready

In my case it took 15 minutes before the cluster comes up as created successfully. You should notice that two storage classes were also created as a part of the install. It however did not create a storage class or CRD for object storage, because we explicitly disabled it in the Helm value file by setting cephObjectStores value to null.

Dashboard

We enabled dashboard. To configure the dashboard view properly, we would need an ingress. For a quick view here, we can play port forwarding tricks. First we fetch the admin password for use in the next step. Then expose the dashboard to the bastion host:

$ kubectl -n rook-ceph get secret rook-ceph-dashboard-password -o jsonpath='{.data.password}' | base64 -d
$ kubectl -n rook-ceph port-forward svc/rook-ceph-mgr-dashboard 8443:8443

Since I don’t have UI on the bastion host, I use the port forwarding trick again from my own MacBook. Start a new terminal and SSH to the bastion host with port-forwarding switch:

$ ssh -L 8443:localhost:8443 kubeadmin@20.116.132.8

The command above suppose the public IP of the bastion host is 20.116.132.8. Then from my MacBook I can browse to localhost:8443 (with Safari browser which gives me the option to bypass certificate error). At the web portal, provide username (admin) and password (as retrieved above):

Ceph console for Kubernetes

Apart from the dashboard, we can also use ceph admin tool from a toolbox pod, following this instruction. For monitoring, Ceph by Rook can expose metrics for Prometheus to scrape.

Performance

With default ceph configuration on AKS, I ran quick performance test using kube-str . The result is as follows:

	read_iops	write_iops	read_bw	write_bw
ceph-block	IOPS=464.507294 BW(KiB/s)=1874	IOPS=243.296143 BW(KiB/s)=989	IOPS=509.928162 BW(KiB/s)=65797	IOPS=248.530762 BW(KiB/s)=32338
ceph-filesystem	IOPS=438.701324 BW(KiB/s)=1770	IOPS=226.270660 BW(KiB/s)=920	IOPS=405.936340 BW(KiB/s)=52456	IOPS=208.869293 BW(KiB/s)=27229

The metrics reflects performance under default configuration. It should not be considered as the best performance that Ceph can deliver on Azure Kubernetes.

Summary

I discussed three storage options for Azure Kubernetes but the idea applies to other Kubernetes platform hosted on a CSP. The native storage has significant limitation. NFS has latency. Block storage does not address high availability at the storage layer. Portworx and LINSTOR fill that gap as a commercial solution. Ceph is based on Object storage.

Intro to Ceph storage

Thu, 21 Jul 2022 18:55:00 -0400

Ceph is a unified, distributed storage system designed for excellent performance, reliability and scalability. In this post, I will introduce Ceph and explain how it stands out from traditional enterprise storage technology.

Software defined storage

In the realm of enterprise storage, I discussed PowerScale (Isilon) from Dell EMC, and touched on ONTAP by NetApp as an alternative. These solutions usually include both enterprise grade hardware, and the software layer that manages those expensive hardware. As the competition with cloud storage arises, those vendors start to decouple the software layer from the hardware to sell them separately. As a result, clients have the options to use commodity hardware. On the other hand, the software layer is built to be more accommodative to different hardware options. Eventually, the software layer evolves into Software Defined Storage (SDS) with the purpose of supporting cheaper storage hardware.

This table shows the full solution offering and SDS offering from NetApp and Dell EMC:

	Full solution offering	SDS offering
NetApp	ONTAP	ONTAP Select
EMC	PowerScale	PowerFlex

It is not easy to make a proprietary SDS appliance support commodity hardware. For example, PowerFlex currently supports (and bundles with) DELL’s commodity hardware only. It is most likely an involuntary move. Then, why would these commercial providers even be motivated to support a broader range of hardware by moving to SDS? It is because they face fierce competition from open-source SDS technologies, which were born to support commodity hardware. In this family of technologies, Ceph is a rising star. This family also includes other technologies such as Gluster and HDFS.

Note that the performance of a storage based on SDS still has to do with the underlying hardware. Therefore, comparing Ceph storage with PowerScale is apple to orange, without identical storage hardware. Now that we decoupled SDS and hardware, let’s take a look at two important aspects of SDS: the distributed technology to manage hardware, and the interface it provides to storage clients.

Distributed storage

The reason to use an SDS layer to manage hardware in a distributed architecture is for better scalability and high availability. The soul of this SDS layer is the ability to manage distributed system. However, a distributed storage introduces problems of its own, such as coordinating consistency. Different storage technologies have their own way to tackle these problems. For example, with PowerScale, OneFS has its own Group Management Protocol. Ceph uses CRUSH for data distribution. GlusterFS uses DHT(Distributed Hash Table) Translator.

Storage architects usually do not need to know these technologies in detail. It is not the intention of this post to cover the details of any distributed technology in any of the storage options above. However, storage architects needs to know supported API very well.

Access API

The supported access API of a storage system determines its compatibility with client systems. One good example is NFS for file storage, which defines the protocol for file share without defining the underlying implementation. Most GNU/Linux distributions come with nfsd (NFS server) which exports directories on XFS or ext4 FS as a file share with NFS protocol. In order to transfer data over network, NFS uses RPC, a request-response protocol. With object storage, S3 is a widespread protocol. Below is a list of storage implementations and their supported access API:

Ceph supports librados, S3, Swift and FUSE
GlusterFS supports SMB, NFS, FUSE,
PowerScale supports NFS, SMB/CIFS, HDFS, Object, POSIX

CephFS is distributed file system built on top of Ceph RADOS. It is also a client-server architecture. A Ceph Client, via librados, interacts directly with OSDs to store and retrieve data. In order to interact with OSDs, the client app must invoke librados and connect to a Ceph Monitor. For compatibility, CephFS namespaces can be export over NFS protocol using NFS-Ganesha NFS server.

Ceph Architecture

Ceph is a high-performance, distributed storage platform. It provides object storage, block storage and distributed file system, all backed by a single, reliable storage cluster running on commodity server hardware. A Ceph Storage Cluster consists of Ceph Nodes on a network. A Ceph Storage cluster requires at least one Ceph monitor (ceph-mon), Ceph Manager (ceph-mgr) and Ceph OSDs (ceph-osd). For file system clients, it also requires Ceph Metadata Server (MDS, ceph-mds) to allow user to execute basic commands on POSIX file system (e.g. ls, find)

Under the hood, Ceph stores data as objects within logical storage pools. Using the CRUSH algorithm, Ceph calculates which placement group (PG) should contain the object, and which OSD should store the placement group. The CRUSH algorithm enables the Ceph Storage Cluster to scale, rebalance, and recover dynamically.

Ceph is based on RADOS (reliable autonomic distributed object store), a self-healing system that distributes and replicates data across nodes. It then layers CephFS (a distributed file system), block storage service (RADOS Block Device or RBD), and s3-compatible object storage (RADOS Gateway or RGW) on top of RADOS. For a better description, refer to this page. The chart above shows how Ceph interacts with different kinds of client. For CephFS, the client can interact with the file system via metadata daemon, as illustrated below. This diagram looks similar to the diagram for NFS.

In a RADOS cluster, each server runs some daemons (i.e. OSD, MON or MDS).

When an I/O request occurs, it needs to be mapped to the specific OSD that keeps the storage units. Here is an illustration of the mapping:

As typically observed in distributed system, there is quite some communication overhead to serve a file.

Ceph Cluster Installation

Installing a VM-based Ceph cluster is no trivial effort and there are several methods. The recommended method is Cephadm. Here is a good instruction, where you will notice a lot of steps on each nodes, such as configuring NTP, installing docker, configuring hostname, Linux user and SSH, etc. You may also check this video for how involving it is. Red Hat adopts Ceph project as a product and has an installation guide on its documentation.

Previously, there was a legacy tool ceph-ansible to help administrators with server configuration. It is similar to the way kubespray helps administrators configure Kubernetes cluster. However, the document suggests that ceph-ansible is not integrated with new orchestrator APIs and therefore is not a viable option anymore. Also I did not find a way to install a single-node ceph cluster just for a quick demo. It involves tweaking the CRUSH map configuration.

If we deploy Ceph on Kubernetes for Kubernetes workload, we use Rook, an orchestrator running on Kubernetes, to integrate storage to a cluster.

Cloud Native Storage

Moving to cloud native storage, instead of presenting storage to operating system, we need to configure storage classes for Pods to use persistent volumes dynamically, using storage provisioners. Ceph also shows good presence in cloud native storage ecosystem. In a self-managed Kubernetes cluster, Ceph gives us the capability to configure storage classes to access connected storage. In public cloud, Ceph allows us to configure storage classes connecting to disks attached to the Nodes, an alternative to the cloud vendor provided native storage classes with high availability across availability zones. This layer enables the organization to normalize how their application connects to persistent volumes, a capability particularly helpful in the multi-cloud strategy of the cluster.

Rook is a CNCF project to orchestrate storage system on Kubernetes. It automates storage administrative tasks such as deployment, bootstrapping, configuring, provisioning and monitoring, using declarative templates. It supports Ceph and a number of other storage backends such as Cassandra, NFS, MinIO.

Knative Eventing Introduction

Fri, 29 Apr 2022 10:16:00 -0400

In the previous post, I mentioned that Knative Serving and Knative Eventing should be seen as two different projects. The former is supposed to be widely used as a serving layer for microservices, whereas the latter has a narrower customer base. There are a dozen companies who need to build Platform as a Service, and will benefit from event-driven architecture. They are the niche customer for Knative Eventing. In this post, we go through two demos for Eventing.

Event Driven Architecture

Before jumping into Knative eventing, let’s first understand the value of event-driven architecture. As covered in Knative presentation in 2020, a growing micro-service architecture involves lots of service-to-service communications that looks like a spider web in a diagram. Two services communicating with each other are tightly coupled: changing one requires a change of the other. Event-driven architecture evolves to decouple those interdependent services. Knative eventing tries to standardize the event data with Cloud Event and also proposes a framework to drive this architecture.

The documentation uses different words to express the same concepts. So let’s take a note first: source = producer which emits events, and sink = subscriber = consumer which is the destination of events. Knative Eventing supports two models of in even-driven architecture.

Brokers and Triggers provides an “event mesh”. Event producers delivers events to a Broker. Triggers then distribute them uniformly by consumers.

Broker is a “hub” for events. It is a central location to receive and send events for delivery. We can choose from the following broker types:

MT (multi-tenant) channel-based broker (which a channel implementation, such as Kaka channel or an in-memory channel)
GCP broker
Apache Kafka broker
RabbitMQ broker

Triggers represents a desire to subscribe to events from a specific broker. It can act as a filter

channel-subscriber model

Channels and Subscriptions provide a “event pipe” model which transforms and routes events between Channels using Subscriptions. This model makes more sense when events out of one system needs to be transformed and then routed to another process.

A channel can be either a generic channel object, or a custom channel implementation (such as Kafka channel or in-memory channel)

The subscription consists of a Subscription object, which specifies the Channel and the Sink (aka the Subscriber) to deliver events to. You can also specify some Sink-specific options, such as how to handle failures.

The key difference is Broker-Trigger model allows you to filter events, where as Channel-Subscription model allows you to transform events.

Note that a Knative Service can act as both a Source and a Sink for events, and for good reason. You may want to consume events from the Broker and send modified events back to the Broker, as you would in any pipeline use-case. Between these components, Knative Eventing uses CloudEvents to send information back and forth.

Knative Eventing Lab

We can create a lab cluster using Kind as per the instruction in real-quicK-cluster. In docker-desktop settings, I configured 6 CPU and 8GB memory as the lab involves a number of nodes. The creation takes a couple minutes. This Knative eventing lab has four paths one can take, based on different model and different implementations:

Lab A demonstrates the configuration of broker-trigger model using Kafka based broker.
Lab B demonstrates the configuration of channel-subscription model, using in-memory channel.

The two labs share some common steps. We first install Knative Eventing using the provided manifests:

$ kubectl apply -f https://github.com/knative/eventing/releases/download/knative-v1.3.0/eventing-crds.yaml
$ kubectl apply -f https://github.com/knative/eventing/releases/download/knative-v1.3.0/eventing-core.yaml
$ kubectl get pods -n knative-eventing

The Pods should come to READY fairly quickly. In the mean time, we can configure the consumer as well using the manifests below:

apiVersion: v1
kind: Namespace
metadata:
 name: event-example
---
apiVersion: apps/v1
kind: Deployment
metadata:
 name: hello-display
 namespace: event-example
spec:
 replicas: 1
 selector:
 matchLabels: &labels
 app: hello-display
 template:
 metadata:
 labels: *labels
 spec:
 containers:
 - name: event-display
 image: gcr.io/knative-releases/knative.dev/eventing/cmd/event_display
---
kind: Service
apiVersion: v1
metadata:
 name: hello-display
 namespace: event-example
spec:
 selector:
 app: hello-display
 ports:
 - protocol: TCP
 port: 80
 targetPort: 8080
---
apiVersion: apps/v1
kind: Deployment
metadata:
 name: goodbye-display
 namespace: event-example
spec:
 replicas: 1
 selector:
 matchLabels: &labels
 app: goodbye-display
 template:
 metadata:
 labels: *labels
 spec:
 containers:
 - name: event-display
 image: gcr.io/knative-releases/knative.dev/eventing/cmd/event_display
---
kind: Service
apiVersion: v1
metadata:
 name: goodbye-display
 namespace: event-example
spec:
 selector:
 app: goodbye-display
 ports:
 - protocol: TCP
 port: 80
 targetPort: 8080

The consumers (aka sinks) will be able to process events at the validation step at the end.

Lab A – Install Kafka and configure Kafka-based Broker

This lab is for broker-trigger model. We first configure a Kafka Broker, which requires a Kafka cluster. The document uses Strimzi operator to quickly install Kafka cluster. Instead, here we use Helm to install the Kafka cluster:

helm install my-cluster-kafka bitnami/kafka -n kafka --create-namespace --set volumePermissions.enabled=true --set replicaCount=3

At the end of installation, the Helm notes indicate that the bootstrap server is my-cluster-kafka.kafka.svc.cluster.local:9092. All Kafka pods should come ready in about 5 minutes. After cluster install, we need to install Kafka controller and the Broker-layer data plane which allow you to map a Kafka instance to Broker CRD later:

kubectl apply -f https://github.com/knative-sandbox/eventing-kafka-broker/releases/download/knative-v1.3.0/eventing-kafka-controller.yaml
kubectl apply -f https://github.com/knative-sandbox/eventing-kafka-broker/releases/download/knative-v1.3.0/eventing-kafka-broker.yaml
kubectl get pods -n knative-eventing
kubectl -n knative-eventing edit cm kafka-broker-config

As the command above suggests, once data plane is installed, we edit the default broker config by updating the bootstrap.servers with the bootstrap server address noted above. We also need to make sure the replication factor is no greater than number of partitions, which should not be a problem in our lab as the replica count was set to 3.

Now we can create a Broker in the same namespace with consumer:

apiVersion: eventing.knative.dev/v1
kind: Broker
metadata:
 name: default
 namespace: event-example
 annotations:
 eventing.knative.dev/broker.class: Kafka
spec:
 config:
 apiVersion: v1
 kind: ConfigMap
 name: kafka-broker-config
 namespace: knative-eventing

Once the manifest above is applied, we should see the status of broker (named default) to be READY. We can also validate broker creation with kn CLI command:

 $ kubectl -n event-example get broker default
NAME URL AGE READY REASON
default http://kafka-broker-ingress.knative-eventing.svc.cluster.local/event-example/default 19m True
$ kn broker list -n event-example
NAME URL AGE CONDITIONS READY REASON
default http://kafka-broker-ingress.knative-eventing.svc.cluster.local/event-example/default 19m 7 OK / 7 True

Then we create Triggers that connects both to the broker with filters, and to the consumers by referencing service name. The manifest to create triggers is provided below:

apiVersion: eventing.knative.dev/v1
kind: Trigger
metadata:
 name: hello-display
 namespace: event-example
spec:
 broker: default
 filter:
 attributes:
 type: greeting
 subscriber:
 ref:
 apiVersion: v1
 kind: Service
 name: hello-display
---
apiVersion: eventing.knative.dev/v1
kind: Trigger
metadata:
 name: goodbye-display
 namespace: event-example
spec:
 broker: default
 filter:
 attributes:
 source: sendoff
 subscriber:
 ref:
 apiVersion: v1
 kind: Service
 name: goodbye-display

We can confirm the setups above by monitoring the resources (like we did for broker), or use knative CLI:

$ kn trigger list -n event-example
NAME BROKER SINK AGE CONDITIONS READY REASON
goodbye-display default service:goodbye-display 3m24s 6 OK / 6 True
hello-display default service:hello-display 3m24s 6 OK / 6 True

Now we can go to the validation step to send event and validate result.

Lab B – Install in-memory channel and configure Channel

This lab is for channel-subscription model. we first install an in-memory channel:

kubectl apply -f https://github.com/knative/eventing/releases/download/knative-v1.3.0/in-memory-channel.yaml

Then we can map this channel to a Channel resource in Eventing. We can do so by applying the manifest below:

apiVersion: messaging.knative.dev/v1
kind: Channel
metadata:
 name: demo-channel-1
 namespace: event-example
---
apiVersion: messaging.knative.dev/v1
kind: Channel
metadata:
 name: demo-channel-2
 namespace: event-example

The creation of Channel resources can be verified as below

$ kn -n event-example channel list
NAME TYPE URL AGE READY REASON
demo-channel-1 InMemoryChannel http://demo-channel-1-kn-channel.event-example.svc.cluster.local 6m34s True
demo-channel-2 InMemoryChannel http://demo-channel-2-kn-channel.event-example.svc.cluster.local 6m34s True

Now we need to create Subscription which connects Channels to Consumers. Subscriptions can be created with the manifest below:

apiVersion: messaging.knative.dev/v1
kind: Subscription
metadata:
 name: demo-subscription-1
 namespace: event-example
spec:
 channel:
 apiVersion: messaging.knative.dev/v1
 kind: Channel
 name: demo-channel-1
 subscriber:
 ref:
 apiVersion: v1
 kind: Service
 name: hello-display
---
apiVersion: messaging.knative.dev/v1
kind: Subscription
metadata:
 name: demo-subscription-2
 namespace: event-example
spec:
 channel:
 apiVersion: messaging.knative.dev/v1
 kind: Channel
 name: demo-channel-2
 subscriber:
 ref:
 apiVersion: v1
 kind: Service
 name: goodbye-display

We can confirm subscription:

$ kn -n event-example subscription list
NAME CHANNEL SUBSCRIBER REPLY DEAD LETTER SINK READY REASON
demo-subscription-1 Channel:demo-channel-1 service:hello-display True
demo-subscription-2 Channel:demo-channel-2 service:goodbye-display True

Now we can go to validation step.

Validation

For validation, we need to first configure an event source. The source in our lab is a throw-away Pod with curl utility. I use my favourite nicolaka netshoot to fire off the events using. To launch the Pod and get to command shell:

$ kubectl run tmp-shell -n event-example --rm -i --tty --image nicolaka/netshoot -- /bin/bash
If you don't see a command prompt, try pressing enter.
bash-5.1#

We can POST message to the corresponding endpoint URL to fire events. The endpoint is different for each lab:

Lab	URL
A (Kafka-based broker)	http://kafka-broker-ingress.knative-eventing.svc.cluster.local/event-example/default
B (channel)	http://demo-channel-1-kn-channel.event-example.svc.cluster.local http://demo-channel-2-kn-channel.event-example.svc.cluster.local

With the correct URL, we can build a payload for POST method. Below is three curl commands with different payload POST to the same endpoint (taking lab A as an example):

# curl -v "http://broker-ingress.knative-eventing.svc.cluster.local/event-example/default" \
-X POST \
-H "Ce-Id: say-hello" \
-H "Ce-Specversion: 1.0" \
-H "Ce-Type: greeting" \
-H "Ce-Source: not-sendoff" \
-H "Content-Type: application/json" \
-d '{"msg":"Hello Knative!"}'

# curl -v "http://broker-ingress.knative-eventing.svc.cluster.local/event-example/default" \
-X POST \
-H "Ce-Id: say-goodbye" \
-H "Ce-Specversion: 1.0" \
-H "Ce-Type: not-greeting" \
-H "Ce-Source: sendoff" \
-H "Content-Type: application/json" \
-d '{"msg":"Goodbye Knative!"}'

# curl -v "http://broker-ingress.knative-eventing.svc.cluster.local/event-example/default" \
-X POST \
-H "Ce-Id: say-hello-goodbye" \
-H "Ce-Specversion: 1.0" \
-H "Ce-Type: greeting" \
-H "Ce-Source: sendoff" \
-H "Content-Type: application/json" \
-d '{"msg":"Hello Knative! Goodbye Knative!"}'

All the requests should receive 202 return code. As highlighted above, we specified type and source for each event fired to the broker. On the consumer side, we can simply check the pod log to verify the events received. Below is an example of events received as a result of the events sent above (take broker-trigger model as example):

$ kubectl -n event-example logs -l app=hello-display --tail=100
2022/03/09 17:05:59 Failed to read tracing config, using the no-op default: empty json tracing config
☁️ cloudevents.Event
Context Attributes,
 specversion: 1.0
 type: greeting
 source: not-sendoff
 id: say-hello
 datacontenttype: application/json
Extensions,
 knativearrivaltime: 2022-03-09T17:14:19.196589801Z
Data,
 {
 "msg": "Hello Knative!"
 }
☁️ cloudevents.Event
Context Attributes,
 specversion: 1.0
 type: greeting
 source: sendoff
 id: say-hello-goodbye
 datacontenttype: application/json
Extensions,
 knativearrivaltime: 2022-03-09T17:14:42.977803608Z
Data,
 {
 "msg": "Hello Knative! Goodbye Knative!"
 }
$ kubectl -n event-example logs -l app=goodbye-display --tail=100
2022/03/09 17:05:58 Failed to read tracing config, using the no-op default: empty json tracing config
☁️ cloudevents.Event
Context Attributes,
 specversion: 1.0
 type: not-greeting
 source: sendoff
 id: say-goodbye
 datacontenttype: application/json
Extensions,
 knativearrivaltime: 2022-03-09T17:14:27.430702588Z
Data,
 {
 "msg": "Goodbye Knative!"
 }
☁️ cloudevents.Event
Context Attributes,
 specversion: 1.0
 type: greeting
 source: sendoff
 id: say-hello-goodbye
 datacontenttype: application/json
Extensions,
 knativearrivaltime: 2022-03-09T17:14:42.977803608Z
Data,
 {
 "msg": "Hello Knative! Goodbye Knative!"
 }

This is the gist of Knative evening. Note that the format of HTTP payload when we fire off events needs to conform to CloudEvents format. The container image (event_display) used in the container is also created for the purpose of processing events received and display them in stdout.

Conclusion

Persistent storage allows for stateful applications on Kubernetes. It also enables architectural patterns that requires persistent storage. Knative eventing is a framework that enables event driven architecture. It supports many messaging channels (including NATs, a cloud-native messaging system) and broker types (MT channel based, GCP, RabbitMQ).

Knative Serving Introduction

Sun, 17 Apr 2022 11:33:00 -0400

Background

As per IBM‘s definition, Knative enables serverless workloads to run on Kubernetes clusters, and makes building and orchestrating containers with Kubernetes faster and easier. It has drawn a lot of attention recently. It released version 1.0 in November 2021, and was accepted as a CNCF incubating project in March 2022.

Glories aside, the value it delivers is the ability to go serverless on a Kubernetes platform. Originally, Knative was built with three components: build, serving and eventing. The build component was deprecated in favour of the Tekton project. Tekton is cloud native CI/CD pipeline. It connects to source code repo and build artifacts. It can also deploy applications with pipeline as code. Tekton’s documentation includes a quality tutorial here linked to interactive terminal. On the other hand, we’ll focus on Knative Serving in this post, and Knative Eventing in the next.

Knative can be seen as a serverless framework with a number of open-source technologies as building blocks, such as Istio for ingress gateway, Kafka or Google Pub/Sub as event-streaming engine, and Prometheus for observability to name a few. Despite of using the same CLI tool (named kn), Knative Serving and Eventing are considered separate capabilities using different groups of CRDs. They are installed separately, using their respective YAML files, or operator.

Serving vs Eventing ?

After reading the article “Did we market Knative wrong” from Ahmet Balkan, my impression is the two are not related. They could have been two separate projects without sharing a common name. As the author puts, these two shared some core logics. But beyond that, they don’t have anything in common.

Serving is lightweight and requires Istio. It provides the capability to scale from N to 0 when the system is idle, and from 0 to N when requests come in. It is the missing serving layer for running microservices on Kubernetes and Ahmet position it as the first thing that people installed after creating a cluster.

The target user of Eventing is much smaller. Eventing is for event-driven architecture and is more complex than Serving. Ahmet admits that they over-estimated how many people on the planet want to build a Heroku-like PaaS layer on top of Knative. There are a couple of dozen companies who would work through the complexity and build their own Kubernetes-based internal PaaS or even public-facing FaaS using Knative. They are the niche audience of Knative Eventing.

For most platform builders who just need to run micro-service, Knative Serving is all they need.

A Demo on Serving

The definition of term “serverless” is very loose. Instead, Knative documentation promotes the ability to scale to zero and refer to it as “some people call this Serverless”. This suggests that “the ability to scale to zero” should be the better term to describe this capability. To deploy it, we need the Service object with apiVersion serving.knative.dev/v1. Below is a guide for a quick hands-on, with some steps modified from the Knative Serving tutorial and installation guide, with optional steps (extensions, DNS) skipped.

Follow the first part of this post or this guide in real-quicK-cluster project, to install minikube with metal LB and configure istio (using istioctl as per instruction). Then Knative serving can be installed with two YAML manifests:

$ kubectl apply -f https://github.com/knative/serving/releases/download/knative-v1.2.0/serving-crds.yaml
$ kubectl apply -f https://github.com/knative/serving/releases/download/knative-v1.2.0/serving-core.yaml

Then we can install Knative istio controller to integrate them so we can see the knative gateways:

$ kubectl apply -f https://github.com/knative/net-istio/releases/download/knative-v1.2.0/net-istio.yaml
$ kubectl -n knative-serving get gateway

Now we can install the dummy service, either using kn CLI tool or using YAML manifest. The manifest is given below:

apiVersion: serving.knative.dev/v1
kind: Service
metadata:
 name: hello
 namespace: default
spec:
 template:
 metadata:
 name: hello-world
 annotations:
 autoscaling.knative.dev/class: "kpa.autoscaling.knative.dev"
 autoscaling.knative.dev/metric: "rps"
 autoscaling.knative.dev/target: "50"
 autoscaling.knative.dev/min-scale: "0"
 autoscaling.knative.dev/max-scale: "10"
 spec:
 containerConcurrency: 0
 containers:
 - env:
 - name: TARGET
 value: World
 image: gcr.io/knative-samples/helloworld-go
 name: user-container
 ports:
 - containerPort: 8080
 protocol: TCP

We can notice that in the annotations there are some settings to overwrite the default configuration. They are explained in the autoscaling section of the Knative serving documentation. Here we use request per second (rps) as metric, and have 50 as target. The scale range is between 3 and 10 replicas. Once the resource is created, we can validate it with one of the two commands:

$ kubectl get services.serving.knative.dev
$ kn service describe hello

The creation of knative service also created a number of Kubernetes resources, such as deployment and services:

$ kubectl get svc
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
hello ExternalName <none> knative-local-gateway.istio-system.svc.cluster.local 80/TCP 52s
hello-world ClusterIP 10.110.127.231 <none> 80/TCP 108s
hello-world-private ClusterIP 10.111.27.73 <none> 80/TCP,9090/TCP,9091/TCP,8022/TCP,8012/TCP 108s
kubernetes ClusterIP 10.96.0.1 <none>
$ kubectl get deploy
NAME READY UP-TO-DATE AVAILABLE AGE
hello-world-deployment 1/1 1 1 71s

Before testing, let’s mock the DNS entry to resolve to the Istio ingress service IP like below in /etc/hosts:

192.168.64.16 hello.default.example.com

Now we can confirm that the service returns the expected result (after a brief pause):

$ curl hello.default.example.com
Hello World!

Wait for a short period without curl or connection from browser, the service should scale down to zero:

$ kubectl get deploy -l serving.knative.dev/service=hello -w
NAME READY UP-TO-DATE AVAILABLE AGE
hello-world-deployment 0/0 0 0 78s

From a separate terminal, emulate HTTP request with hey command, so the service receives more than 50 rps per pod:

$ hey -n 200 -c 20 -z 20s -m GET http://hello.default.example.com

We can watch the deployment get scaled up to 10. Once the hey command completes its task, the deployment will scale back down:

$ kubectl get deploy -l serving.knative.dev/service=hello -w
NAME READY UP-TO-DATE AVAILABLE AGE
hello-world-deployment 0/0 0 0 78s
hello-world-deployment 0/1 0 0 3m54s
hello-world-deployment 0/1 0 0 3m54s
hello-world-deployment 0/1 0 0 3m54s
hello-world-deployment 0/1 1 0 3m54s
hello-world-deployment 1/1 1 1 4m2s
hello-world-deployment 1/10 1 1 4m4s
hello-world-deployment 1/10 1 1 4m4s
hello-world-deployment 1/10 1 1 4m4s
hello-world-deployment 1/10 10 1 4m5s
hello-world-deployment 2/10 10 2 4m9s
hello-world-deployment 3/10 10 3 4m10s
hello-world-deployment 4/10 10 4 4m10s
hello-world-deployment 7/10 10 7 4m10s
hello-world-deployment 9/10 10 9 4m10s
hello-world-deployment 10/10 10 10 4m11s
hello-world-deployment 10/5 10 10 5m18s
hello-world-deployment 10/5 10 10 5m18s
hello-world-deployment 5/5 5 5 5m18s
hello-world-deployment 5/2 5 5 5m20s
hello-world-deployment 5/2 5 5 5m20s
hello-world-deployment 2/2 2 2 5m20s
hello-world-deployment 2/1 2 2 5m22s
hello-world-deployment 2/1 2 2 5m22s
hello-world-deployment 1/1 1 1 5m22s
hello-world-deployment 1/0 1 1 5m52s
hello-world-deployment 1/0 1 1 5m52s
hello-world-deployment 0/0 0 0 5m52s

Autoscaling in Knative serving is backed by KPA (Knative Pod Autoscaler, used in the example above) or HPA. Also note that the trigger to scale up and down is client connection (RPS). This is different than KEDA. KEDA also supports the capability to scale to zero. However, with KEDA, it is an event that scales a deployment up from zero. In Knative serving, it is the connection to the Service itself that wakes up the service.

Cost of scale-to-zero

As discussed, both KEDA and Knative serving supports the ability to scale to zero and wake up from zero. The trigger to wake up from zero is different. With KEDA, workload wakes up by an event. With Knative Serving, workload has to wake up upon receiving a connection to the service. In the design to solve this problem, Knative has four sub-components :

Activator – When a service is scaled to zero, its request are routed to the activator which waits for a Pod to become ready (wake up) and proxies the traffic while editing the underlying Endpoints to route the traffic directly to the Pod(s)
Autoscaler – KPA
Controller – The main component responsible for watching API objects for Knative CRDs (KService, Configuration, Route, Revision) and manage their lifecycle, create the underlying Kubernetes resources and garbage-collect old objects.
Webhook – a Kubernetes Admission Webhook acting both as validating admission controller and mutating admission controller.

The key player is activator, which receives requests when a service is scaled to zero. The wake-up process is summarized as follows:

After receiving request, activator will buffer (hold onto) the request
Then, it will look at request’s hostname to find which KService it is for.
Then, it will scale up the Kubernetes Deployment and wait for a Pod to become ready.
In the meanwhile, it updates Kubernetes Service to point to Pod IP addresses (so that activator gets out of the network path if the KService is awake).
Finally, activator proxies the request to the started pod

So the “plumbing” of the service connection managed by the Activator pod is the cost of scale to zero. This is also called Load Balancing, and the behaviour can be tweaked using two parameters: activator capacity and target burst capacity:

Target burst capacity: once the target deploy has one more more Pod, service requests may still route via activator or bypassing the activator. Target burst capacity determines at what point service request should start to bypass activator.
Activator capacity: determines how many requests can activator hold on to. Considering the service-to-service connection also routes via Service CRD, there can be a lot of connections via Activator so its capacity needs to be adjusted.

The additional overhead of managing and optimizing the “Load Balancer” is also part of operational cost for the ability to scale to zero.

In comparison, KEDA now has an HTTP add-on still at beta but allows connection-based wake-up. The design is a little different. As discussed in this previous post.

Conclusion

Knative consists of two disparate components: Serving and Eventing. Serving uses KPA to provide scaling based on service request, and the ability to scale to zero. It is often compared with KEDA, a single-purpose lightweight tool for autoscaling. Here is a blurb on their difference by KEDA. The takeaways is that KEDA is more focused on scalability, whereas Knative serving covers more aspects. For example, the Service object of Knative also supports Traffic Splitting. Knative serving can integrate with istio.

Autoscaling on Kubernetes Platform

Mon, 28 Mar 2022 13:14:00 -0400

Introduction

The concept of autoscaling on Kubernetes platform dates from the era where virtualization first became widespread and the overhead of provisioning a new server became lightweight through the use of cloud-init. With public cloud, customers operate on usage-based billing. Autoscaling allows workload to scale down during idle times to reduce cost, and scale up during peak time to meet the demand of business traffic. Vertical autoscaling replaces a VM with one of higher capacity, which is usually interruptive. Horizontal autoscaling adds or removes VMs to adjust capacity, and works in conjunction with load balancing mechanism to assign load to a specific target in the group. Unless otherwise specified, we simply refer to horizontal autoscaling as autoscaling. Depending on what triggers autoscaling, it can be metrics based or event driven.

Metric-based autoscaling is based on VM (or Pod) metrics in the autoscaling group. The metrics are mostly about CPU usage, memory, IOPS, number of connections, etc. For example, when the average CPU utilization across all VMs in the last five minutes hits 70% threshold, then the scaler introduces a new VM into the autoscaling group. The trigger can factor in a variety of metrics. Advanced autoscaling APIs can also support lifecycle hooks, i.e. custom activities upon creation of new VMs during scale-up, or upon deletion of existing VMs during scale-down. Other aspects of custom behaviours include a cool-off period, i.e a no-activity window after the previous scaling activity. Since scaling activities are re-active.

Using real-time metrics as a trigger of scaling is not always a good idea. For example, a buggy order processing program may consume 100% of CPU due to an infinite loop, or 99% of memory due to memory leak. Metric-based scaling may fire off even though there is currently no order pending in the queue. Event-driven approach is more flexible. Event can fire from any type of source. For example, in Kubernetes, when scheduler fails to schedule a Pod due to constraints, it is an event This event can trigger scaling. In some case, a metric hitting a threshold fires an event. For example, scale up when size of order queue reaches 20. In this sense, metric-based autoscaling is a special case of event-driven autoscaling.

With Kubernetes, let’s examine node group autoscaling (aka cluster autoscaling) and workload autoscaling (Pod autoscaling).

Node Group Autoscaling

Cluster autoscaler is the mechanism to auto-scale node groups for Kubernetes. As per its documentation, any metric-based cluster/node group autoscalers are NOT compatible with CA. They are also not particularly suitable for Kubernetes in general. Take AKS for example, the events to trigger scale-up and scale-down are as below:

The cluster autoscaler component can watch for pods in your cluster that can’t be scheduled because of resource constraints. The cluster then automatically increases the number of nodes.
The cluster autoscaler decreases the number of nodes when there has been unused capacity for a period of time. Pods on a node to be removed by the cluster autoscaler are safely scheduled elsewhere in the cluster.

Both are in essence event driven. The behaviours can be fine-tuned with a number of parameters as below:

scan-interval
scale-down-delay-after-add
scale-down-delay-after-delete
scale-down-delay-after-failure
scale-down-unneeded-time
scale-down-unready-time
scale-down-utilization-threshold
max-graceful-termination-sec
balance-similar-node-groups
expander: random, most-pods, least-waste, priority
skip-nodes-with-local-storage
skip-nodes-with-system-pods
max-empty-bulk-delete
new-pod-scale-up-delay
max-total-unready-percentage
max-node-provision-time
ok-total-unready-count

The parameters above constitute the autoscaler profiler, and are effective if cluster autoscaler is enabled. For many implementations, cluster autoscaler can be enabled and disabled even after the cluster has been created, and the parameters can be changed. The overhead of provisioning a new node should not be overlooked, because that is usually the window that a Pod needs to wait to get scheduled. As stated in CA’s FAQ, the main purpose of CA is to get pending pods a place to run, instead of pre-emptively accommodating to increasing workload.

The delay in pod scheduling while adding a new node can be controlled to a certain degree with one of the two workarounds below:

With HPA or KEDA, set lower threshold so the workload level scaling acts more aggressive than the increase of demand. This buys some buffer time
Use a tool to puff up utilization, such as cluster overprovisioner, which deploys pods that
- request enough resources to reserve virtually all resources for a node
- consume no actual resources
- use a priority class that causes them to be evicted as soon as any other Pod needs it.

In practice the cluster autoscaler setup should be conservative and keep node size as stable as it can. For example, a 20 minutes idle-window (low utilization) on a node is not worth the overhead to remove a node and add it back in 20 minute later.

When the cluster do need to scale down by removing a node, one common symptom is failing to scale down because some Pods have nowhere else to schedule to. Here is a list of possible causes as troubleshooting tips.

If node scaling should be triggered sparsely, then pod scaling is by design very dynamic. Cloud native applications should assume that pod scaling occurs very frequently.

In late 2021, AWS released the open-source project Karpenter for cluster autoscaler. Karpenter addresses some challenges with native Cluster Autoscaler on EKS. Karpenter is gaining momentum and now adding support for other cloud service providers including Azure.

Workload Autoscaling

Stateless workload are controlled by a Deployment object, which is associated with a replicaSet object. For stateless workload we can use HorizontalPodAutoscaler, or HPA. There is a VerticalPodAutoscaler (VPA) which is much less common. HPA is metrics based with flexible options such as specifying an object, depending on what metrics are available via metrics API. There are two versions of HorizontalPodAutoscaler: autoscaling/v1 and autoscaling/v2. The latter supports scaling policies, such as adjusting downscale stabilization window, and limiting scale down rate. No matter which API version, the metric-based triggers in HPA are fairly limited.

We already know that metrics are not always the best indicator to trigger scaling. We need an option to trigger scaling based on the status of other components such as queue size. KEDA (Kubernetes Event Driven Autoscaling) is a great option to consider for horizontal workload scaling. KEDA works with HPA, and significantly enriches trigger options. Apart from metrics, KEDA can use a number of external mechanisms as triggers, for example:

RabbitMQ/Kafka/SQS: scale based on queue size
Azure Log Analytics: scale based on a kusto query result against Azure Log Analytics
AWS CloudWatch, Azure Monitor: scale based on metrics from Azure Monitor/AWS CloudWatch
Azure Pipelines: scale based on agent pool queues of Azure Pipeline
Elasticsearch: scale based on elasticsearch query result
Kubernetes Workload: scale based on the count of running pods of a specified workload
MSSQL, MySQL, Postgres, Cassandra: scale based on a query result
Prometheus: scale based on prometheus query result

KEDA is a single-purpose and lightweight component. With KEDA, we don’t need to explicitly define HPA. It allows us to select from a longer list of triggering mechanisms for our auto scaler. We shall not underestimate the work needed to select the most suitable trigger because having an incorrect trigger (e.g. bad metrics) is costly. Let’s take Java applications as an example. Java workload operates in a JVM inside of the container. JVM request the entire heap size from operating system. The garbage collection activities also consumes a good portion of CPU cycles. This pattern makes CPU and memory metrics inaccurate as an indicator for scaling activity. Because of this we need to find out what is the best scaler for Java application, based on understanding of how the entire solution stack works as a whole.

The other aspect that KEDA beats HPA is its ability to scale to 0. This can be helpful when a service is idle most of the time but cannot shut down.

KEDA lab

let’s use Kafka as an example to configure KEDA for a dummy workload. We create a mock cluster using Kind with a simple configuration file. Then, let’s start with the following dummy workload with replica count set to 1:

apiVersion: v1
kind: Namespace
metadata:
 name: workload
---
apiVersion: apps/v1
kind: Deployment
metadata:
 name: aks-helloworld-one
 namespace: workload
spec:
 replicas: 1
 selector:
 matchLabels:
 app: aks-helloworld-one
 template:
 metadata:
 labels:
 app: aks-helloworld-one
 spec:
 containers:
 - name: aks-helloworld-one
 image: neilpeterson/aks-helloworld:v1
 ports:
 - containerPort: 80
 env:
 - name: TITLE
 value: "Welcome to Azure Kubernetes Service (AKS)"
---
apiVersion: v1
kind: Service
metadata:
 name: aks-helloworld-one
 namespace: workload
spec:
 type: LoadBalancer
 ports:
 - port: 80
 selector:
 app: aks-helloworld-one

We need to install KEDA and Kafka using Helm:

helm repo add kedacore https://kedacore.github.io/charts
helm install keda kedacore/keda -n keda --create-namespace
helm repo add bitnami https://charts.bitnami.com/bitnami
helm install kafka bitnami/kafka -n kafka --create-namespace --set volumePermissions.enabled=true --set replicaCount=3

Watch for all Pods to come up. Also read the notes from Kafa installation and confirm the Kafka service address. Now, we will apply KEDA scaled object, defined as below:

apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
 name: kafka-scaledobject
 namespace: workload
spec:
 scaleTargetRef:
 apiVersion: apps/v1
 kind: Deployment
 name: aks-helloworld-one
 pollingInterval: 10
 cooldownPeriod: 30
 idleReplicaCount: 0
 minReplicaCount: 2
 maxReplicaCount: 5
 fallback:
 failureThreshold: 3
 replicas: 1
 advanced:
 restoreToOriginalReplicaCount: true
 horizontalPodAutoscalerConfig:
 behavior:
 scaleDown:
 stabilizationWindowSeconds: 300
 policies:
 - type: Percent
 value: 100
 periodSeconds: 15
 triggers:
 - type: kafka
 metadata:
 bootstrapServers: kafka-0.kafka-headless.kafka.svc.cluster.local:9092
 consumerGroup: my-group # Make sure that this consumer group name is the same one as the one that is consuming topics
 topic: test
 # Optional
 lagThreshold: "5"
 offsetResetPolicy: latest

The field definitions are on KEDA deploy documentation. Items under Kafka trigger are on the trigger documentation. In this lab we set the idelReplicaCount to 0. It will scale up with average lag of all partitions reaching 5. In the next few steps, we’ll mock up some messages posted to the Kafka topic named “test” for consumer group my-group. We can watch deployment size grow along with the growth of lags.

To emulate Kafka client activity, we can spin up a Kafka test Pod:

kubectl run kafka-client --restart='Never' --image docker.io/bitnami/kafka:2.8.1-debian-10-r73 --namespace kafka --command -- sleep infinity
kubectl exec --tty -i kafka-client --namespace kafka -- bash

From within the Pod, we can leverage the client-side scripts located in /opt/bitnami/kafka/bin/. For example, to post message to a topic (e.g. named test):

kafka-console-producer.sh --topic test --broker-list kafka-0.kafka-headless.kafka.svc.cluster.local:9092,kafka-1.kafka-headless.kafka.svc.cluster.local:9092,kafka-2.kafka-headless.kafka.svc.cluster.local:9092

The Helm installer also gives the command with broker list. To consume messages from a topic (e.g. test) to a given consumer group:

kafka-console-consumer.sh --topic test --bootstrap-server kafka.kafka.svc.cluster.local:9092 --group my-group

We can have two command terminals, post test messages on one terminal and watch it consumed nearly immediately on the other terminal.

The Kafka trigger documentation suggests that the number of replicas will not exceed the number of partitions on a topic when a topic is specified. To make this lab work, we need to have set 5 partitions:

kafka-topics.sh --alter --bootstrap-server kafka.kafka.svc.cluster.local:9092 --topic test --partitions 5
kafka-topics.sh --describe --bootstrap-server kafka.kafka.svc.cluster.local:9092 --topic test

Once we confirm five partitions, we can spin up two command terminals, one to produce message and the other to consume messages. If working, we can stop the consumer and use the command below to watch for the lag for each partition.

kafka-consumer-groups.sh --bootstrap-server kafka.kafka.svc.cluster.local:9092 --describe --group my-group

Now we can artificially trigger scaling by increasing average lag. We keep posting messages on the producer (each carriage return posts a message), and we can check the lag after posting:

Growth of average lags

The size of deployment starts with 0 as defined in the scaled object. As the average exceeds 5, we can see deployment size growing.

Growth of deployment size

This lab is an oversimplified scenario to illustrate the idea of scaling. Kafka is a typical queue construct and other queue configuration such as RabbitMQ or AWS SQS works in very similar ways. Real life use case involves more aspects to consider, such as multiple topics, and authentication.

Connection triggered wake-up

KEDA uses Events to scale workload from zero to one (wake up). There is no way to scale (wake up) based on an incoming web request. In many cases, such as serverless configuration, we need to scale the deployment size from zero to N once the service receives incoming web request. This is not supported by KEDA. By definition KEDA uses Events to wake up.

There is an HTTP-add-on for KEDA still at beta but it is trying to address this problem. This page shows the design. Suppose a service has scaled down to zero, the followings will happen to wake it up:

The incoming request is routed to an interceptor behind the service
interceptor keeps track of number of pending HTTP request
The scaler periodically watches for the size of the pending queue on the interceptor
Based on the queue size, the scaler reports scaling metrics as appropriate to KEDA
As the queue size increases, the scaler instructs KEDA to scale up as appropriate

The periodical check activity is the key to make it work and also what makes it a pseudo-trigger.

Summary

Fine-tuning autoscaling is important to the performance of workload on Kubernetes. At node level, we briefed on cluster autoscaler and suggest that we only use it sparsely. At pod level, we introduced native HPA as well as KEDA, with an example. We also discussed KEDA has limitations and the HTTP-add-on. In the next post, we’ll explore Knative’s autoscaling capability.

File storage vs object storage in the cloud

Thu, 23 Sep 2021 22:54:00 -0400

File storage (e.g. NFS) used to be prevalent until object storage comes in for competition.

The competition

Traditionally, enterprise storage product lines are built around three capabilities, as listed in this table below:

Capability	Typical Implementation	Data served
T1 – Block stroage	DAS (e.g. SAS cable) or SAN (Fibre Cable for FCP protocol, or Ethernet for iSCSI protocol)	Mission critical data that are extremely sensitive to latency (e.g. database). Client has block-level access.
T2 – File storage	NAS (connect via CIFS or NFS protocols). Storage arrays are typically a mix of HDD and SSD. Storage servers are usually deployed in the same location over low latency network. DR location is usually in the same region.	Hot data. Multiple client access at file level. The size of each data request varies from small to medium (e.g. text document)
T3 – Object storage	Hardware agnostic, connect via layer-7 protocol (e.g. S3). Storage backend can be either on premise, or in the cloud, over WAN connection.	Warm and code data. Multiple client access at object level. Traditionally for backup but use cases are expanding. The size of each data request varies significantly, from small to very large (e.g. media content).

In the last couple decades, leading players for T2 have been enterprise storage vendors. They each have developed their secret sauces to tackle the challenges. For example, EMC has OneFS, a parallel distributed file system as the foundation of PowerScale (formerly Isilon) product line. NetApp develops ONTAP, featuring proprietary techniques for storage efficiency (deduplication, compaction and compression).

The leading players in T3 are mostly public cloud provider, such as Amazon’s S3. They might work with enterprise storage vendor behind the scene. But the T3 services appear to the end users as provided by the public cloud. Originally, the use case for T3 was archive only for its virtually unlimited capacity. This is not entirely true today. With the drastic improvement in modern network infrastructure, T3 can also brings satisfactory performance to serve hot data.

A competition between T2 and T3 arises. After all, both offer storage service over Ethernet, and both support multiple clients. Today when developers architect the storage layer of their applications, they need to weigh between supporting T2 and T3. Since NFS is the typical protocol for T2 storage (sorry Windows guys) and S3 is typical T3 storage. This competition essentially boils down to NFS versus S3.

For many, the fancy S3 is a no-brainer. While I have suffered from many NFS drawbacks, and there’s even a whole article by Linux folks about why NFS sucks, is it sentenced to death today? Does it beat S3 in some cases? Do so many organizations still stick to NFS just out of inertia?

To answer these questions, I examine four aspects to explore the differences between file storage via NFS protocol, and object storage in S3.

Data request size

Storage client can make request by byte range of a file. Therefore, data request size, instead of file size, is what ultimately matters. I pick a few data request sizes (1K, 4K, 16K, 64K, 246K, 1024K and 4096K) in my experiment, and want to see how much network traffic a write operation produces using NFS and using S3.

To emulate request size, I created files at each size (using dd command), and copy the entire file to each backend. In the mean time, I use tcpdump to write out traffic across the wire into capture files. The size of capture file gives me an idea of how much network traffic went through the network interface, which is closely related to latency.

For NFS, I mounted the target with sync option. This requires NFS client to write out to server synchronously on file copy. I’ve also set the wsize to be 1M. For S3, I simply use the following CLI command to copy file:

aws s3 cp 1kb.img s3://digihunch5ffafe32ab0fd40f

On the network interface, I use tcpdump to filter traffic through specific TCP port (443 for S3, or 2049 for NFS) and record the size of the capture file:

sudo tcpdump -s0 -pi eth0 dst port 443 or src port 443 -w /tmp/4096kb.cap

The key indicator is the payload size (file size) as a percentage of the capture size. I call it payload ratio. The closer it is to 1, the better. I have the following result from my experiment:

Request	Payload	S3 capture size (byte)	NFS capture size (byte)	S3 payload ratio	NFS payload ratio
1K	1024	9352	4332	0.11	0.24
4K	4096	12640	7404	0.32	0.55
16K	16384	25969	20472	0.63	0.80
64K	65536	79183	69746	0.83	0.94
256K	262144	290035	271408	0.90	0.97
1024K	1048576	1085366	1074076	0.97	0.98
4096K	4194304	4381547	4286910	0.96	0.98

This result indicates that NFS has a higher ratio in all groups. However, its advantage diminishes as the data request size grows. What it tells us is that if your applications workload issues most request in small chunks of data, such as 1K, 4K, then NFS will require much less traffic over the network, and thus less latency.

This essentially explains the use case of NFS against S3: workload with small data requests.

Client Support

NFS is natively supported by Linux operating system kernel. NFS client sits below the virtual file system (VFS) layer, which sits below the system call layer. The NFS client translate system calls into RPC (remote procedure calls). Communication between client and server is completed with RPC, on top of TCP.

NFS architecture

Because of the native support, in most cases, developer can treat NFS mounts as if they were local. For performance to be sustainable as file system grows, the directory structure on NFS should follow a certain naming conventions so that files are evenly distributed across directories. The client should also use list operation as sparse as it can because that operation is expensive across the network.

From developer’s perspective, NFS support is brought in by operating system and does not require much effort. On the other hand, S3 client support is not included by default in the operating system. S3 support requires special library, code changes, and integration effort to manage dependency and library version.

NFS has an advantage on client supportability. However, as we move applications to containers, and as container storage options mature, we will need an intermediary layer (storage class, storage provisioner, CSI driver, etc), NFS, or in general file storage, does not have this advantage any more.

Client-side Cache

The NFS support behind VFS layer also means it can leverage the I/O caching mechanism on the client side, that comes with operating system. Client operating system with sufficient memory can take advantage of this mechanism to give it a performance boost. Check out this guide for NFS cache tuning.

In comparison, S3 does not have a cache mechanism by itself. Either the application needs to implement its own cache mechanism, or a cache architecture needs to be introduced, such as CloudFront.

Consistency and concurrency

A common consistency problem is whether client can read the changes immediately after it writes the file. S3 and NFS make a tie in this round.

S3 originally came with eventual consistency model for read after write since 2006. As of Dec 2020 it introduced strong read-after-write consistency. For more information, refer to the guide here.

NFS has a similar consistency guarantee called close-to-open cache coherency. Any changes made by client are flushed to the server on closing the file, and a cache revalidation occurs when you re-open it.

There are more to consider in terms of consistency. For example, multiple clients tries to write the same file/object at the same time.

On the S3 side, there is a locking mechanism called S3 object lock at object level (no byte-range lock). Without an object lock, when two PUT requests are simultaneously made to an object, the request with the latest timestamp wins. Refer to the section Concurrent application on this page.

As far as NFS goes, managing this kind of consistency problem is not in the scope of the standard. Although there are some tinkers. For example, NFS v4 includes a file locking mechanism. Client can choose to lock the entire file, or a byte range within the file. Locking can be mandatory or advisory.

The convergence

NFS and S3 each has their respective advantage. Enterprise NAS customers have been looking for ways to expand into the cloud for lower storage cost. To combine the advantages of the two, solution providers started to converge file storage and object storage. There are two types of solutions that reflects this trend of convergence. In the first trend, enterprise NAS deployed on premise now have the ability to scale out into the cloud. In the second trend, public cloud just brought enterprise NAS into their product offerings.

Scale-out NAS

NAS is traditionally expensive to scale because it requires physical storage media. The idea of scale-out NAS allows NAS to connect to object storage in the public cloud, making it a hybrid architecture. This essentially makes T3 storage as a backend of T2 and it can be implemented with a virtual storage appliance (VSA). The VSA translate file system activities into API calls for object storage operations. One example is AWS storage gateway. EMC has a similar appliance called ECS and this white paper explains how it proxies file system calls and interact with object backends. NetApp, a vested enterprise NAS provider, also has a counterpart called Cloud Volumes ONTAP (CVO). It works well with NetApp on-premise deployment, but the architecture is similar. Here‘s NetApp’s take on how CVO is different than AWS Storage Gateway.

In the scale-out NAS architecture, the public cloud acts merely as extension to on-premise storage solution, to provide capacity. The NAS on premise serves the storage workload primarily.

Cloud hosted NAS

For applications hosted in public cloud, it makes sense for public cloud provider to operate enterprise NAS storage as a service. The underlying storage technology is provided by storage vendor. It is just installed in the data centre managed by the public cloud vendor, instead of customer’s own data centre.

One example is Azure NetApp Files (ANF). ANF is fully managed services, presented to users as storage volumes. The underlying storage technology is NetApp ONTAP. Because it is offered as a fully managed service, the customers are not able to manage the fine details of the storage, as they could with an ONTAP cluster on premise. This takes a lot of flexibility away from the user.

FSx ONTAP is a managed NetApp storage service by AWS, launched in September 2021. The NetApp arrays are installed in AWS data centre, ready for users to provision from AWS console, or using CLI. The Terraform provider support is not available as of yet. Unlike ANF, FSx ONTAP exposes the ONTAP CLI to users, allowing for advanced storage managed by storage gurus. They can use ONTAP CLI commands to configure custom policy for Snapshot, setup SnapMirror replication, and so forth.

Likewise, PowerScale landed on GCP as public cloud partner to launch Dell Cloud PowerScale for Google Cloud in 2020. However, it seems to require a purchase agreement before APIs are enabled.

Conclusion

Object storage has a great momentum and some sees that as a replacement of file storage in the long run. However file storage has its advantages for small data requests, OS-level cache support, and built-in POSIX compatibility. It will continue to be an option for customers with specific workload. Customer stickiness to file storage is so firm, that public cloud providers now install them in their data centres.

From competition to collaboration, it will be interesting to watch what happens next for enterprise storage.

Follow-up Reading

Tom Lyon’s presentation on why NFS must die.

Spark, Cassandra and Python

Tue, 15 Sep 2020 16:24:09 -0400

In this post we touch briefly on Apache Spark as a cluster computing framework that supports a number of drivers to pipe data in, and that its stunning performance thanks much to resilient distributed dataset (RDD) as its architectural foundation. In this hands-on guide, we expand on how to configure Spark, and use Python to connect to Cassandra data source.

Spark supports Sala, Java and Python shells. I’m not familiar with Scala but I have had Python background and know it’s importance in big data processing. One key data structure with big data processing in Python is Pandas data frame. Spark has the ability to map its own data frame to Pandas data frame.

Spark also needs a third party connector to connect to Cassandra. This connector is provided by Datastax in this open-source project called spark-cassandra-connector. The Github page includes a README with compatibility matrix, which is very important to understand before any configuration works. However, the Github is only the source code repository for anyone to build the project themselves. An alternative source of the dependency is this page from Maven repository. When running Spark we can simply reference that page URL as dependency.

Suppose we install spark onto CentOS, we download and unzip this package to somewhere such as user directory (~). Assuming we already have Open JDK 1.8 installed, when we run spark binary, it places cache and jar files in ~/.ivy2, potentially we need to manually move the following dependencies to ~/.ivy2/jars:

org.codehaus.groovy_groovy-json-2.5.7.jar
com.github.jnr_jffi-1.2.19.jar
org.codehaus.groovy_groovy-2.5.7.jar

These jar files are available for download from Maven’s repository as well if you wish provide them as package dependencies. We have two flavours of interactive shells to connect to Spark: the Scala shell (spark-shell) and python shell (PySpark)

Scala Shell

We can enter the default scala shell by

$ ./bin/spark-shell --packages com.datastax.spark:spark-cassandra-connector_2.11:2.5.1 --conf spark.cassandra.connection.host=10.10.10.151 --verbose

During the start, note a stdout line that says:

Spark context Web UI available at http://spark-host:4040

Then we can open that tcp port on iptables and view that job in browser. From within scala shell we can test connectivity to Cassandra with the following commands:

>>> val new_exam = spark.read.format("org.apache.spark.sql.cassandra").options(Map("table" -> "new_exam","keyspace" -> "examarchive")).load()

Python Shell

Python Shell (aka PySpark) brings Python shell which is known to many engineers from system admin or development background. By default, python 2 will be used. To specify python version, set some environment variables before we start pyspark with cassandra connector package specified:

$ export PYSPARK_PYTHON=python3
$ export PYSPARK_DRIVER_PYTHON=python3
$ export SPARK_HOME=/home/dhunch/spark-2.4.6-bin-hadoop2.7
$ export PATH=$SPARK_HOME/bin:$PATH
$ ./bin/pyspark --packages com.datastax.spark:spark-cassandra-connector_2.11:2.5.1 --conf spark.cassandra.connection.host=10.10.10.151

Once you’re in the interactive shell, you can start with loading required python libraries, and test your connectivity:

>>> from pyspark import SparkContext, SparkConf
>>> from pyspark.sql import SQLContext

>>> load_options = { "table": "new_exam", "keyspace": "examarchive"}
>>> df=spark.read.format("org.apache.spark.sql.cassandra").options(**load_options).load()
>>> df.show()
>>> df.write.csv('/tmp/mycsv.csv')

>>> #df.registerTempTable("ne")
>>> df.createTempView("ne")

>>> tw1=sqlContext.sql("select count(*) from ne")
>>> tw1.show()

>>> qrdf2=sqlContext.sql("select study_key, image_count from ne where current_exam_version=exam_version")
>>> qrdf2.write.csv('/tmp/tw2')

Note that the load method returns type pyspark.sql.dataframe.DataFrame, which is already a distributed data structure. So there is no need to parallelize it with parallelize() method. As of Spark 2.0, we are supposed to use createTempView() method instead of the old registerTempTables() method. Read this for further information.

Python Application

With interactive shell you run one or several commands at a time. We can build a python script and submit the whole script as an application. This is an example command:

./bin/spark-submit --packages com.datastax.spark:spark-cassandra-connector_2.11:2.5.1 sample.py

Note that the sample.py script name must be provided after –packages switch. Otherwise, you will get an error saying missing dependency (Failed to find data source: org.apache.spark.sql.cassandra). In the script, we can manipulate the data from Cassandra with greater flexibility. For example, we can map one field to several fields. For example, if one of the fields stores an XML document, the script can drill down the XML tree structure parse out values at different levels of child nodes, into separate data base columns. Here is an example of python script where we register a custom UDF declared in python and apply it to some existing columns to build new columns:

#! /usr/bin/python3
# To submit this script as an application to spark:
# ./bin/spark-submit --packages com.datastax.spark:spark-cassandra-connector_2.11:2.5.1 examstat.py
# Note that the script name must be placed after --packages 

import sys,datetime,re
import xml.etree.ElementTree as ET
from pyspark import SparkContext, SparkConf
from pyspark.sql import SQLContext, SparkSession
from pyspark.sql.functions import udf 
from pyspark.sql.types import StringType,StructType,StructField

cluster_seeds=['dest_cass_host']

def pTrimExamCode(raw_code):
 return 'NULL' if raw_code is None or raw_code=='None' else str(raw_code).replace(',','').rstrip('\r\n')

def is_valid_date(date_str):
 isValidDate=bool(re.match("^(19|20)\d\d(0[1-9]|1[012])(0[1-9]|[12][0-9]|3[01])$",date_str))
 if isValidDate:
 try:
 datetime.datetime(int(date_str[:4]),int(date_str[4:6]),int(date_str[6:8]))
 except ValueError:
 isValidDate=False
 return isValidDate

def pPullTags(study_key,raw_xml_field):
 ns={"vc":"http://medical.nema.org/mint"}

 StudyDateTag='None'
 StudyDescriptionTag='None'
 try:
 if raw_xml_field is not None:
 summary_tree=ET.fromstring(str(raw_xml_field)) # str function outputs 'None' or null object
 xml_find_res=summary_tree.find("vc:attributes/vc:attr[@tag='00080020']",ns)
 if xml_find_res is not None: StudyDateTag=str(xml_find_res.attrib.get('val'))
 
xml_find_res=summary_tree.find("vc:attributes/vc:attr[@tag='00081030']",ns)
 if xml_find_res is not None: StudyDescriptionTag=str(xml_find_res.attrib.get('val'))
 except:
 print("----------------------->>>>>>>>> examstat: error parsing metadta for study_key "+study_key)
 
 return (StudyDateTag,StudyDescriptionTag)

# custom StructType for the output tuple
XMLExtractType=StructType([
 StructField("StudyDate",StringType(),False),
 StructField("StudyDescription",StringType(),False)])

if __name__ == "__main__":
 sparkSession=SparkSession.builder \
 .appName('examstat') \
 .config('spark.cassandra.connection.host',','.join(cluster_seeds)) \
 .master('local[*]') \
 .getOrCreate()
 load_options = {"table": "new_exam", "keyspace": "examarchive"}
 sqlContext=SQLContext(sparkSession)

 # pyspark.sql.dataframe.DataFrame is already a distributed data structure. No need to parallelize it.
 df0=sqlContext.read.format('org.apache.spark.sql.cassandra').options(**load_options).load()
 df0.createTempView("new_exam")

 # pyspark.sql.functions.udf(python function,output type)
 sparkSession.udf.register("uTrimExamCode",udf(pTrimExamCode,StringType()))
 sparkSession.udf.register("uPullTags",udf(pPullTags,XMLExtractType))

 # use custom UDFs uTrimExamCode and uPullTags to calculate new columns and remove dups and deleted studies
 df1=sqlContext.sql("select study_key as StudyKey,uTrimExamCode(exam_id) as ExamCode,image_count as ImgCnt,Total_pixel_data_size as PixelSize, uPullTags(study_key,metadata_summary) as XMLExtract, metadata_summary from new_exam where exam_version=current_exam_version and is_deleted=False")
 df1.createTempView("uniq_study")
 
 # map the four fields in XMLExtract to separate columns. we take this as separate step as we don't want uPullTags to execute multiple times in previous step 
 df2=sqlContext.sql("select StudyKey,ExamCode,ImgCnt,PixelSize,XMLExtract.StudyDate as StudyDate,XMLExtract.StudyDescription as StudyDescription from uniq_study")
 df2.createTempView("uniq_study_stat")

 # Run analytical query
 df3=sqlContext.sql("SELECT ExamCode, round(avg(PixelSize)/1024/1024) as avg_size_mb, round(sum(PixelSize)/1024/1024/1024,2) as total_size_gb,count(StudyKey) as study_count FROM uniq_study_stat GROUP BY ExamCode order by study_count desc")
 #data frames are lazily loaded and processing not started until the following call
 df3.write.csv('/tmp/examstat_'+datetime.datetime.now().strftime("%m%d%H%M%S"))

It is important to understand the concept of lazy evaluation in Spark RDD here. The execution of function to RDD does not start until an action is triggered (eg. show method, or write method). Spark maintains the record of which operation is being called through DAG (directed acyclic graph). Such record is referred to as a transformation. We need to understand whether each RDD method is a transformation, or an action so we know whether it will be lazily evaluated (here’s more information).

This is a major difference between Apache Spark and Hadoop MapReduce. With MapReduce, developer spend a lot of time in minimizing the number of MapReduce passes. It happens by clubbing the operations together.

Kafka high-level Overview

Tue, 21 Jul 2020 23:19:00 -0400

Zookeeper

General definition of distributed system: a software system that is composed of independent computing entities linked together by a computer network whose components communicate and coordinate with each other to achieve a common computational goal. Implementing coordination among components of a distributed system is hard. For example, designated master node becomes single point of failure; cluster needs to detect availability of new nodes as it joins cluster.

Zookeeper is designed to simplify cluster coordination. Zookeeper implements key aspects in cluster coordination, such as distributed consensus, group management, presence protocols and leader election. In order to coordinate a cluster, zookeeper itself also runs in its own cluster, called ensemble. Zookeeper exposes a simple but powerful interface of primitives. Applications can be designed on these primitives implemented through ZooKeeper APIs to solve the problems of distributed synchronization, cluster configuration management, group membership, etc.

Zookeeper Ensemble

Clients can connect to a Zookeeper service by connecting to any member of the ensemble. The members of the ensemble are aware of each other’s state. As long as a majority of the nodes are available, the service will be available. Zookeeper cli (zkCli.sh) can be used to connect to Zookeeper server. they can be downloaded from here.

Zookeeper is integrated with many other services apart from Kafka, such as Nifi and Hadoop.

Kafka

Kafka is a messaging system that is horizontally scalable, fault tolerant. It can also serve as queue storage system and stream processing system. It is distributed and use Zookeeper for cluster coordination. Each node is called a broker.

Topics in Kafka (think of table in database) is a category or feed name to which messages (records) are published. Topic is broken up into ordered commit logs called partitions. Each partition has an ID. Each message in a partition is assigned an offset. Topics that are created in Kafka are distributed across brokers based on the partition, replication, and other factors. Each partition is replicated across several brokers depending on replication factor. For each partition, Kafka elect one replica as the leader of partition.

Writes to a partition is generally sequential. Reading messages can either be from the beginning, or rewind or skip to any port in partition given an offset value. Data in a topic is retained for a configurable period of time.

A message is a unit of data in Kafka, in the format of key-value pair. A key is used to control the message that is to be written to partitions. Messages with the same keys are always written to the same partition (hash map)

A producer publishes new message to a topic. Producers do not care which partition the message is written to and will balance messages over every partition of a topic evenly. Directing messages to a partition is done using the message key and a partitioner, this will generate a hash of the key and map it to a partition.

A consumer is subscribed to one or more topics and read messages sequentially. The consumer keeps track of messages it has consumed by keeping track on the offset of the message. The offset is a bit of metadata (an integer value that continually increases) that kafka adds to each message. Each partition has a unique offset which is stored with the offset of the last consumed message. A consumer can stop and start without losing its current state.

A Kafka broker is designed to operate as part of a cluster. One broker in the cluster also function as the cluster’s controller, which is responsible for administrative operations such as: assigning partitions to brokers; monitoring for broker failures in cluster. A particular partition is owned by a broker and that broker is called the leader of the partition.

All consumers and producers operating on that partition must connect to the leader.

Kafka cluster may replicate across cluster using MirrorMaker.

Reference: Kafka: The Definitive Guide: Real-Time Data and Stream Processing at Scale

Docker Compose, Docker Stack and Docker Swarm

Sun, 24 May 2020 21:58:03 -0400

This posting covers some basic docker orchestration tools.

Docker Compose

Docker Compose’s predecessor is a tool called Fig developed by Orchard, which was acquired by Docker in 2014, with Fig renamed to Docker Compose. Docker Compose is the official container management tool. It is essentially a python script that parses yaml file, to make Docker API calls to manage containers dynamically. It is installed along with Docker on MacOS and Windows. On Linux, you will have to download package with curl command and install manually. Docker Compose has three versions so far and we should create new template with v3.

The Docker compose yaml template consists of three parts:

services: similar to docker run
- build: specify Dockerfile to build image
- cap_add, cap_drop: specify kernel capabilities (e.g. NET_ADMIN, SYS_ADMIN)
- command: override default startup command by container
- container_name
- depends_on
- devices: map host device to container
- dns
- dns_search:
- entryppoint: override entry point from image
- env_file: specify file that stores environment variable
- environment: specify environment variable
- image: specify the location of image
- pid: share the PID namespace with host
- ports: expose network ports. HOST:CONTAINER
- networks
- volumes: mount host volume to container
networks: similar to docker network create
volumes: similar to docker volume create

Here is a typical structure of docker compose yaml template (wordpress):

version: "3.8"
services:
 mysql:
 image:mysql:5.7
 volumes:
 - mysql_data:/var/lib/mysql
 restart: always
 environment:
 MYSQL_ROOT_PASSWORD:root
 MYSQL_DATABASE:mywordpress
 MYSQL_USER:digihunch
 MYSQL_PASSWORD:hunchdigi
 wordpress:
 depends_on:
 - mysql
 image: wordpress:php7.4
 ports:
 - "8080:80"
 restart:always
 environment:
 WORDPRESS_DB_HOST:mysql:3306
 WORDPRESS_DB_USER:digihunch
 WORDPRESS_DB_PASSWORD: hunchdigi
 WORDPRESS_DB_NAME: digihunch 
networks:
 frontend:
 backend:
volumes
 mysql-data: {}

In summary, Docker Compose is an orchestration tool for single host, typically seen in development and test environment with dependencies between services.

Docker Stack

A stack is a set of related services and infrastructure that gets deployed and managed as a unit. A docker stack file has the same format as Docker Compose file, with the only requirement that the version: key specify a value of 3.0. The other difference between Docker Stacks and Docker Compose, is that stacks do not support builds. All images have to be built prior to deploying the stack.

From the stack file, Docker first executes the network section and create networks that do not exist. Then it goes through other elements. A service is a JSON collection(dictionary) that contains a bunch of keys. The image key is the only mandatory key in the service objects, which will be pulled from Docker Hub by default. Ports key maps the port of Swarm to the port of each service replica. By default, all ports are mapped using ingress mode. This means they’ll be mapped and accesible from every node in the Swarm -even nodes not running a replica. The alternative is host mode, where ports are only mapped on Swarm nodes running replicas for the service.

The environment key lets you inject environment variables into services replica.

The secrets key defines two secrets – revprox_cert and revprox_key. These must be defined in the top-level secrets key, and must exist on the system. Secrets get mounted into service replicas as a regular file. The secrets defined in this service will be mounted in each service replica as /run/secrets/revprox_cert and /run/secrets/revprox_key, unless otherwise specified.

The volumes key is used to mount pre-created volumes and host directories into a service replica.

The networks key ensures that all replicas for the service will be attached to the front-tier network. The network specified here must be defined in the networks top-level key, and if it doesn’t already exist, Docker will create it as an overlay.

The service also defines a placement constraint under the deploy key. This ensures that replicas for this service will always run on Swarm worker nodes. Placement constraints are a form of topology-aware scheduling, and can be a great way of influencing scheduling decisions.

When Docker stops a container, it issues a SIGTERM to the process with PID 1 inside the container. The container (its PID 1 process) then has a 10-second grace period to perform any clean-up operations. If it doesn’t handle the signal, it will be forcibly terminated after 10 seconds with a SIGKILL. The stop_grace_period property overrides this 10 second grace period.”
Although you may scale a docker service as part of a stack with scale command, it is not recommended. Instead, stack file should be used as the ultimate source of truth (declarative method vs imperative method). All changes to the stack should be made to the stack file, and the updated stack file used to redeploy the app.

Docker Swarm

For multi-host cluster, Docker Swarm facilitates the deployment of micro-services. Docker Swarm is:

a cluster of Docker hosts: enterprise-grade, secure communication, PKI with automation, dynamic addition of nodes
an orchestration engine, with deployment automation, deploying native swarm apps (using Docker API) and Kubernetes apps.

A Docker nodes can be physical servers, VMs, cloud instances, etc. Nodes are configured as managers or workers. Managers look after the control plane of the cluster, and dispatches tasks to workers. Managers forms a distributed management cluster on its own, and they use Raft protocol to ensure consistency. Workers accept tasks from managers and execute them. Swarm mandatorily uses TLS to encrypt communications, authenticate nodes, and authorize roles, with Automatic key rotation.

The atomic unit of scheduling on a swarm is the service. When a container is wrapped in a service, we call it a task or a replica, and the service construct adding things like scaling, rolling updates and simple rollbacks.

To initialize a swarm, we need to have the following ports open. Then we can initialize the first manager node, join additional manager nodes, and then join workers.

2377/tcp: for secure client-to-swarm communication
7946/tcp & udp: for control plane gossip
4789/udp: for VXLAN-based overlay networks

A Docker node can exist either in single-engine mode as stand alone, or in swarm mode as part of a swarm. Service only exist in swarm mode. Running docker swarm init on a Docker host in single-engine mode will switch that node into swarm mode, create a new swarm, and make the node the first manager of the swarm. Then additional nodes can be joined as managers or workers.

Swarm managers have native support for high availability, through an active-passive, multi-manager HA. Only one manager is considered active (the leader), which is the only one that will ever issue live commands against the swarm. If a passive manager receives commands for the swarm, it proxies them across to the leader.

Managers are either leaders or followers. This is Raft terminalogy because swarm uses an impelementation of the Raft consensus althorithm to power manager HA. As to HA, the following two best practices apply:

deploy an odd number of managers
don’t deploy too many managers (3 or 5 recommended, never more than 7)

Having an odd number of managers reduces the chances of split-brain conditions. Having less than 7 managers ensures that achieving consensus is quick.

With a service, we can specify name, port mappings, network to attach to, and images, as well as desired state for an application service. It is recommended in production environment to use docker-compose template to specify service. Services have replication mode, and the default is replicated. This will deploy a desired number of replicas and distribute them as evenly as possible across the cluster. The other mode is global, which runs a single replica on every node in the swarm.

Running “docker service scale” command can scale the number of service replicas from 5 to 10, which in the background updates the service’s desired state to the newly specified number of replicas. Behind the scenes, Swarm also runs a scheduling algorithm that defaults to balancing replicas as evenly as possible across the node in the swarm. Docker makes it super easy to push updates to deployed applications. With rolling update, you may specify number of replicas to update at a time, and cool-off period per update.

Here is a great article on the difference between Docker Swarm and Kubernetes.