2
1

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

More than 3 years have passed since last update.

【2021年4月版】Kubernetes ナビ その5: Serverless

Last updated at Posted at 2021-04-09

Kubernetes ナビ

本シリーズでは kubernetes クラスタを構成する様々なコンポーネント・技術をざっとご紹介しております。

今回はその5となります。

Kubernetes で使える Serverless フレームワークいろいろ

以下で k8s にデプロイできる Serverless フレームワークをご紹介いたします。

1. Knative

元々は Google 社で開発されていたフレームワークをオープンにしたもの。Istio ベースで HTTPサービスもサポートしている。

サポートされる Event ソース

以下は公式サイトに上げられている代表的な Event コンポーネントです。

Name API Version Maintainer Description
APIServerSource v1 Knative Brings Kubernetes API server events into Knative. The APIServerSource fires a new event each time a Kubernetes resource is created, updated or deleted.
AWS SQS v1alpha1 Knative Brings AWS Simple Queue Service messages into Knative. The AwsSqsSource fires a new event each time an event is published on an AWS SQS topic.
Apache Camel v1alpha1 Knative Enables use of Apache Camel components for pushing events into Knative. A CamelSource is an event source that can represent any existing Apache Camel component, that provides a consumer side, and enables publishing events to an addressable endpoint. Each Camel endpoint has the form of a URI where the scheme is the ID of the component to use. CamelSource requires Camel-K to be installed into the current namespace. See the CamelSource example.
Apache CouchDB v1alpha1 Knative Brings Apache CouchDB messages into Knative.
Apache Kafka v1beta1 Knative Brings Apache Kafka messages into Knative. The KafkaSource reads events from an Apache Kafka Cluster, and passes these events to a sink so that they can be consumed. See the Kafka Source example for more details.
Container Source v1 Knative The ContainerSource will instantiate container image(s) that can generate events until the ContainerSource is deleted. This may be used, for example, to poll an FTP server for new files or generate events at a set time interval. Given a spec.template with at least a container image specified, ContainerSource will keep a Pod running with the specified image(s). K_SINK (destination address) and KE_CE_OVERRIDES (JSON CloudEvents attributes) environment variables are injected into the running image(s). It is used by multiple other Sources as underlying infrastructure. Refer to the Container Source example for more details.
GitHub v1alpha1 Knative Registers for events of the specified types on the specified GitHub organization or repository, and brings those events into Knative. The GitHubSource fires a new event for selected GitHub event types. See the GitHub Source example for more details.
GitLab v1alpha1 Knative Registers for events of the specified types on the specified GitLab repository, and brings those events into Knative. The GitLabSource creates a webhooks for specified event types, listens for incoming events, and passes them to a consumer. See the GitLab Source example for more details.
Heartbeats N/A Knative Uses an in-memory timer to produce events at the specified interval.
PingSource v1beta2 Knative Produces events with a fixed payload on a specified Cron schedule. See the Ping Source example for more details.
RabbitMQ Active development None
SinkBinding v1 Knative The SinkBinding can be used to author new event sources using any of the familiar compute abstractions that Kubernetes makes available (e.g. Deployment, Job, DaemonSet, StatefulSet), or Knative abstractions (e.g. Service, Configuration). SinkBinding provides a framework for injecting K_SINK (destination address) and K_CE_OVERRIDES (JSON cloudevents attributes) environment variables into any Kubernetes resource which has a spec.template that looks like a Pod (aka PodSpecable). See the SinkBinding example for more details.
WebSocket N/A Knative Opens a WebSocket to the specified source and packages each received message as a Knative event.
Auto Container Source Proof of Concept None AutoContainerSource is a controller that allows the Source CRDs without needing a controller. It notices CRDs with a specific label and starts controlling resources of that type. It utilizes Container Source as underlying infrastructure.
Amazon CloudWatch Supported TriggerMesh Collects metrics from Amazon CloudWatch.
Amazon CloudWatch Logs Supported TriggerMesh Subscribes to log events from an Amazon CloudWatch Logs stream.
Amazon CodeCommit Supported TriggerMesh Registers for events emitted by an Amazon CodeCommit source code repository.
Amazon Cognito Identity Supported TriggerMesh Registers for events from Amazon Cognito identity pools.
Amazon Cognito User Supported TriggerMesh Registers for events from Amazon Cognito user pools.
Amazon DynamoDB Supported TriggerMesh Reads records from an Amazon DynamoDB stream.
Amazon Kinesis Supported TriggerMesh Reads records from an Amazon Kinesis stream.
Amazon SNS Supported TriggerMesh Subscribes to messages from an Amazon SNS topic.
Amazon SQS Supported TriggerMesh Consumes messages from an Amazon SQS queue.
BitBucket Proof of Concept None Registers for events of the specified types on the specified BitBucket organization/repository. Brings those events into Knative.
CloudAuditLogsSource v1 Google Registers for events of the specified types on the specified Google Cloud Audit Logs. Brings those events into Knative. Refer to the CloudAuditLogsSource example for more details.
CloudPubSubSource v1 Google Brings Cloud Pub/Sub messages into Knative. The CloudPubSubSource fires a new event each time a message is published on a Google Cloud Platform PubSub topic. See the CloudPubSubSource example for more details.
CloudSchedulerSource v1 Google Create, update, and delete Google Cloud Scheduler Jobs. When those jobs are triggered, receive the event inside Knative. See the CloudSchedulerSource example for further details.
CloudStorageSource v1 Google Registers for events of the specified types on the specified Google Cloud Storage bucket and optional object prefix. Brings those events into Knative. See the CloudStorageSource example.
DockerHubSource v1alpha1 None Retrieves events from Docker Hub Webhooks and transforms them into CloudEvents for consumption in Knative.
FTP / SFTP Proof of concept None Watches for files being uploaded into a FTP/SFTP and generates events for those.
Heartbeat Proof of Concept None Uses an in-memory timer to produce events as the specified interval. Uses AutoContainerSource for underlying infrastructure.
Konnek Active Development None Retrieves events from cloud platforms (like AWS and GCP) and transforms them into CloudEvents for consumption in Knative.
K8s Proof of Concept None Brings Kubernetes cluster events into Knative. Uses AutoContainerSource for underlying infrastructure.
RedisSource v1alpha1 None Brings Redis Stream into Knative.
Slack v1alpha1 TriggerMesh Subscribes to events from Slack.
VMware Active Development None Brings vSphere events into Knative.
Zendesk v1alpha1 TriggerMesh Subscribes to events from Zendesk.

MaintainerがKnativeとなっているEventソースが公式のサポートするものです。
さすがサポートされるイベントソースが多いですね。

2. KEDA

CNCF 傘下の Event-driven Auto-Scaling プロジェクト。KEDA ではイベントソースを "Scaler" としています。システムそのものを Metrix に振り切った感じがいいですね。

サポートされる Scalers

以下は公式サイトに上げられているScalersです。

Name Availability Maintainter Description
ActiveMQ Artemis v1.5+ Community Scale applications based on ActiveMQ Artemis queues
Apache Kafka v1.0+ Microsoft Scale applications based on an Apache Kafka topic or other services that support Kafka protocol.
AWS CloudWatch v1.0+ Community Scale applications based on AWS CloudWatch.
AWS Kinesis Stream v1.1+ Community Scale applications based on AWS Kinesis Stream.
AWS SQS Queue v1.0+ Community Scale applications based on AWS SQS Queue.
Azure Blob Storage v1.1+ Community Scale applications based on the count of blobs in a given Azure Blob Storage container. Notioce:As of now, this Azure Blob Storage scaler scales based on the count of the blobs in a container as opposed to the Azure Functions behavior where code is only triggered on new blobs.
Azure Event Hubs v1.0+ Community Scale applications based on Azure Event Hubs. Notice:As of now, the Event Hub scaler only supports reading checkpoints from Blob Storage, as well as scaling only Event Hub applications written in C#, Java, Python or created with Azure Functions.
Azure Log Analytics v2.0+ Community Scale applications based on Azure Log Analytics query result
Azure Monitor v1.3+ Community Scale applications based on Azure Monitor metrics.
Azure Service Bus v1.0+ Microsoft Scale applications based on Azure Service Bus Queues or Topics.
Azure Storage Queue v1.0+ Microsoft Scale applications based on Azure Storage Queues.
CPU v2.0+ Community Scale applications based on cpu metrics.
Cron v1.5+ Community Scale applications based on a cron schedule.
External v1.0+ Microsoft Scale applications based on an external scaler.
External Push v2.0+ Microsoft Scale applications based on an external push scaler.
Google Cloud Platform‎ Pub/Sub v1.0+ Community Scale applications based on Google Cloud Platform‎ Pub/Sub.
Huawei Cloudeye v1.1+ Community Scale applications based on a Huawei Cloudeye.
IBM MQ v2.0+ Community Scale applications based on IBM MQ Queue
InfluxDB v2.1+ Community Scale applications based on InfluxDB queries
Liiklus Topic v1.0+ Community Scale applications based on Liiklus Topic.
Memory v2.0+ Community Scale applications based on memory metrics. Notice: This scaler will never scale to 0 and even when user defines multiple scaler types (eg. Kafka + cpu/memory, or Prometheus + cpu/memory), the deployment will never scale to 0.This scaler only applies to ScaledObject, not to Scaling Jobs.
Metrics API v2.0+ Community Scale applications based on a metric provided by an API
MongoDB v2.1+ Community Scale applications based on MongoDB queries.
MSSQL v2.2+ Microsoft Scale applications based on Microsoft SQL Server (MSSQL) query results.
MySQL v1.2+ Community Scale applications based on MySQL query result.
NATS Streaming v1.0+ Community Scale applications based on NATS Streaming.
OpenStack Swift v2.1+ Community Scale applications based on the count of objects in a given OpenStack Swift container.
PostgreSQL v1.2+ Community Scale applications based on a PostgreSQL query.
Prometheus v1.0+ Community Scale applications based on Prometheus.
RabbitMQ Queue Latest v1.0+ Microsoft Scale applications based on RabbitMQ Queue.
Redis Lists v1.0+ Community Scale applications based on Redis Lists.
Redis Lists (supports Redis Cluster) v2.1+ Community Redis Lists scaler with support for Redis Cluster topology
Redis Streams v1.5+ Community Scale applications based on Redis Streams.
Redis Streams (supports Redis Cluster) v2.1+ Community Redis Streams scaler with support for Redis Cluster topology

こちらの対応する Scaler もなかなかの層の厚さです。
Microsoft が意外と協力してますね。。。

3. Kubeless

Bitnamiがお届けする Serveless フレームワークです。上記のプロジェクトに比べるとまだまだ小ぢんまりしてますが、WEB UIなども提供されるので OSS版の AWS Lambda といった感じでしょうか。

サポートされるイベントトリガー

今のところサポートされるイベントトリガーは以下の通りです。

  • HTTP Trigger
  • CronJob Trigger
  • PubSub Triggers
  • Kafka Trigger

正直、こんなもんでいいのではないでしょうか・・・

4. fission

こちらもミニマルさがウリ?の fission でございます。コンポーネントにビルドツールなどがありますので、イベントドリブンに書いたコードがサクッとデプロイできる、というのは使いやすいかもしれません。

サポートされるイベントトリガー

サポートされるイベントトリガーは以下の通りです。

  • HTTP
  • NATS Streaming
  • Kafka
  • AWS SQS (KEDA based)
  • AWS Kinesis (KEDA based)
  • NATS Streaming (KEDA based)
  • Timer Triggers

次期バージョン?では KEDA連携が予定されている模様です。。。組み合わせてくるかー。。。

5. OpenWhisk

Apache 傘下の Serverless プロジェクトです。k8s 専用のプロジェクトではないのでdocker-composeやmesos、ansible、Vagrant デプロイできるらしいです。

対応言語(環境)は一通り…

  • nodejs
  • Docker
  • Python
  • go
  • Swift
  • PHP
  • Java
  • Ruby

と、対応するコンポーネントは

  • alarm
  • cloudant
  • kafka
  • pushnotification
  • rss
  • jira

などあります。

これらに対する "トリガー" と "ルール" と "アクション" を定義して対象環境にデプロイ、といった感じです。
さすが Apache、裾野が広い。。。

6. OpenFaaS

シンプルさがウリ?の Serverless フレームワーク。K8s だけではなく OpenShift にも対応している模様です。

将来的には アクションのシェアなどしたい模様がうかがえます。Serverless を突き詰めて IFTTT 的なコミュニティベースの イベント/アクション Hub みたいな感じにしたいのでしょうか。まだまだ途上のプロジェクトの模様です。

番外 Tekton

Knativeからビルドコンポーネントだけスピンアウトした CI/CD プロジェクトです。

とても分かりやすい説明がありましたので詳しくは以下のページをご覧ください。

CI/CDやるにはそれなりの"装備"が必要なので、CI/CD 自体の維持にかかるコストもバカになりません。それらもまとめて提供してくれるのはありがたいです。

雑感/サーバレスが変えるパラダイム

これまで Servelessフレームワーク を使って開発は行っていたので "Severless" という概念や仕組みは理解していましたが、単に AWS や GCP の上のサービスメニューの一つとして lambda や Functions があってそれらをどう組み合わせて使うか、という単なるオプションの一つであって、大手のクラウドプロバイダを使わないのであれば関係ない、と思っておりましたが…
k8s でガチの Serverless が、Event-Driven + MicroService のアーキテクチャが、どのようなインフラでも動くとなると話は変わってきます。
k3s や k0s であれば手元の Raspberry Pi でも Serverless は動きます。というか、Event-Driven + MicroService は IoT のエッジ側にこそふさわしい技術なんじゃないか とまで思えてきます。
エッジも Serverless でサーバーサイドも Serverless となり、これまでのWebやアプリ、クライアントを起点とする "Pull" 型のモデルから、Push型メインのモデルへと本格的にパラダイムシフトしてしまった感があります。

KEDAのSQLベースのトリガーなんてヤベーよ。。

ということで本日はここまでとしたいと思います。

参考にしたページ

2
1
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
2
1

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?