TiDB Operator Architecture

This document describes the architecture of TiDB Operator and how it works.

Architecture

The following diagram is an overview of the architecture of TiDB Operator.

TidbCluster, TidbMonitor, TidbInitializer, Backup, Restore, BackupSchedule, and TidbClusterAutoScaler are custom resources defined by CRD (CustomResourceDefinition).

• TidbCluster describes the desired state of the TiDB cluster.
• TidbMonitor describes the monitoring components of the TiDB cluster.
• TidbInitializer describes the desired initialization Job of the TiDB cluster.
• Backup describes the desired backup of the TiDB cluster.
• Restore describes the desired restoration of the TiDB cluster.
• BackupSchedule describes the scheduled backup of the TiDB cluster.
• TidbClusterAutoScaler describes the automatic scaling of the TiDB cluster.

The following components are responsible for the orchestration and scheduling logic in a TiDB cluster:

• tidb-controller-manager is a set of custom controllers in Kubernetes. These controllers constantly compare the desired state recorded in the TidbCluster object with the actual state of the TiDB cluster. They adjust the resources in Kubernetes to drive the TiDB cluster to meet the desired state and complete the corresponding control logic according to other CRs;
• tidb-scheduler is a Kubernetes scheduler extension that injects the TiDB specific scheduling policies to the Kubernetes scheduler;
• tidb-admission-webhook is a dynamic admission controller in Kubernetes, which completes the modification, verification, operation, and maintenance of Pod, StatefulSet, and other related resources.
• discovery is a service for inter-components discovery. Each TiDB cluster contains a discovery Pod which is used for the components to discover other existing components in the same cluster.

Control flow

The following diagram is the analysis of the control flow of TiDB Operator. Starting from TiDB Operator v1.1, the TiDB cluster, monitoring, initialization, backup, and other components are deployed and managed using CR.

The overall control flow is described as follows:

1. The user creates a TidbCluster object and other CR objects through kubectl, such as TidbMonitor;
2. TiDB Operator watches TidbCluster and other related objects, and constantly adjust the StatefulSet, Deployment, Service, and other objects of PD, TiKV, TiDB, Monitor or other components based on the actual state of the cluster;
3. Kubernetes' native controllers create, update, or delete the corresponding Pod based on objects such as StatefulSet, Deployment, and Job;
4. If you configure the components to use tidb-scheduler in the TidbCluster CR, the Pod declaration of PD, TiKV, and TiDB specifies tidb-scheduler as the scheduler. tidb-scheduler applies the specific scheduling logic of TiDB when scheduling the corresponding Pod.

Based on the above declarative control flow, TiDB Operator automatically performs health check and fault recovery for the cluster nodes. You can easily modify the TidbCluster object declaration to perform operations such as deployment, upgrade, and scaling.