TiDB Operator Architecture
This document describes the architecture of TiDB Operator and how it works.
The following diagram is an overview of the architecture of TiDB Operator.
TidbClusterAutoScaler are custom resources defined by CRD (
TidbClusterdescribes the desired state of the TiDB cluster.
TidbMonitordescribes the monitoring components of the TiDB cluster.
TidbInitializerdescribes the desired initialization Job of the TiDB cluster.
Backupdescribes the desired backup of the TiDB cluster.
Restoredescribes the desired restoration of the TiDB cluster.
BackupScheduledescribes the scheduled backup of the TiDB cluster.
TidbClusterAutoScalerdescribes the automatic scaling of the TiDB cluster.
The following components are responsible for the orchestration and scheduling logic in a TiDB cluster:
tidb-controller-manageris a set of custom controllers in Kubernetes. These controllers constantly compare the desired state recorded in the
TidbClusterobject 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-scheduleris a Kubernetes scheduler extension that injects the TiDB specific scheduling policies to the Kubernetes scheduler;
tidb-admission-webhookis a dynamic admission controller in Kubernetes, which completes the modification, verification, operation, and maintenance of Pod, StatefulSet, and other related resources.
In addition, TiDB Operator provides
tkctl, the command-line interface for TiDB clusters in Kubernetes. It is used for cluster operations and troubleshooting cluster issues.
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:
- The user creates a
TidbClusterobject and other CR objects through kubectl, such as
- TiDB Operator watches
TidbClusterand other related objects, and constantly adjust the
Service, and other objects of PD, TiKV, TiDB, Monitor or other components based on the actual state of the cluster;
- Kubernetes' native controllers create, update, or delete the corresponding
Podbased on objects such as
- In the
Poddeclaration of PD, TiKV, and TiDB, the
tidb-schedulerscheduler is specified.
tidb-schedulerapplies the specific scheduling logic of TiDB when scheduling the corresponding
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.