Support deploying one TiDB Cluster across multiple regions or data centers

[DanielZhangQD]

trafficstars

Description

Several users request the ability to deploy TiDB Clusters across multiple Kubernetes clusters that deployed in different regions or data centers to achieve HA in region or data center level. In addition, users can also deploy applications in different regions/data centers to access the TiDB/TiKV Pods in the local region/data center to reduce the latency.

Category

Feature

Value

• Achieve HA in the region or data center level
• Reduce the latency of cross-region access

TODO list

• P0 features

• [x] Support configuring cluster domain in the TidbCluster CR #2929

• [x] Create headless for discovery #2930

• [x] Support configuring Discovery address in the TidbCluster CR #2933

• P1 features

• [x] TLS Support for across Kubernetes deployment #2934

• [x] Support failover for across Kubernetes clusters deployment #2935

• [ ] Monitoring aggregation for the TidbClusters deployed across Kubernetes clusters #3118

• [x] e2e for across Kubernetes clusters deployment #3510

• P2 features

• [x] Docs for across Kubernetes clusters deployment #2936

• P3 features

• [x] Docs for across Kubernetes clusters deployment on EKS #3175

• [x] Docs for across Kubernetes clusters deployment on GKE #3176

• [x] Deploy heterogeneous cluster across Kubernetes clusters https://github.com/pingcap/tidb-operator/issues/3169

• [ ] Consider introducing a federated TidbCluster CR for the management of the different CRs in different Kubernetes cluster.

• [ ] Failover if one Kubernetes cluster is down

Workload Estimation

60

Time

DevProgress: 45% QAProgress: 0% DocProgress: 0%

GanttProgress: 20% GanttStart: 2020-07-13 GanttDue: 2020-12-30

[mikechengwei]

I can help with this feature.

[DanielZhangQD]

@mikechengwei Thanks for your help! Which issue above do you want to take?

[mikechengwei]

I can finish PO feature and i can do #2929 firstly. I have a question about other cluster join current cluster by specify discovery url. Should the domain systems of these two clusters ensure that they can communicate with each other?

[mikechengwei]

I go to complete other issues, this pr is left to others to do.

[L3T]

/assign

[DanielZhangQD]

/assign @L3T

[scsldb]

Reason for reprioritization: development complete, testing in progress, more important current development tasks.