QPS: The number of SQL statements executed per second on all TiDB instances, which is counted according to SELECT, INSERT, UPDATE, and other types of statements.
connection_count: The  change in connection count may be caused by changes in workload or performance problems of the TiDB cluster. If there is no major change in workload, please check the comparison of other duration-related metrics and analyze the specific reasons.
Active_connection: Active connection is the new active value of each tidb, and workload changes may lead to changes in this data.
Disconnection: The number of lost connections for each tidb. If it increases, you need to check the workload changes or the status of the tidb nodes.
CPS_by_Type: CPS stands for Command Per Second, and Command stands for the command type of the MySQL protocol. Changes in this indicator may be affected by workload changes or cluster performance.
Queries_Using_Plan_Cache_OPS : The number of times the TiDB cluster hits the execution plan cache per second changes. If it decreases, you need to check the workload or whether the business layer request uses the plan cache correctly.
Database_Time_by_SQL_Type: Represents database time, and changes in workload will result in numerical changes.
Database_Time_by_SQL_Phase: Represents database time, and changes in workload will result in numerical changes.
SQL_Execute_Time_Overview: For the time-consuming of Execute, split it according to the dismantling of different stages, if a huge change is found, it is necessary to check the business change.
Failed_Query_OPM_Detail: If the value is greater than 0, it indicates that there are failed Queries, and it is recommended to troubleshoot the corresponding SQL.
Duration: Duration includes the average latency for each SQL type and overall.
Connection_Idle_Duration: This value represents the idle time of the connection in or outside the transaction. If the value is large, check whether there is a problem on the application side.
P99_Get_Token_Duration: The get token phase is usually only a few microseconds long. In case of the number of connections of a TiDB instance reaches the token limit, the connection will be limited when creating a new connection.
Parse_Duration: The Parse stage is one of the stages of SQL processing inside TiDB, and it is the time to parse the query statement into an abstract syntax tree (AST). If the time of the Parse phase increases significantly, the possible reason is that the query statement is very long, and the text parsing consumes a lot of CPU.
Compile_Duration: The Compile stage is one of the stages of SQL processing inside TiDB. It is the stage of compiling an execution plan based on the AST and statistical information output by the parse stage. The delay in the compile phase may reach a few milliseconds or tens of milliseconds. If the execution plan cache cannot be hit, logical and physical optimizations are required in the compile phase, which will consume a lot of CPU and memory, further affecting the performance of other TiDB components.
Execute_duration: The execute phase in the internal processing of SQL in TiDB includes TSO waiting time, KV request time, and data processing time by the TiDB executor itself. If the workload remains unchanged and the indicator increases significantly, need to check the potential performance problem of the cluster.
KV_request_OPS: The total number of KV requests per second for all TiDB instances. If the workload does not change and the number of requests decreases significantly, need to check the TiKV performance.
Avg_TiDB_KV_Request_Duration: The average latency of KV requests measured by TiDB. If this metric increases significantly, it may be caused by a network problem or a TiKV performance bottleneck.
Avg_TiKV_GRPC_Duration: The average delay of GRPC message processing in TiKV. If this metric increases significantly, it may be caused by a network problem or a TiKV performance bottleneck.
TSO_request_OPS: The get token phase is usually only a few microseconds long and can be ignored. Unless the number of connections of a single instance of TiDB reaches the token limit , the connection will be limited when creating a connection.
PD_TSO_Wait_Duration: The TiDB executor waits for the TSO delay (wait). If the metric data increases significantly, it may be caused by a network problem or a PD performance bottleneck.
PD_TSO_RPC_Duration:  Network latency for TSO requests (rpc). If the metric data increases significantly, it may be caused by a network problem or a PD performance bottleneck.
TiDB_Memory: The memory usage of each TiDB instance. It is divided into the memory occupied by the process and the memory requested by Golang on the heap.
TiDB_CPU: CPU usage per TiDB instance.
TiKV_Memory: Memory usage per TiKV instance.
TiKV_CPU: CPU usage per TiKV instance.
TiKV_Disk_IO_MBPS: Write and read speed per TiKV instance.
TiKV_Storage_Usage: Percentage of storage space used by each TiKV instance.
PD_CPU: CPU usage per PD instance.
KV_request_time: Time waiting for  KV request during SQL execution. The total KV request wait time might exceed SQL execution time, because KV requests are concurrent.
KV_request_ops: Total number of KV requests per second in all TiDB instances.
TPS: Transactions per second.