YugabyteDB supports additional attributes for latency metrics, which enable you to calculate throughput.

These attributes include the following:

For example, if SELECT * FROM table is executed once and returns 8 rows in 10 microseconds, the handler_latency_yb_ysqlserver_SQLProcessor_SelectStmt metric would have the following attribute values: total_count=1, total_sum=10, min=10, max=10, and mean=10. If the same query is run again and returns in 6 microseconds, then the attributes would be as follows: total_count=2, total_sum=16, min=6, max=10, and mean=8.

Although these attributes are present in all handler_latency metrics, they may not be calculated for all the metrics.

YSQL query processing

YSQL query processing metrics represent the total inclusive time it takes YugabyteDB to process a YSQL statement after the query processing layer begins execution. These metrics include the time taken to parse and execute the SQL statement, replicate over the network, the time spent in the storage layer, and so on. The preceding metrics do not capture the time to deserialize the network bytes and parse the query.

The following are key metrics for evaluating YSQL query processing. All metrics are counters and units are microseconds.

The YSQL throughput can be viewed as an aggregate across the whole cluster, per table, and per node by applying the appropriate aggregations.

Database IOPS (reads and writes)

The YB-TServer is responsible for the actual I/O of client requests in a YugabyteDB cluster. Each node in the cluster has a YB-TServer, and each hosts one or more tablet peers.

The following are key metrics for evaluating database IOPS. All metrics are counters and units are microseconds.

These metrics can be viewed as an aggregate across the whole cluster, per table, and per node by applying the appropriate aggregations.