SingleStore Managed Service

Simplified Statistics

The statistics listed in the table below are present in the following management views in the information_schema database:

  • mv_activities

  • mv_activities_cumulative

Column name

Description

elapsed_time_ms

Wall clock milliseconds elapsed during execution.

cpu_time_ms

Milliseconds spent running on all the CPUs across the cluster.

cpu_wait_time_ms

Milliseconds spent waiting for a CPU to become available. A non-zero value indicates that there were more concurrent tasks in the workload than there were CPUs on which to run them. If this value is a significant fraction of elapsed time, the workload is likely issuing its queries faster than the hardware can handle. Investigate the tasks with the highest cpu_time_ms and optimize the relevant queries, reduce your workload, or provision more CPUs for your cluster. Note that this column is also an Advanced Statistics, and requires that the global variable read_advanced_counters is set to ON.

disk_time_ms

Milliseconds spent waiting for physical disk I/O to complete. If this value is high, it may indicate excessive time to commit transactions to storage. You may need hardware which supports a faster write rate, or you may need to investigate whether other processes are issuing too much I/O and introducing unwanted variability. Check which other activities are using disk_physical_read_b or disk_physical_write_b. Note that this column is also an Advanced Statistics, and requires that the global variable read_advanced_counters is set to ON.

disk_b

Number of bytes that SingleStore DB read and wrote to the filesystem or the in memory transaction log.

lock_time_ms

Milliseconds spent waiting on locks and condition variables, excluding row locks. This isn’t solely wait time due to contention on data structures because it also includes some time spent waiting for events in general.

memory_bs

Byte-seconds of memory used.

network_b

Number of bytes SingleStore DB sent and received to sockets.

network_time_ms

Milliseconds spent waiting for sockets to be ready to send or receive data. This may indicate that leaves are slow to respond to a query, or that a distributed join involves a large data transfer. If this counter has a noticeable delay, check what other activities are consuming network_logical_recv_b and network_logical_send_b.

Notice

For simple statistics, elapsed_time_ms is aggregated over aggregators only, whereas all other statistics show total usage over both aggregators and leaves. For this reason, it is not meaningful to compare elapsed_time_ms to cpu_time_ms.