Query I/O statistics and latency distribution using views - PolarDB

Use the polar_stat_io_info and polar_stat_io_latency views to diagnose I/O performance at the file level — including operation counts, throughput, and latency distributions broken down by file type and I/O kind.

Prerequisites

Before you begin, ensure that you have:

The polar_monitor extension installed in the database

To install the extension, run:

CREATE EXTENSION polar_monitor;

I/O statistics view

polar_stat_io_info answers the question: how much I/O is happening, and how long is it taking? Query this view to find which file type is generating the most read or write traffic, or which operation is accumulating the most latency.

Column	Type	Description
`filetype`	text	The file type, such as `WAL`, `DATA`, and `LOG`.
`fileloc`	text	The local file.
`open_count`	numeric	Number of file open operations.
`open_latency_us`	double precision	Total latency of file open operations, in microseconds.
`close_count`	numeric	Number of file close operations.
`read_count`	numeric	Number of file read operations.
`write_count`	numeric	Number of file write operations.
`read_throughput`	numeric	Total bytes read.
`write_throughput`	numeric	Total bytes written.
`read_latency_us`	double precision	Total latency of read operations, in microseconds.
`write_latency_us`	numeric	Total latency of write operations, in microseconds.
`seek_count`	numeric	Number of file seek operations.
`seek_latency_us`	double precision	Total latency of seek operations, in microseconds.
`creat_count`	numeric	Number of file create operations.
`creat_latency_us`	double precision	Total latency of file create operations, in microseconds.
`fsync_count`	numeric	Number of fsync operations.
`fsync_latency_us`	double precision	Total latency of fsync operations, in microseconds.
`falloc_count`	numeric	Number of falloc operations.
`falloc_latency_us`	double precision	Total latency of falloc operations, in microseconds.

I/O latency distribution view

polar_stat_io_latency answers the question: where does latency concentrate? Instead of an average, it shows how many operations completed within each latency bucket, making it easier to detect tail latency.

Column	Type	Description
`iokind`	text	The I/O operation type, such as `OPEN`, `CREATE`, `READ`, and `WRITE`.
`num_lessthan200us`	numeric	Operations that completed in less than 200 microseconds.
`num_lessthan400us`	numeric	Operations that completed in less than 400 microseconds.
`num_lessthan600us`	numeric	Operations that completed in less than 600 microseconds.
`num_lessthan800us`	numeric	Operations that completed in less than 800 microseconds.
`num_lessthan1ms`	numeric	Operations that completed in less than 1 millisecond.
`num_lessthan10ms`	numeric	Operations that completed in less than 10 milliseconds.
`num_lessthan100ms`	numeric	Operations that completed in less than 100 milliseconds.
`num_morethan100ms`	numeric	Operations that took longer than 100 milliseconds.