Edit

Share via

sys.dm_os_linux_disk_stats (Transact-SQL)

Applies to: SQL Server 2025 (17.x) on Linux

Returns a table with disk I/O statistics for each Linux device, showing total activity beyond SQL Server, in SQL Server 2025 (17.x) Cumulative Update (CU) 1 and later versions.

The DMV returns one row per Linux disk device.

Column name Data type Nullable Description
dev_name nvarchar(256) No Device name.
major_num bigint No Major device number.
minor_num bigint No Minor device number.
reads_completed bigint No Number of reads completed.
reads_merged bigint No Number of adjacent reads merged into a single request.
sectors_read bigint No Number of sectors read.
read_time_ms bigint No Milliseconds spent servicing reads.
writes_completed bigint No Number of writes completed.
writes_merged bigint No Number of adjacent writes merged into a single request.
sectors_written bigint No Number of sectors written.
write_time_ms bigint No Milliseconds spent servicing writes.
ios_in_progress bigint No Number of IOs currently in request queues.
io_time_ms bigint No Milliseconds the device spent doing I/O.
weighted_io_time_ms bigint No Weighted number of milliseconds spent doing I/Os.

Permissions

Requires VIEW SERVER PERFORMANCE STATE permission on the server.

Remarks

  • Host level scope: The results reflect all activity on the device, including SQL Server, other services, and background system operations. To attribute database workload effects, correlate these results with SQL Server wait statistics, such as WRITELOG and PAGEIOLATCH_*.

  • Interpret time columns:

    Measure Calculation
    Average read latency (ms/op) read_time_ms / NULLIF(reads_completed, 0)
    Average write latency (ms/op) write_time_ms / NULLIF(writes_completed, 0)
    Device utilization (%) over an interval io_time_ms / (elapsed_ms) * 100
    Average queue length weighted_io_time_ms / elapsed_ms

  • Sector units: Linux commonly reports sectors in 512-byte units. Validate your environment's sector size and adjust calculations accordingly.

  • Use with care in multitenant hosts: High values can originate from non-SQL Server workloads. Correlate with system tools or other DMVs for attribution.

Use this DMV with other Linux-specific DMVs for holistic monitoring:

Usage scenarios

  • Log flush slowness (availability groups or standalone): High write_time_ms / writes_completed with elevated io_time_ms indicates device-level write latency. Corroborate with WRITELOG waits.

  • Checkpoint or read heavy workloads: Rising read_time_ms / reads_completed with sustained read_MBps suggests throughput constrained by latency. Consider storage tier or queue settings.

  • Noisy neighbors on shared hosts: Spikes in device_utilization_pct and avg_queue_length without corresponding SQL Server workload changes imply external I/O pressure. Validate with host monitoring.

Examples

A. Current device activity snapshot

This query returns an activity snapshot for the current storage device.

SELECT dev_name,
       reads_completed,
       read_time_ms,
       writes_completed,
       write_time_ms,
       ios_in_progress,
       io_time_ms,
       weighted_io_time_ms
FROM sys.dm_os_linux_disk_stats
ORDER BY io_time_ms DESC;

B. Latency and throughput over a 10-second sample window

The script samples the DMV twice and computes deltas for latency (milliseconds per operation), throughput (MB per second), utilization (percentage), and queue length. Adjust @SectorBytes to match your device configuration.

DECLARE @SectorBytes AS INT = 512;

-- verify sector size for your environment
DECLARE @SampleMs AS INT = 10000;

IF OBJECT_ID('tempdb..#before') IS NOT NULL
    DROP TABLE #before;

IF OBJECT_ID('tempdb..#after') IS NOT NULL
    DROP TABLE #after;

SELECT dev_name,
       reads_completed,
       reads_merged,
       sectors_read,
       read_time_ms,
       writes_completed,
       writes_merged,
       sectors_written,
       write_time_ms,
       ios_in_progress,
       io_time_ms,
       weighted_io_time_ms
INTO #before
FROM sys.dm_os_linux_disk_stats;

WAITFOR DELAY '00:00:10';

SELECT dev_name,
       reads_completed,
       reads_merged,
       sectors_read,
       read_time_ms,
       writes_completed,
       writes_merged,
       sectors_written,
       write_time_ms,
       ios_in_progress,
       io_time_ms,
       weighted_io_time_ms
INTO #after
FROM sys.dm_os_linux_disk_stats;

WITH deltas
AS (SELECT a.dev_name,
           a.reads_completed - b.reads_completed AS d_reads,
           a.read_time_ms - b.read_time_ms AS d_read_ms,
           a.sectors_read - b.sectors_read AS d_read_sectors,
           a.writes_completed - b.writes_completed AS d_writes,
           a.write_time_ms - b.write_time_ms AS d_write_ms,
           a.sectors_written - b.sectors_written AS d_write_sectors,
           a.io_time_ms - b.io_time_ms AS d_io_ms,
           a.weighted_io_time_ms - b.weighted_io_time_ms AS d_weighted_io_ms
    FROM #after AS a
         INNER JOIN #before AS b
             ON a.dev_name = b.dev_name)
SELECT dev_name,
       -- latency (ms/op)
       CAST (d_read_ms / NULLIF (d_reads, 0) AS DECIMAL (18, 2)) AS avg_read_latency_ms,
       CAST (d_write_ms / NULLIF (d_writes, 0) AS DECIMAL (18, 2)) AS avg_write_latency_ms,
       -- throughput (MB/s)
       CAST ((d_read_sectors * @SectorBytes) / (@SampleMs / 1000.0) / 1048576.0 AS DECIMAL (18, 2)) AS read_MBps,
       CAST ((d_write_sectors * @SectorBytes) / (@SampleMs / 1000.0) / 1048576.0 AS DECIMAL (18, 2)) AS write_MBps,
       -- utilization (%)
       CAST (d_io_ms / @SampleMs * 100.0 AS DECIMAL (5, 2)) AS device_utilization_pct,
       -- average queue length
       CAST (d_weighted_io_ms / @SampleMs AS DECIMAL (18, 2)) AS avg_queue_length
FROM deltas
ORDER BY device_utilization_pct DESC;

C. Identify devices with high merge ratios

This example identifies devices with high merge ratios, indicating possible I/O coalescing.

SELECT dev_name,
       reads_completed,
       reads_merged,
       writes_completed,
       writes_merged,
       CAST (reads_merged / NULLIF (reads_completed, 0) AS DECIMAL (10, 2)) AS read_merge_ratio,
       CAST (writes_merged / NULLIF (writes_completed, 0) AS DECIMAL (10, 2)) AS write_merge_ratio
FROM sys.dm_os_linux_disk_stats
ORDER BY write_merge_ratio DESC, read_merge_ratio DESC;

D. Watch queue depth live

This example shows the live queue depth, which is useful for incident triage.

SELECT TOP (20) dev_name,
                ios_in_progress,
                io_time_ms,
                weighted_io_time_ms
FROM sys.dm_os_linux_disk_stats
ORDER BY ios_in_progress DESC,
         weighted_io_time_ms DESC;

Related content

  • Last updated on