[BUG] Linux Host CPU is shown at 100% all the time (SQL Server CPU consumption probably correct)

[bogdanben2000]

Component

Full Dashboard

Performance Monitor Version

2.11.0

SQL Server Version

SQL Server 2019 CU12

Windows Version

Linux (OpenSuse)

Describe the Bug

Result of Top command:

Steps to Reproduce

.

Expected Behavior

.

Actual Behavior

.

Error Messages / Log Output

.

Screenshots

.

Additional Context

Graph with Host CPU (not just SQL Server CPU consumption) would help

[erikdarlingdata]

Thanks for the clear report — the top output alongside the Dashboard screenshot made the root cause easy to pin down.

What's happening: the host CPU figure is derived from the RING_BUFFER_SCHEDULER_MONITOR ring buffer as 100 - SystemIdle - SQLProcessUtilization. On SQL Server on Linux, SystemIdle is always reported as 0 — it's a documented platform limitation (Microsoft's own sample query even carries the comment -- SystemIdle on Linux will be 0). So the math collapses to 100 - 0 - sqlcpu, and the host total pins at 100% no matter what the box is actually doing. Your SQL Server CPU number was correct the whole time; only the host/"other process" figure was fabricated. Unfortunately there's no DMV that exposes true host CPU on Linux.

The fix (just merged to dev in #1049): the collector now detects Linux and stores the host figure as "unavailable" instead of a bogus value, and the Dashboard falls back to showing the correct SQL-Server-only CPU rather than a flat 100%.

Could you help verify on the nightly build? I don't have a Linux SQL host to reproduce on, so a confirmation from your environment would be hugely helpful. The fix lands in the next nightly (auto-built from dev around 06:00 UTC daily):

1. Grab the latest assets from the nightly release — you'll want both PerformanceMonitorInstaller-*.zip and PerformanceMonitorDashboard-*.zip. (Make sure the version stamp is dated after 2026-06-03 so it includes the fix.)
2. Run the CLI installer against your instance to upgrade the PerformanceMonitor database — this deploys the Linux-aware collector and a small schema migration.
3. Let the CPU collector run a cycle or two, then open the upgraded nightly Dashboard.

A couple of things to expect:

• Only data collected after the upgrade is corrected — the collector doesn't rewrite history, so the existing 100% points will still be there. Look at fresh samples.
• On Linux the chart will show the SQL CPU line only (no separate host line), since real host CPU isn't obtainable from SQL Server there. If you need true OS-level CPU on Linux, that has to come from an OS tool (top, node_exporter, etc.).

If the new samples track your actual SQL CPU instead of sitting at 100%, that confirms the fix. Thanks again!

[bogdanben2000]

Hi,

Thanks for helping.
I ran the nightly installer on the SQL Server hosted on OpenSuse with existing database overwrite.
When I open the nightly dashboard I have the same alert regarding host CPU:

Thanks

[erikdarlingdata]

Thanks for testing and for the screenshot — that pinned it down. The #1049 fix was incomplete, and your result showed exactly where.

What was still wrong: #1049 corrected every path that reads CPU from the collected data (the chart, the views, the stored collector), so the CPU chart should now track your real SQL CPU. But the alert you're seeing comes from a different code path — the Dashboard's alert engine runs its own live query against the scheduler ring buffer and computes host CPU as 100 - SystemIdle - ProcessUtilization. That one query was never touched by #1049, so on Linux (where SystemIdle is always 0) it kept returning 100 - sqlcpu, and the total pinned at 100% → the alert fired forever.

The fix (#1055, just merged to dev): the alert engine now uses the same Linux detection as the collector and reports host/other CPU as unavailable on Linux, falling back to the correct SQL-only figure. The alert should clear.

Good news — this one is Dashboard-only. No schema or database change, so you do not need to re-run the installer this time. Just:

1. Grab the next nightly PerformanceMonitorDashboard-*.zip (version stamp dated after 2026-06-04).
2. Open it against your existing upgraded database.

The host-CPU alert should be gone, and the CPU chart should show your SQL CPU line only (real host CPU still isn't obtainable from SQL Server on Linux — that needs an OS tool like top or node_exporter).

Sorry for the round trip, and thanks again for verifying on real hardware — I don't have a Linux SQL host to reproduce on, so your confirmation is what makes this fixable. A quick note on whether the alert clears on the new nightly would close this out.

[bogdanben2000]

Hi,

With the latest nightly release it seems that the issue is not occurring anymore:

Btw, Other: 0% it is showing the real CPU usage of the other process running on the Linux VM (except SQL) or it was set to 0% just to avoid the issue? The question is if you get the other process CPU usage from SQL Server, but using omething similar to xp_cmdshell on Windows maybe.

Thanks