mirror of
https://github.com/prometheus-community/windows_exporter.git
synced 2026-10-30 15:39:04 +01:00
contrib: expand Windows mixin rules and publish runbooks (#2575)
Signed-off-by: Jan-Otto Kröpke <mail@jkroepke.de>
This commit is contained in:
@@ -0,0 +1,37 @@
|
||||
---
|
||||
title: WindowsCPUHighUsage
|
||||
---
|
||||
|
||||
## Meaning
|
||||
|
||||
Average non-idle CPU usage across logical processors exceeds the configured
|
||||
threshold for 15 minutes. The default is 90%, measured over a five-minute rate
|
||||
window. This measures CPU time; Task Manager's frequency-adjusted utility can
|
||||
show a different value.
|
||||
|
||||
## Impact
|
||||
|
||||
The host has little CPU capacity for additional work. Applications may show
|
||||
increased latency, timeouts, or reduced throughput.
|
||||
|
||||
## Diagnosis
|
||||
|
||||
1. Inspect `windows:cpu_usage:ratio` for the affected host and compare it with
|
||||
its workload and normal baseline.
|
||||
2. Inspect `windows:cpu_time:ratio_rate` by `mode`. High user time points toward
|
||||
application work; high privileged, interrupt, or DPC time can indicate kernel
|
||||
or driver activity. DPC and interrupt modes overlap privileged time; do not
|
||||
sum all modes to estimate usage.
|
||||
3. Use Task Manager, Resource Monitor, or Performance Monitor to identify the
|
||||
busy processes. Compare individual core usage to detect a single-threaded
|
||||
bottleneck.
|
||||
4. Check scheduled tasks, antivirus scans, recent deployments, and hypervisor
|
||||
CPU contention.
|
||||
|
||||
## Mitigation
|
||||
|
||||
Reduce or redistribute the workload, reschedule expensive background tasks, or
|
||||
add CPU capacity. Investigate a process or driver consuming unexpected CPU
|
||||
before restarting it. Confirm CPU usage and application latency return to normal.
|
||||
For expected sustained workloads, tune the threshold against service objectives
|
||||
and available capacity.
|
||||
Reference in New Issue
Block a user