Skip to main content

Host performance

6readings
6conditions
1themes fed
Plannedstatus

Whole-host CPU and memory posture over a measured window.

Topic id: performance.

This topic is still rolling out. Endpoints on the general release do not send it yet.

Conditions

A condition is a state that holds for a while. The agent opens it when the host enters it, keeps it open while it lasts, and closes it when the host comes back out, so one episode answers for the whole stretch instead of one alert per sample.

ConditionSeverityHow an episode ends
cpu busy (cpu_busy)Display to NoticeIt closes when the measurement falls back past its recovery point.
cpu interrupt storm (cpu_interrupt_storm)Warning to ErrorIt closes when the measurement falls back past its recovery point.
cpu kernel dominated (cpu_kernel_dominated)NoticeIt closes when any one of the several recovery conditions is met.
cpu throttled under load (cpu_throttled_under_load)Warning to ErrorIt closes on a recovery rule written for this condition, which reads more than one measurement together.
committed memory high (ram_commit_high)Display to CriticalIt closes when the measurement falls back past its recovery point.
hard fault storm (ram_hard_fault_storm)SeriousIt closes when any one of the several recovery conditions is met.

Example

Inventory (every 5 minutes)

cpu 12% avg, commit 48%, run queue 0.3/core, interrupts 1.2%, hard faults 6/s.

sparklogs.data.performance.cpu_pct_time_over_90: 0.0
sparklogs.data.performance.run_queue_per_core_p90_10s: 0.3
sparklogs.data.performance.cpu_interrupt_dpc_pct_avg: 1.2
sparklogs.data.performance.cpu_busy_pct_avg: 12.0
sparklogs.data.performance.cpu_kernel_excl_drivers_pct_of_busy_avg: 18.0
sparklogs.data.performance.commit_pct: 48.0
sparklogs.data.performance.ram_pct_time_in_hard_fault_storm: 0.0

SparkLogs: CONTEXT, Info, performance: INVENTORY: cpu 12% avg, commit 48%, run queue 0.3/core, interrupts 1.2%, hard faults 6/s.

Selected conditions

cpu_busy

CPU is busy.

Example

started; time over 90% CPU 100% (threshold 80%)

sparklogs.data.performance.cpu_pct_time_over_90: 100.0
sparklogs.data.performance.run_queue_per_core_p90_10s: 5.0
sparklogs.data.performance.cpu_busy_age_h: 0.0

SparkLogs: cpu_busy, Notice, performance: cpu_busy: NOTABLE: started; time over 90% CPU 100% (threshold 80%)

cpu_interrupt_storm

CPU time is dominated by interrupt and DPC handling.

Impact: Device or driver interrupt load can starve ordinary work on the host.

Example

started; interrupt+DPC 34% (threshold 30%)

sparklogs.data.performance.cpu_interrupt_dpc_pct_avg: 34.0
sparklogs.data.performance.cpu_interrupt_storm_age_h: 0.0

SparkLogs: cpu_interrupt_storm, Warning, performance: cpu_interrupt_storm: NOTABLE: started; interrupt+DPC 34% (threshold 30%)

CaseSeverityTicket class
onsetTrace to Fatalperformance
heldTrace to Fatalperformance
recoveredTrace to Fatalperformance

cpu_kernel_dominated

Busy CPU time is mostly kernel time rather than application work.

Impact: Application throughput can be lower than the busy figure alone suggests.

Example

started; CPU busy 84% (threshold 70%)

sparklogs.data.performance.cpu_busy_pct_avg: 84.0
sparklogs.data.performance.cpu_kernel_excl_drivers_pct_of_busy_avg: 85.26
sparklogs.data.performance.cpu_kernel_dominated_age_h: 0.0

SparkLogs: cpu_kernel_dominated, Notice, performance: cpu_kernel_dominated: NOTABLE: started; CPU busy 84% (threshold 70%)

CaseSeverityTicket class
onsetTrace to Fatalperformance
heldTrace to Fatalperformance
recoveredTrace to Fatalperformance

cpu_throttled_under_load

The CPU is running below its rated frequency while under load.

Impact: Work takes longer than the hardware would otherwise allow; thermal, power or firmware limits are the usual cause.

Example

started; CPU busy 82

sparklogs.data.performance.cpu_throttled_under_load_age_h: 0.0

SparkLogs: cpu_throttled_under_load, Warning, performance: cpu_throttled_under_load: NOTABLE: started; CPU busy 82

CaseSeverityTicket class
onsetTrace to Fatalperformance
heldTrace to Fatalperformance
recoveredTrace to Fatalperformance

ram_commit_high

Committed memory is high.

Example

started; commit 92% (threshold 90%)

sparklogs.data.performance.commit_pct: 92.0
sparklogs.data.performance.ram_commit_high_age_h: 0.0

SparkLogs: ram_commit_high, Notice, performance: ram_commit_high: NOTABLE: started; commit 92% (threshold 90%)

ram_hard_fault_storm

The host is thrashing memory.

Impact: Paging pressure can degrade every workload on the host.

Example

started; commit 97% (threshold 95%)

sparklogs.data.performance.commit_pct: 97.0
sparklogs.data.performance.ram_pct_time_in_hard_fault_storm: 100.0
sparklogs.data.performance.ram_hard_fault_storm_age_h: 0.0

SparkLogs: ram_hard_fault_storm, Serious, performance: ram_hard_fault_storm: NOTABLE: started; commit 97% (threshold 95%)

CaseSeverityTicket class
onsetTrace to Fatalperformance
heldTrace to Fatalperformance
recoveredTrace to Fatalperformance