1runqslower(8) System Manager's Manual runqslower(8)
2
3
4
6 runqslower - Trace long process scheduling delays.
7
9 runqslower [-p PID] [-t TID] [min_us]
10
12 This measures the time a task spends waiting on a run queue (or equiva‐
13 lent scheduler data structure) for a turn on-CPU, and shows occurrences
14 of time exceeding passed threshold. This time should be small, but a
15 task may need to wait its turn due to CPU load. The higher the CPU
16 load, the longer a task will generally need to wait its turn.
17
18 This tool measures two types of run queue latency:
19
20 1. The time from a task being enqueued on a run queue to its context
21 switch and execution. This traces ttwu_do_wakeup(), wake_up_new_task()
22 -> finish_task_switch() with either raw tracepoints (if supported) or
23 kprobes and instruments the run queue latency after a voluntary context
24 switch.
25
26 2. The time from when a task was involuntary context switched and still
27 in the runnable state, to when it next executed. This is instrumented
28 from finish_task_switch() alone.
29
30 The overhead of this tool may become significant for some workloads:
31 see the OVERHEAD section.
32
33 This works by tracing various kernel scheduler functions using dynamic
34 tracing, and will need updating to match any changes to these func‐
35 tions.
36
37 Since this uses BPF, only the root user can use this tool.
38
40 CONFIG_BPF and bcc.
41
43 -h Print usage message.
44
45 -p PID Only show this PID (filtered in kernel for efficiency).
46
47 -t TID Only show this TID (filtered in kernel for efficiency).
48
49 min_us Minimum scheduling delay in microseconds to output.
50
52 Show scheduling delays longer than 10ms:
53 # runqslower
54
55 Show scheduling delays longer than 1ms for process with PID 123:
56 # runqslower -p 123 1000
57
59 TIME Time of when scheduling event occurred.
60
61 COMM Process name.
62
63 PID Process ID.
64
65 LAT(us)
66 Scheduling latency from time when task was ready to run to the
67 time it was assigned to a CPU to run.
68
70 This traces scheduler functions, which can become very frequent. While
71 eBPF has very low overhead, and this tool uses in-kernel maps for effi‐
72 ciency, the frequency of scheduler events for some workloads may be
73 high enough that the overhead of this tool becomes significant. Measure
74 in a lab environment to quantify the overhead before use.
75
77 This is from bcc.
78
79 https://github.com/iovisor/bcc
80
81 Also look in the bcc distribution for a companion _examples.txt file
82 containing example usage, output, and commentary for this tool.
83
85 Linux
86
88 Unstable - in development.
89
91 Ivan Babrou, original BCC Python version Andrii Nakryiko, CO-RE version
92
94 runqlen(8), runqlat(8), pidstat(1)
95
96
97
98USER COMMANDS 2016-02-07 runqslower(8)