diff options
| author |   Julien Meunier <julien.meunier@6wind.com> | 2018-03-30 10:27:52 +0200 |
|---|---|---|
| committer | Julien Meunier <julien.meunier@6wind.com> | 2018-04-10 17:09:14 +0200 |
| commit | 4ba4c2e82f8fbc59b9992f53201e26049ede44c8 (patch) | |
| tree | 2de99e59cd0b2b11467ae36ca6f35ed4f18801db /tools/report | |
| parent | 66029803af7f3d21a0697e3054b7dc3564bc5e7d (diff) | |
sysmetrics/pidstat: monitor all threads
Since sysstat commit 52977c479d3d ("Introduce new SP_VALUE_100() macro
and use it in pidstat"), pidstat can return a wrong CPU usage for all
monitored PIDs.
For example, if a process like ovs-vswitchd uses 2 threads, the CPU
usage of each thread (TID) is equal to 100%, so the CPU usage of the
main process (PID) must be equal at least to 200%. However, with this
sysstat commit, the CPU usage is restricted to 100%. It is not possible
to change this behavior.
Now, pidstat is started in order to monitor all threads created by a
process. As the output header has changed, readapt the existing code.
In order to fix this issue, only the CPU usage is accumulated with all
threads. For all other measures, main process should report correct
values.
JIRA: VSPERF-569
Change-Id: I98aa94f545d04f4de1b994c420fb5756c6f2a387
Signed-off-by: Julien Meunier <julien.meunier@6wind.com>
Diffstat (limited to 'tools/report')
| -rw-r--r-- | tools/report/report_rst.jinja | 4 |
1 files changed, 3 insertions, 1 deletions
diff --git a/tools/report/report_rst.jinja b/tools/report/report_rst.jinja index eda0c01e..6b51807a 100644 --- a/tools/report/report_rst.jinja +++ b/tools/report/report_rst.jinja @@ -90,7 +90,9 @@ Testing Activities/Events ~~~~~~~~~~~~~~~~~~~~~~~~~ pidstat is used to collect the process statistics, as such some values such as %CPU and %USER maybe > 100% as the values are summed across multiple cores. For -more info on pidstat please see: http://linux.die.net/man/1/pidstat. +more info on pidstat please see: http://linux.die.net/man/1/pidstat. Please +note that vsperf recalculates the CPU consumption of a process by aggregating +the CPU usage of each thread. Known issues: Some reported metrics have the value "unkown". These values are marked unknown as they are not values retrieved from the external tester |