aiol test doesn't increase aio-max-nr enough
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Stress-ng |
Fix Released
|
High
|
Colin Ian King |
Bug Description
Noted this on a certification run for a Purley system:
Running stress-ng aiol stressor for 240 seconds....
stress-ng: info: [55752] dispatching hogs: 208 aiol
stress-ng: info: [55752] cache allocate: using built-in defaults as unable to determine cache details
stress-ng: info: [55752] cache allocate: default cache size: 2048K
stress-ng: error: [55831] stress-ng-aiol: io_setup failed, ran out of available events, consider increasing /proc/sys/
stress-ng: error: [55832] stress-ng-aiol: io_setup failed, ran out of available events, consider increasing /proc/sys/
<snip>
stress-ng: error: [55752] process [55959] (stress-ng-aiol) aborted early, out of system resources
stress-ng: error: [55752] process [55960] (stress-ng-aiol) aborted early, out of system resources
stress-ng: info: [55752] successful run completed in 242.79s (4 mins, 2.79 secs)
return_code is 3
I wonder if it's more apropos to increase aio-max-nr before testing, or have fwts (either by default or by optional arg) increase aio-max-nr temporarily at runtime.
This is a Purley system running 4.4.0-91-generic 16.04 LTS
Changed in stress-ng: | |
status: | Incomplete → In Progress |
Changed in stress-ng: | |
importance: | Low → High |
You can increase aio-max-nr with simple commands (with root) such as
------- ------- ------- ------- ----- fs/aio- max-nr` fs/aio- max-nr
# save & change aio-max-nr
TMP_AIO=`cat /proc/sys/
echo 200000 > /proc/sys/
# doing some tests
...
# restore aio-max-nr fs/aio- max-nr ------- ------- ------- -----
echo $TMP_AIO > /proc/sys/
-------
There is no need for fwts to support changing aio-max-nr.