avoid test timeouts caused by scsi_debug delay#221
avoid test timeouts caused by scsi_debug delay#221gitPo1son wants to merge 1 commit intolinux-blktests:masterfrom
Conversation
|
@gitPo1son @zhongchanghui Thanks for this PR. As I noted in the issue #219, I would like to ask the confirmation by Yu Kuai. Also, this commit needs proper commit log. Its title should have a prefix, like, "throtl/rc: ...". And the commit message should summarize what is reported in the issue #219. If that helps, I can do such commit message edit. |
5cc80ad to
0e2d304
Compare
|
Hi, @kawasaki yes, you are right, my test kernel has CONFIG_HZ=100. Thanks, |
when kernel set CONFIG_HZ=100,throtl/001 will failed on the sdebug device, due to 1 jiffy delay for each I/O takes 0.01 second,then 256 I/Os takes around 2.56 seconds which is reported as 3 seconds, making the expected 1 second test actually take 3 seconds. Signed-off-by: Changhui Zhong <czhong@redhat.com>
0e2d304 to
9cc31a8
Compare
|
I ran the throtl test group using the old kernel v6.17-rc3 and blktests with this PR. Then I confirmed that the test case throtl/002 failed with the symptom that recorded in the commit d91fa93. It indicates that the testing capability of throtle group is not affected by this PR. Good. |
When the kernel sets CONFIG_HZ=100, throtl/001 fails on the scsi_debug device. scsi_debug adds 1 jiffy delay to each I/O which takes 0.01 second. Then 256 I/Os take around 2.56 seconds which is reported as 3 seconds. This makes the test case fail which expects I/Os to complete within 1 second. To avoid the unexpected failure, set zero value to the scsi_debug delay parameter. With this, scsi_debug adds no delay, then the 256 I/Os will take less than 1 second to complete the I/Os. This behavior will be the same as null_blk. Link: #221 Signed-off-by: Changhui Zhong <czhong@redhat.com> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
|
@gitPo1son Thanks for improving the commit message. I did a bit more improvement in the commit message, and applied it to the master branch. This PR was not merged, but its purpose has been achieved. Thank you :) Let me close this. |
No description provided.