WIP: Per Job Start Delay #1164 #1224
Conversation
|
Just my feedback here, the commit message should be precise enough to explain what you want to achieve with an example. Reading code without a clear description of the goal to achieve is kind of difficult |
There was a problem hiding this comment.
House style for commit messages is:
area: brief description
Longer description here
Signed-off-by: Name <[email protected]>
- Use old-style C comment markers (
/* */) - Update the documentation
@ahribeng nice to see someone making an attempt at implementing this one!
| .is_time = 1, | ||
| .category = FIO_OPT_C_GENERAL, | ||
| .group = FIO_OPT_G_RUNTIME, | ||
| }, |
There was a problem hiding this comment.
So I can specify a per job start delay and a job delay versus the start of all jobs? I feel this would be better as a boolean unless there's a reason why you would specify both together...
| struct thread_data *td; | ||
| unsigned int i, todo, nr_running, nr_started; | ||
| struct timespec last_finish_time; | ||
| unsigned int i, todo, nr_running, nr_started, prev_nr_running; |
There was a problem hiding this comment.
I would have kept the unsigned above the struct line to make the diff smaller/more obvious.
| if (td->o.per_job_start_delay > per_job_spent) | ||
| continue; | ||
| } | ||
|
|
There was a problem hiding this comment.
So we only arrive here if we're not waiting for someone else to finish? If so why check for stonewall - surely we can always do the per job delay? I'm curious if we need the last_finish_time... What are your thoughts?
4 participants
Working on #1164 to try to get some experience pushing to FIO repo. From testing it looks like this would work, but unsure if it is exactly what is desired or the ideal way of implementing