On Wed, Nov 27, 2024 at 03:43:11PM +0900, Damien Le Moal wrote: > I thought about that one. The problem with it is the significant performance > penalty that the context switch to the zone write plug BIO work causes. But > that is for qd=1 writes only... If that is acceptable, that solution is > actually the easiest. The overhead is not an issue for HDDs and for ZNS SSDs, > zone append to the rescue ! So maybe for now, it is best to just do that. The NOWAIT flag doesn't really make much sense for synchronous QD=1 writes, the rationale for it is not block the submission thread when doing batch submissions with asynchronous completions. So I think this should be fine (famous last words..)
