On 12/18/24 11:00 PM, Christoph Hellwig wrote: > On Tue, Dec 17, 2024 at 12:41:50PM -0700, Jens Axboe wrote: >> As per earlier replies, it's either -EAGAIN being mishandled, OR it's >> driving more IOs than the device supports. For the latter case, io_uring >> will NOT block, but libaio will. > > And that's exactly the case here as zoned devices only support a single > I/O to a zone and will have to block for the next one. > >> Like Christoph alluded to in his first reply, driving more than 1 >> request inflight is going to be trouble, potentially. > > If you rely on order. If you are doing O_APPEND-style I/O on zonefs or > using a real file systems it's perfectly fine. Yes this is the way, rather than attempt guarantee ordering for multiple IOs. It's fragile and will repeatedly break. -- Jens Axboe
