On 2024-03-23 14:25, Maciej W. Rozycki wrote:
On Sat, 23 Mar 2024, Johnny Billquist wrote:As for recovering - well. It's not like the system can just throw the disk reads/writes away if you have a timeout. It's going to keep trying at it. The only other option is to just crash.Well, the usual outcome is supposed to be EIO rather than a crash, though there are more complex scenarios too, e.g. when a page cannot be faulted in from storage.
True, if we're talking about a direct write from user level to disk, sure. But when it hits the cache, then you won't notice until the cache is flushed to disk. Where would the EIO go then, and what is the expected outcome? Same with meta-data. And yes, paging is yet another issue in there.
Reads are generally not going to cause as much buffer starvations I think. Writes are the bigger problem in my head when just thinking about it.
But I might be missing something...?
Johnny
--
Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: bqt%softjar.se@localhost || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol