From: Nikolaus R. <Nik...@ra...> - 2010-09-28 04:39:56
|
John Bent <joh...@pu...> writes: >> >> > What we've discovered is that if we have multiple processes on a >> >> > single node doing IO to a single file that FUSE is serializing these >> >> > IO's. >> >> Yes, that's the high level API. In that case I believe you should see >> concurrent operations as long as none of the operations is a rename() or >> unlink(), the way Miklos describes in the mail you linked to. The >> options you describe are not related to this. >> > Yeah. Looks like the implementation got more than the intent. > >> I think a much easier and more reliable way to detect if there is >> concurreny problem on your system is to take the example file system, >> add a sleep(10) and a debug message into the write() function and test >> it with two concurrently running equally simple test programs that write >> at two different offsets (probably you should make them at least >> <pagesize> apart). >> >> Does that show concurrent requests for you? >> > Nope. I believe that Miklos will happily accept a patch in this case. But he'll probably reply here himself as soon as he finds the time. Best, -Nikolaus -- »Time flies like an arrow, fruit flies like a Banana.« PGP fingerprint: 5B93 61F8 4EA2 E279 ABF6 02CF A9AD B7F8 AE4E 425C |