Thanks for your answering. Actually, I am not trying to do lock-and-wait on file lock level. In my app, all operation is done on single process but different threads. I just want to simulate this using thread lock. So it may not cost too much time to do thread-switching. And thread lock have a timeout. I will do some testing to check whether it isefficient. Thanks for youransweringagain.
???? ???:R Smithrsmith at rsweb.co.za ???:sqlite-userssqlite-users at mailinglists.sqlite.org ????:2016?4?6?(??)?15:45 ??:Re: [sqlite] Why SQLite use busy-retry but not lock-and-wait? On 2016/04/06 6:03 AM, sanhua.zh wrote: Recently, I am reading the source code of SQLite. I found that on OS level, SQLite use file lock to solve multi-processes problem and use VFS to solve multi-threads problem. But all of them might failed with racing and SQLite will return a SQLITE_BUSY result code to make it sleep-and-retry. I get confused of this code. Why SQLite use lock-and-wait? For example, in the unixLock, we can use F_SETLKW instead of F_SETLK, so that code will return immediatly when file unlocked. We have not need to sleep-and-retry, which may waste our running time. So I think it might be a kind oftrick, but I don?t why SQLite do so. I already find out the www.sqlite.org, but it tells nothing about this. Does anybody know why SQLite design it so? SQLite is platform independent - i.e. it works on all systems and has to make do with what is available everywhere. F_SETLKW is not available on all platforms and can't be a standard way for SQLite to interface. You can however make your own VFS for SQLite that implements it on Unix systems, if you like. This is quite easy. Does F_SETLKW have a timeout? what must SQLite do if the file still doesn't become available after a long time? How would this be better than the current way? You probably know this already, but the lock-checking mechanism is quite efficient, and a locked file gets re-checked (if a timeout is specified) very quick, then successively longer and longer intervals until the timeout expires. I doubt you will see a great increase in lock-situation performance using the file-lock wait cycle - but if you do make that VFS, perhaps we could do some testing of the hypothesis. Cheers, Ryan _______________________________________________ sqlite-users mailing list sqlite-users at mailinglists.sqlite.org http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users
