[Dev] Tizen 3.0 proposal for fixing OSP/WRT/Core hard-coded UID issue

Jussi Laako jussi.laako at linux.intel.com
Thu Oct 17 09:50:21 GMT 2013


On 17.10.2013 12:31, Patrick Ohly wrote:
> For example, exclusive write access to the sqlite DB in EDS or the Tizen
> contacts service simplifies change notification, because a single
> process knows what changes are made at a semantic level and can send
> change notifications to readers more efficiently. With shared write
> access, all a reader can get from the OS is "file modified".

You can have an auxiliary notification queue with sqlite or you could 
use postgresql and it's NOTIFY/LISTEN.

> There's also nothing that prevents multithreading in such a design. I
> don't see how you came to that conclusion. EDS is internally
> multithreaded and allows concurrent reads directly from the sqlite file
> (only writing is centralized).

There's a write bottleneck, because with record-locking you could 
simultaneously write to different portions of file without conflicts. 
But unfortunately sqlite doesn't support that, so it needs some more 
advanced database.

When you expand this, or read/write locking (for multi-reader, 
single-writer) outside of the process you already solved the problem.



More information about the Dev mailing list