On Wed, Mar 16, 2016 at 10:38 AM, Ryan Moats <rmo...@us.ibm.com> wrote:
> > ---- Original Message ---- > > Currently ovsdb_jsonrpc_session are grouped together in a linked > > list within 'ovsdb_jsonrpc_remote'. This makes sense since most > > session operations applies to sessions within a remote. > > > > However, in order to scale up ovsdb-server with multi-threading, it is > > more convenient to distribute a sessions to any thread available, > > regardless which remote it is associated with. > > > > This patch introduces a set of APIs that provide operations on > > a list of sessions. Instead of group sessions by remote, they > > are linked together in a new ovs_list field 'all_sessions' in the > > ovsdb_jsonrpc_server struct. > > > > With multi-threading, the design is that all sessions managed > > by a thread will have them linked together on a thread private > > linked list. At that time, the 'all_sessions' field in > > ovsdb_jsonrpc_server struct will have all session managed > > the main process. > > > > Signed-off-by: Andy Zhou <az...@ovn.org> > > I've given this patch a bit of a test spin and while it appears > relatively sane for simple cases, I'm concerned that the patch > should be using LIST_FOR_EACH_SAFE in more places than it currently > does. The code snippet that got me wondering was the use of > LIST_FOR_EACH in ovsdb_jsonrpc_sessions_count()...\ > Thanks for testing it. LIST_FOR_EACH_SAFE is required to protect against possible list elements removal during the list traversal. Since a session is always assigned to a single thread, I don't see how it can be remove from the list during the ovsdb_jsonprc_session_count(), Or I am still missing something here. > > Ryan (regXboi) > _______________________________________________ > dev mailing list > dev@openvswitch.org > http://openvswitch.org/mailman/listinfo/dev > _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
