On Mon, Feb 27, 2012 at 02:04:54PM -0800, Ethan Jackson wrote: > > I would argue that run_connection() should only read a new request > > if the connection is not backlogged (this is already tested in > > unixctl_server_wait() so maybe that was your intention all along?). > > > > I don't think that run_connection() really needs a loop. One try > > should be enough. (I think I must have overthought the logic in > > there.) Ditto for the loop around pstream_accept() in > > unixctl_server_run(). > > I'm not sure I agree with this entirely. What if a client sends > multiple requests in a row very quickly. Doesn't it make sense to > process all of them at once in a single run loop instead of having to > go through an entire run loop per request? I suppose the problem with > the current code is that a client could arbitrarily fill the server's > backlog by sending lots of requests. It certainly would be easiest to > deal with this by not processing requests until the backlog is empty. > Alternatively we could set a maximum on the backlog and not process > requests when that value is hit. I'm not sure what the correct thing > to do is, do you have thoughts?
I was suggesting an optional simplification. Obviously when I wrote the existing code I had the idea of back-to-back requests in mind, or I wouldn't have written it that way. But the only existing client only sends a single request per program execution, so multiple back-to-back requests just isn't an important case. The existing code is also fine, if you prefer it. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
