BIND 10 #419: Parallel xfrout session should be allowed

[BIND 10 Development]

#419: Parallel xfrout session should be allowed
-------------------------------------+-------------------------------------
                 Reporter:           |                Owner:  zzchen_pku
  zhanglikun                         |               Status:  reviewing
                     Type:  defect   |            Milestone:  A-Team-
                 Priority:  major    |  Sprint-20110223
                Component:  xfrout   |           Resolution:
                 Keywords:           |            Sensitive:  0
Estimated Number of Hours:  0.0      |  Add Hours to Ticket:  0
                Billable?:  1        |          Total Hours:  0
                Internal?:  0        |
-------------------------------------+-------------------------------------
Changes (by vorner):

 * owner:  vorner => zzchen_pku


Comment:

 Hello

 I have serious problems understanding, why there's both
 process_request_thread that seems to finish something off and
 _handle_request_nonblock (which, in fact, seems to be blocking, so the
 name seems little bit misleading).

 However, I noticed one thing. New thread is started each time the socket
 becomes readable. However, the reading itself is done inside the thread,
 which brings a problem. The socket stays readable until everything that is
 there is read. Imagine you start thread (because the socket is readable).
 But before the thread reads all the data, another iteration of the select
 loop is run, so another thread is started to read the same data.

 And the read in the thread is not atomic. Imagine two requests come at
 almost same time and two threads are started. First, one reads the file
 descriptor and before it can read the request data, the second thread gets
 CPU and tries to read file descriptor as well. But that will fail, because
 there is the packet to handle, not fd.

 May I suggest to read the request first and start the thread after reading
 the data, only to handle the request? This way the read should be short
 (as all data of the request should be sent at once) and parallel transfers
 would be still handled in separate threads.

-- 
Ticket URL: <https://bind10.isc.org/ticket/419#comment:10>
BIND 10 Development <http://bind10.isc.org>
BIND 10 Development