[QFJ-484] Session responder lock change inadequate Created: 17/Nov/09 Updated: 29/Apr/11 Resolved: 29/Apr/11 |
|
Status: | Closed |
Project: | QuickFIX/J |
Component/s: | Engine |
Affects Version/s: | None |
Fix Version/s: | None |
Type: | Bug | Priority: | Default |
Reporter: | James Olsen | Assignee: | Unassigned |
Resolution: | Won't Fix | Votes: | 0 |
Labels: | None |
Attachments: | QFJ-484(1_5_0).patch QFJ-484.patch | ||||||||
Issue Links: |
|
Description |
Using a synchronized block can result in a thread being starved of access to the responder. This should be changed to use a Lock with fair semantics. Starvation was observed in a scenario using SocketSynchronousWrites=Y where a slow consumer could not be disconnected because the message writing thread kept re-obtaining the monitor. |
Comments |
Comment by James Olsen [ 17/Nov/09 ] |
Relates to change proposed by |
Comment by James Olsen [ 17/Nov/09 ] |
No option to attach patch, so will try in-lining: Index: core/src/main/java/quickfix/Session.java import quickfix.Message.Header; private boolean enabled;
@@ -330,19 +332,25 @@
@@ -1436,13 +1444,16 @@
|
Comment by James Olsen [ 16/Feb/10 ] |
Attached Parwdiner's change is critical when using synchronous writes when you have multiple sessions running. |
Comment by James Olsen [ 17/Jan/11 ] |
Copy of patch as applied to 1_5_0 Session source. |
Comment by Eric Deshayes [ 28/Apr/11 ] |
committed on integration branch rev #1024 |
Comment by Eric Deshayes [ 29/Apr/11 ] |
After discussion, the usage of Reentrant Lock would likely cause some throughput/latency performance. |