[Varnish] #64: Overflow in varnishstat

[Varnish]

#64: Overflow in varnishstat
--------------------------+-------------------------------------------------
 Reporter:  ay at linpro.no  |        Owner:  phk  
     Type:  task          |       Status:  new  
 Priority:  normal        |    Milestone:       
Component:  varnishstat   |      Version:  trunk
 Severity:  normal        |   Resolution:       
 Keywords:  overflow      |  
--------------------------+-------------------------------------------------
Changes (by phk):

  * type:  defect => task

Comment:

 This is a known issue.

 The situation is simply that maintaining the counters in question with
 proper locking would be quite expensive in terms of locking operations,
 and therefore they are not properly locked.

 If we are unlucky, the race conditions result in the loss of increments
 and therefore the counters will eventually go lower than zero.

 Doing so for an unsigned 64 bit counter gives you these incredibly large
 numbers.

 I have not decided what to do about this.  It has no practical effect, but
 it is obviously annoying.   One option is to delay the statistics and just
 keep track per worker thread and agrregate these once every N seconds
 under proper locking.

-- 
Ticket URL: <http://varnish.projects.linpro.no/ticket/64>
Varnish <http://varnish.projects.linpro.no/>
The Varnish HTTP Accelerator