[Varnish] #1142: nuke_limit documentation needs expanding to explain 'random' 503's

[Varnish]

#1142: nuke_limit documentation needs expanding to explain 'random' 503's
----------------------+-----------------------------------------------------
 Reporter:  timbunce  |        Type:  documentation
   Status:  new       |    Priority:  normal       
Milestone:            |   Component:  documentation
  Version:  3.0.0     |    Severity:  normal       
 Keywords:            |  
----------------------+-----------------------------------------------------
 See https://www.varnish-cache.org/trac/ticket/1012 for some background.

 The current behavior is unexpected, counter-intuitive, (seemingly)
 undocumented, hard to predict and harmful to users.

 Ideally it should be possible for varnish to 'pass-through' the backend
 response even if it can't find space in the cache. If that's not
 feasible/practical then docs need updating to remove the element of
 surprise for the users.

 Currently nuke_limit and the related behaviour is under-documented.

 I suggest that it should mention:

 * what happens when nuke_limit is reached (ie the 503 response)

 * explain that it's more likely if the site serves a mix of large and
 small objects

 * give guidance on the effect of setting it to high or two low

 * explain how to tell from the stats if this is happening

-- 
Ticket URL: <https://www.varnish-cache.org/trac/ticket/1142>
Varnish <https://varnish-cache.org/>
The Varnish HTTP Accelerator