[Varnish] #280: Workspace overflows give confusing error messaqe

[Varnish]

#280: Workspace overflows give confusing error messaqe
----------------------+-----------------------------------------------------
 Reporter:  marlier   |        Owner:  phk      
     Type:  defect    |       Status:  new      
 Priority:  normal    |    Milestone:  After 2.0
Component:  varnishd  |      Version:  trunk    
 Severity:  normal    |   Resolution:           
 Keywords:            |  
----------------------+-----------------------------------------------------
Changes (by tfheen):

  * summary:  VCL fault with restart logic if backend returns 403 =>
              Workspace overflows give confusing error
              messaqe
  * milestone:  => After 2.0

Old description:

> Issue description:
> With restart logic enabled, "403: Forbidden" responses are not handled
> correctly.  It appears that they loop forever (sending requests to the
> backend, or backends), until they cause this fault:
>
> {{{
> Child (25712) died signal=6
> Child (25712) Panic message: Missing errorhandling code in cnt_lookup(),
> cache_center.c line 558:
>   Condition((p) != 0) not true.  thread = (cache-worker)sp = 0x857546c {
>   fd = 9, id = 9, xid = 1541762087,
>   client = 127.0.0.1:35186,
>   step = STP_LOOKUP,
>   err_code = 403, err_reason = (null),
>   ws = 0x85754b4 { overflow
>     id = "sess",
>     {s,f,r,e} = {0x8575934,,+8168,(nil),+8192},
>   },
>     worker = 0x6f2970e0 {
>     },
>     vcl = {
>       srcname = {
>         "/etc/varnish/restarts.vcl",
>         "Default",
>       },
>     },
>   backend = 0x855fc98 {
>     vcl_name = "tea",
>   },
> },
>

> Child cleanup complete
> }}}
>

> Steps to reproduce:
> This issue is consistently reproducible.
>
> * Configure varnish with a backend that includes forbidden URLs.
> * Add logic to vcl_fetch that will cause a restart if the returned object
> has a response code other than 200
> * Request the forbidden URL via varnish (eg, with wget)
>

> Workaround:
> There is a simple workaround to this issue.  Instead of enabling restart
> logic with this:
> {{{
> sub vcl_fetch {
>     if (obj.status != 200) {
>         restart;
>     }
> }
> }}}
>
> use this:
> {{{
> sub vcl_fetch {
>     if (obj.status != 200 && obj.status != 403) {
>         restart;
>     }
> }
> }}}
>
> Attached VCL file is broken, and includes the workaround commented out.

New description:

 Issue description:

 If a request overflows the workspace (such as if the configuration uses
 "restart" and it ends up in an endless loop), the child dies with a
 somewhat confusing error message such as:

 {{{
 Child (25712) died signal=6
 Child (25712) Panic message: Missing errorhandling code in cnt_lookup(),
 cache_center.c line 558:
   Condition((p) != 0) not true.  thread = (cache-worker)sp = 0x857546c {
   fd = 9, id = 9, xid = 1541762087,
   client = 127.0.0.1:35186,
   step = STP_LOOKUP,
   err_code = 403, err_reason = (null),
   ws = 0x85754b4 { overflow
     id = "sess",
     {s,f,r,e} = {0x8575934,,+8168,(nil),+8192},
   },
     worker = 0x6f2970e0 {
     },
     vcl = {
       srcname = {
         "/etc/varnish/restarts.vcl",
         "Default",
       },
     },
   backend = 0x855fc98 {
     vcl_name = "tea",
   },
 },


 Child cleanup complete
 }}}

 We should try to detect what's happened and report a better error.
 Ideally, the child should not die.

-- 
Ticket URL: <http://varnish.projects.linpro.no/ticket/280#comment:3>
Varnish <http://varnish.projects.linpro.no/>
The Varnish HTTP Accelerator