subroutine & action - varnish

 

subroutine & action - varnish

 

subroutine, code section, called in the flow, manipulate http header and other things,

action, indicate the next step, used inside subroutine, in format "return (action);" ,

 

see: https://www.varnish-cache.org/docs/3.0/reference/vcl.html#subroutines

 

------

format of subroutine

 

sub xxx {

...

return (action);

}

 

------

call a subrouine

 

format:

call xxx;

 

what we do:

we just re-define subroutine logic, it will be used in the flow automatically,

 

------

terminate of subroutine

 

use:

return(action);

 

'action' is the result which indicate the next step,

 

------

multi subroutine

 

if more than one subroutine with the same name is defined, they will be concated into a single subroutine in the order defined,

 

------

action

 

actions:

* error code [reason]

return error with specified reason, and abandon the request,

* pass

switch to pass mode, 

if current subroutine is not vcl_pass, control will be pass to vcl_pass,

* pipe

switch to pipe mode, 

if current subroutine is not vcl_pipe, control will be pass to vcl_pipe,

* lookup

lookup the requested object in the cache,

if found, control will be pass to vcl_hit, otheriwse to vcl_miss,

* deliver

deliver data to client, 

if current subroutine is not vcl_deliver, control will be pass to vcl_deliver,

* restart

restart the request, and increase the restart counter, if the counter > max_restarts, an error will be send,

* fetch

retrieve the requested object from the backend, 

if current subroutine is not vcl_fetch, control will be pass to vcl_fetch,

* ok

means ok, used in vcl_init & vcl_fini,

* hit_for_pass

used in vcl_fetch,

create a hit_for_pass object, set the object's ttl to the current value of beresp.tll,

current request will be handled by vcl_deliver, 

but subsequence request will go directly to vcl_pass according to hit_for_pass object, means use vcl_pass and cache will not be used until the hit_for_pass object's ttl expired,

* 

 

return the same action:

if a subroutine return a action the same as it self, that means the request is already proceed and finished in this way,

e.g.

in vcl_pipe(), return (pipe); , means request is proceed by the way pipe,

 

------

subroutines

 

* vcl_init

* vcl_recv

* vcl_pipe

* vcl_pass

* vcl_hash

* vcl_hit

* vcl_miss

* vcl_fetch

* vcl_deliver

* vcl_error

* vcl_fini

 

------

vcl_init

 

for init, called before any request come,

 

possible action:

* ok

vcl will go on

* 

 

------

vcl_recv

 

called when request is received and parsed,

 

decide:

* whether to serve this request

* how to do it

* which backend to use

 

possible action:

* error code

* pass

* pipe

* lookup

* 

 

------

vcl_pipe

 

request is passed to backend, varnish act just like a pipe between client & backend for furture data until the connection is closed,

 

possible action:

* error code

* pipe

indicate proceed & finished in pipe mode,

* 

 

------

vcl_pass

 

similar as vcl_pipe, but the response from backend don't enter cache, means the response will not be cached,

the subsequence request of the same connection is proceed normally,

 

possible action:

* error code

* pass

indicate proceed & finished in pipe mode,

* restart

* 

 

------

vcl_hash

 

add data to hash, 

call hash_data() to add a data,

 

possible action:

* hash

Proceed,

* 

 

------

vcl_hit

 

called after a cache is found in lookup,

 

possible action:

* deliver

* error code

* pass

* restart

* 

 

------

vcl_miss

 

called after a cache not found in lookup, 

 

decide:

* whether to attempt to retrieve the document from backend,

* which backend to use

 

possible action:

* error code

* pass

* fetch

 

------

vcl_fetch

 

called after a document is fetched from backend,

 

possible action:

* deliver

add object to cache, then deliver it,

* error code

* hit_for_pass

* restart

* 

 

------

vcl_deliver

 

called before a object is deliver to the client,

 

possible action:

* deliver

deliver object to client,

* error code

* restart

* 

 

------

vcl_error

 

called when an error occur,

 

possible action:

* deliver

deliver the error to client,

* restart

* 

 

------

vcl_fini

 

clean up VCL, called after all request have exited VCL,

 

possible action:

* ok

vcl will be discarded,

* 

 

------