Plasma GitLab Archive
Projects Blog Knowledge

Class Http_client.pipeline


class pipeline : object .. end
A pipeline is a queue of HTTP calls to perform

A

pipeline object is a FIFO queue of HTTP calls. It is called "pipeline" because it is processed asynchronously: Requests may be sent to the HTTP server independently of whether responses of the previous requests already arrived or not.

Furthermore, a pipeline object may keep connections to several servers at once. (More exactly, it has a FIFO queue for every server it is connected with.)

The pipeline object keeps track what is happening, so you need not to care about the details of communications. The API is simple: Create a pipeline object, do some setup (add authentication methods; configure the proxy to use), add the requests, and run the pipeline. The rest is done automatically. To get the results, you can either memorize the requests you wanted to know yourself and ask every request object about the reply of the server; or you can specify that a callback function should be called once the request is processed (with positive or negative result). It is possible to add further requests to the pipeline from within these callback functions.

If you want to have several pipelines, or some cooperation with other network services, you may specify a Unixqueue.event_system. For example, to have two pipelines working concurrently:

 let ues = Unixqueue.create_unix_event_system() in
 let p1 = new pipeline in
 let p2 = new pipeline in
 p1 # set_event_system ues;
 p2 # set_event_system ues;
 Unixqueue.run ues             (* run p1 and p2 in parallel *)
 

This works not only with pipelines, but with every network client or server which is compatible with the Unixqueue design.

method event_system : Unixqueue.event_system
Returns the event system
method set_event_system : Unixqueue.event_system -> unit
Sets the event system. Must be called before the first call is added
method connection_cache : connection_cache
The current connection cache. By default, a private restrictive cache is used.
method set_connection_cache : connection_cache -> unit
Set the connection cache. This must happen before the first call is added.
method add_authentication_method : basic_auth_method -> unit
adds an old-style authentication method
method add_auth_handler : auth_handler -> unit
adds a new-style authentication handler
method set_proxy : string -> int -> unit
set_proxy name port: sets that a proxy name listening on port is to be used
method set_proxy_auth : string -> string -> unit
sets user and password for the proxy. Only the "basic" authentication method is implemented.
method avoid_proxy_for : string list -> unit
sets a list of host names or domain suffixes for which no proxy should be used. e.g. "localhost"; ".our.net"
method set_proxy_from_environment : unit -> unit
Inspect the environment variables http_proxy and no_proxy and set the proxy options from them.
method reset : unit -> unit
Empties the pipeline and inactivates any open connection. The currently active operation is interrupted, and every request with response is set to No_reply (i.e. you get the exception No_reply if you try to access the response). If there are callbacks for these requests, the callback functions are invoked. The queues of open requests and replies are cleared. All connections to all servers are inactivated.

Inactivation means that open connections are given back to the connection cache for further reuse if the state of the connection allows this; otherwise the connections are closed.

method add : http_call -> unit
Adds the call to the end of the pipeline. One must not add calls that have already been served.
method add_with_callback : http_call -> (http_call -> unit) -> unit
Adds the call to the end of the pipeline.

After the call has been processed, the callback function is called. This function is called for every call that leaves the pipeline, it does not matter whether processing was successful or not. Invoke status on the message to get what happened; either some status information from the server is available (perhaps OK status), or an exception is indicated.

method run : unit -> unit
Runs through the requests in the pipeline. If a request can be fulfilled, i.e. the server sends a response, the state of the request is set and the request is removed from the pipeline. If a request cannot be fulfilled (no response, bad response, network error), the exception is stored in the http_call object and will be raised once the state of the object is queried.

Under certain conditions (serious network errors) run does not catch the exception; it simply cleans its own state up (aborting the errorneous network connection). In this case, simply invoke run again to continue. run terminates normally if the pipeline becomes empty.

The engine handles the following HTTP return codes itself:

  • 100: This is an intermediate return code
  • 301: The redirection is followed if configured
  • 302: The redirection is followed if configured
  • 401: Content server authentication
  • 407: Proxy server authentication
All other return codes remain uninterpreted, it is up to the caller of this function to react on them.
method get_options : http_options
method set_options : http_options -> unit
Get/Set the available options for the HTTP engine. The new options will take into effect immediately.
method number_of_open_messages : int
Returns the number of messages which are still in the pipeline.
method number_of_open_connections : int
Returns the number of connections which are open at the same time and currently being used by this object (i.e. connections returned to the cache do not count)
method connections : (string * int * int) list
Reports which connections exist: (host, port, queue_length)
method cnt_new_connections : int
Counts new connections (or better: attempts to establish connections)
method cnt_timed_out_connections : int
Counts connections given up because of timeouts
method cnt_crashed_connections : int
Counts connections with network or protocol errors
method cnt_server_eof_connections : int
Counts connections the server terminated with EOF
method cnt_successful_connections : int
Counts connections closed because pipelines become empty
method cnt_failed_connections : int
Counts totally failed connections (no more reconnects allowed)
method reset_counters : unit -> unit

Notes on counters:

  • cnt_new_connections: Is increased when a new connection attempt is started (that may fail or timeout in the future). Reconnects do not count.
  • cnt_timed_out_connections: Is increased whenever an established connection times out. Usually, it is immediately reconnected.
  • cnt_crashed_connections: Is increased whenever an established connection crashes. Usually, it is immediately reconnected.
  • cnt_failed_connections: Is increased when a timed out or crashed connection exceeds the maximum number of errors, and it is not tried to reconnect.
  • cnt_successful_connections: Is increased when all HTTP calls have been replied.
When the client has done all of its jobs, we have

 cnt_new_connections = cnt_failed_connections + cnt_successful_connections 
This web site is published by Informatikbüro Gerd Stolpmann
Powered by Caml