(* $Id: netplex_main.mli 1836 2013-02-10 20:58:39Z gerd $ *)
(** Main program for Netplex servers *)
open Netplex_types
type cmdline_config
val args :
?defaults:cmdline_config ->
unit -> ((Arg.key * Arg.spec * Arg.doc) list * cmdline_config)
(** [let (opt_list, cmdline_cfg) = args()]:
* Returns [opt_list] for inclusion in the [Arg.parse] option list.
* The effects made available by the returned [cmdline_cfg] value.
*
* The defaults (unless overridden by [defaults]): The config file
* is derived from the name of the executable (by appending [.conf]
* instead of the current extension). There is no config tree.
*
* @param defaults The default argument values
*)
val create : ?config_filename:string ->
?config_tree:config_tree ->
?pidfile:string option ->
?foreground:bool ->
unit -> cmdline_config
(** Creates the command-line configuration object.
By setting [config_tree] a special configuration can be injected -
the file [config_filename] is not loaded in this case (but may
still appear in error messages)
*)
val modify : ?config_filename:string ->
?config_tree:config_tree ->
?pidfile:string option ->
?foreground:bool ->
cmdline_config -> cmdline_config
(** Modifies the command-line configuration object *)
val config_filename : cmdline_config -> string
(** Returns the filename of the configuration file, or the default
if it has not been set on the command-line
*)
val config_filename_opt : cmdline_config -> string option
(** Returns the filename of the configuration file, or [None] if it
has not been set on the command-line
*)
val config_tree_opt : cmdline_config -> config_tree option
(** Returns the tree of the configuration file, or [None] if it
has not been set by [create] or [modify]. Note that [args]
never sets the config tree.
*)
val pidfile : cmdline_config -> string option
(** Returns the location of the PID file (if any) *)
val foreground : cmdline_config -> bool
(** Returns whether the daemon runs in the foreground *)
val startup :
?late_initializer:(config_file -> controller -> unit) ->
?config_parser:(string -> config_file) ->
parallelizer ->
logger_factory list ->
workload_manager_factory list ->
processor_factory list ->
cmdline_config ->
unit
(** Establishes a configuration and starts the Netplex daemon.
*
* If a ready-made configuration tree is included in the passed configuration
* it is taken as the configuration. If otherwise only the name of the
* config file is available, this file is parsed.
* Fails with [Netplex_config.Config_error] when an error in the
* configuration is detected.
*
* The [late_initializer] is called after the Netplex controller has been
* fully initialized, and before the main event loop is entered. You can
* perform here further initializations, e.g. starting helper threads.
*
* The [config_parser] is by default [Netplex_config.read_config_file].
* You can override it by whatever parser you would like to use. The
* parser is only used if a configuration tree is unavailable, and
* the configuration needs to be loaded from a file.
*
* As side-effect, the current logger of the {!Netlog} module is set
* to selected Netplex logger. Note that this setting remains active
* even after [startup] returns to the caller. Also note that log messages
* submitted via {!Netlog} appear as from component "netplex.other"
* if they are sent from outside of a container.
*
* It is changed to [/] as working directory (as required for a daemon).
*)
val run :
?config_parser:(string -> config_file) ->
late_initializer:(config_file -> controller -> 'a) ->
extract_result:(controller -> 'a -> 'b) ->
parallelizer ->
logger_factory list ->
workload_manager_factory list ->
processor_factory list ->
cmdline_config ->
'b
(** Similar to [startup], but this function is tailored for Netplex
systems that compute results (e.g. Netmulticore jobs). In this
scenario, it is
not sufficient to just fire off child processes, but you also want
to return a result to the caller once the processes have finished
their tasks. One consequence of this is that you cannot daemonize
the program - it always runs in the foreground (no matter whether
[-fg] is given or not).
[run] expects that the caller passes a function [extract_result].
This function is invoked once the Netplex system is done, and
the user can extract the final result from the controller. The
result of [extract_result] is also the result of [run].
At the moment [extract_result] is invoked, the child processes are
already terminated. They should have put their result at a place
that is in the controller, or can be reached with the help of the
controller. The controller functionality is still fully available.
Note that the controller is shut down as soon as [extract_result]
returns. Further interactions with the controller are not possible.
The [late_initializer] is now obligatory for typing reasons
(so you can get the result of the [late_initializer] from
[extract_result]). If you don't need it, just return [()].
The working directory is left unchanged.
*)
(** {2 Tutorial}
*
* The typical main program for a [Netplex] server system looks like:
*
* {[
* let my_factories = ...
*
* let start() =
* let opts, cmdconf = Netplex_main.args() in
* Arg.parse
* opts
* (fun s -> raise(Arg.Bad ("Unknown arg: " ^ s)))
* "usage: protoserver";
* let par = Netplex_mp.mp() in (* or Netplex_mt.mt() *)
* Netplex_main.startup
* par
* Netplex_log.logger_factories
* Netplex_workload.workload_manager_factories
* my_factories
* cmdconf
*
* Sys.set_signal Sys.sigpipe Sys.Signal_ignore;
* start()
* ]}
*
* This main program enables:
* - The standard command-line arguments [-conf], [-pid] and [-fg] are
* understood
* - The configuration file is parsed
* - The configuration can refer to all loggers and workload managers
* coming with Netplex
* - The parallelizer is selected: Here, it is multi-processing
* (Netplex_mp). One could also select multi-threading (Netplex_mt)
* - The processors defined by [my_factories] are made available
* for connection processing
*)
