(define-module (8sync actors)
#:use-module (oop goops)
#:use-module (srfi srfi-9)
- #:use-module (srfi srfi-9 gnu)
#:use-module (ice-9 control)
#:use-module (ice-9 format)
#:use-module (ice-9 match)
+ #:use-module (ice-9 atomic)
#:use-module (ice-9 pretty-print)
+ #:use-module (ice-9 receive)
+ #:use-module (fibers)
+ #:use-module (fibers channels)
+ #:use-module (fibers conditions)
+ #:use-module (fibers operations)
#:use-module (8sync agenda)
+ #:use-module (8sync inbox)
#:use-module (8sync rmeta-slot)
#:export (;; utilities... ought to go in their own module
big-random-number
big-random-number-string
- simple-message-id-generator
<actor>
actor-id
actor-message-handler
- %current-actor
-
;;; Commenting out the <address> type for now;
;;; it may be back when we have better serializers
;; <address>
make-address address?
address-actor-id address-hive-id
+ actor-init! actor-cleanup!
+
address->string
actor-id-actor
actor-id-hive
define-actor
- <hive>
- make-hive
- ;; There are more methods for the hive, but there's
- ;; no reason for the outside world to look at them maybe?
- hive-id
+ ;; <hive>
+ ;; make-hive
+ ;; ;; There are more methods for the hive, but there's
+ ;; ;; no reason for the outside world to look at them maybe?
+ ;; hive-id
bootstrap-actor bootstrap-actor*
create-actor create-actor*
message-auto-reply?
- <- <-* <-wait <-wait* <-reply <-reply* <-reply-wait <-reply-wait*
-
- call-with-message mbody-receive mbody-val
+ <- <-wait
- run-hive
- bootstrap-message
-
- serialize-message write-message
- serialize-message-pretty pprint-message
- read-message read-message-from-string))
+ spawn-hive run-hive))
;; For ids
-(define %random-state
- (make-parameter (random-state-from-platform)))
+(set! *random-state* (random-state-from-platform))
;; Same size as a uuid4 I think...
(define random-number-size (expt 2 128))
(define (big-random-number)
- (random random-number-size (%random-state)))
+ (random random-number-size))
;; Would be great to get this base64 encoded instead.
(define (big-random-number-string)
;; @@: This is slow. Using format here is wasteful.
(format #f "~x" (big-random-number)))
-;; @@: This is slow. A mere ~275k / second on my (old) machine.
+;; @@: This is slow-ish. A mere ~275k / second on my (old) machine.
;; The main cost seems to be in number->string.
(define (simple-message-id-generator)
;; Prepending this cookie makes message ids unique per hive
;;; Messages
;;; ========
-
-;; @@: We may want to add a deferred-reply to the below, similar to
-;; what we had in XUDD, for actors which do their own response
-;; queueing.... ie, that might receive messages but need to shelve
-;; them to be acted upon after something else is taken care of.
-
(define-record-type <message>
(make-message-intern id to from action
- body in-reply-to wants-reply
- replied)
+ body in-reply-to wants-reply)
message?
+ ;; @@: message-ids are removed. They could be re-enabled
+ ;; if we had thread-safe promises...
(id message-id) ; id of this message
(to message-to) ; actor id this is going to
(from message-from) ; actor id of sender
(action message-action) ; action (a symbol) to be handled
(body message-body) ; argument list "body" of message
(in-reply-to message-in-reply-to) ; message id this is in reply to, if any
- (wants-reply message-wants-reply) ; whether caller is waiting for reply
- (replied message-replied ; was this message replied to?
- set-message-replied!))
+ (wants-reply message-wants-reply)) ; whether caller is waiting for reply
(define* (make-message id to from action body
- #:key in-reply-to wants-reply
- replied)
+ #:key in-reply-to wants-reply)
(make-message-intern id to from action body
- in-reply-to wants-reply replied))
-
-(define (message-auto-reply? message)
- (eq? (message-action message) '*auto-reply*))
-
-(define (message-needs-reply? message)
- "See if this message needs a reply still"
- (and (message-wants-reply message)
- (not (message-replied message))))
-
+ in-reply-to wants-reply))
(define (kwarg-list-to-alist args)
(let loop ((remaining args)
;; This is the internal, generalized message sending method.
;; Users shouldn't use it! Use the <-foo forms instead.
-;; @@: Could we get rid of some of the conditional checks through
-;; some macro-foo?
-
-(define-inlinable (send-message send-options from-actor to-id action
- replying-to-message wants-reply?
- message-body-args)
- (if replying-to-message
- (set-message-replied! replying-to-message #t))
- (let* ((hive (actor-hive from-actor))
- (new-message
- (make-message (hive-gen-message-id hive) to-id
- (actor-id from-actor) action
- message-body-args
- #:wants-reply wants-reply?
- #:in-reply-to
- (if replying-to-message
- (message-id replying-to-message)
- #f))))
- (if wants-reply?
- (abort-to-prompt (hive-prompt (actor-hive from-actor))
- from-actor new-message send-options)
- ;; @@: It might be that eventually we pass in send-options
- ;; here too. Since <-wait and <-reply-wait are the only ones
- ;; that use it yet, for now it kind of just makes things
- ;; confusing.
- (8sync (hive-process-message hive new-message)))))
-
-(define (<- to-id action . message-body-args)
- "Send a message from an actor to another actor"
- (send-message '() (%current-actor) to-id action
- #f #f message-body-args))
-
-(define (<-* send-options to-id action . message-body-args)
- "Like <-*, but allows extra parameters via send-options"
- (define* (really-send #:key (actor (%current-actor))
- #:allow-other-keys)
- (send-message send-options actor to-id action
- #f #f message-body-args))
- (apply really-send send-options))
-
-(define (<-wait to-id action . message-body-args)
- "Send a message from an actor to another, but wait until we get a response"
- (wait-maybe-handle-errors
- (send-message '() (%current-actor) to-id action
- #f #t message-body-args)))
-
-(define (<-wait* send-options to-id action . message-body-args)
- "Like <-wait, but allows extra parameters, for example whether to
-#:accept-errors"
- (define* (really-send #:key (actor (%current-actor))
- #:allow-other-keys)
- (apply wait-maybe-handle-errors
- (send-message send-options actor to-id action
- #f #t message-body-args)
- send-options))
- (apply really-send send-options))
-
-;; TODO: Intelligently ~propagate(ish) errors on -wait functions.
-;; We might have `send-message-wait-brazen' to allow callers to
-;; not have an exception thrown and instead just have a message with
-;; the appropriate '*error* message returned.
-
-(define (<-reply original-message . message-body-args)
- "Reply to a message"
- (when (message-needs-reply? original-message)
- (send-message '() (%current-actor) (message-from original-message) '*reply*
- original-message #f message-body-args)))
-
-(define (<-reply* send-options original-message . message-body-args)
- "Like <-reply, but allows extra parameters via send-options"
- (define* (really-send #:key (actor (%current-actor))
- #:allow-other-keys)
- (send-message send-options actor
- (message-from original-message) '*reply*
- original-message #f message-body-args))
- (when (message-needs-reply? original-message)
- (apply really-send send-options)))
-
-(define (<-auto-reply actor original-message)
- "Auto-reply to a message. Internal use only!"
- (send-message '() actor (message-from original-message) '*auto-reply*
- original-message #f '()))
-
-(define (<-reply-wait original-message . message-body-args)
- "Reply to a messsage, but wait until we get a response"
- (if (message-needs-reply? original-message)
- (wait-maybe-handle-errors
- (send-message '() (%current-actor)
- (message-from original-message) '*reply*
- original-message #t message-body-args))
- #f))
-
-(define (<-reply-wait* send-options original-message
- . message-body-args)
- "Like <-reply-wait, but allows extra parameters via send-options"
- (define* (really-send #:key (actor (%current-actor))
- #:allow-other-keys)
- (apply wait-maybe-handle-errors
- (send-message send-options actor
- (message-from original-message) '*reply*
- original-message #t message-body-args)
- send-options))
- (when (message-needs-reply? original-message)
- (apply really-send send-options)))
-
-(define* (wait-maybe-handle-errors message
- #:key accept-errors
- #:allow-other-keys)
- "Before returning a message to a waiting caller, see if we need to
-raise an exception if an error."
- (define action (message-action message))
- (cond ((and (eq? action '*error*)
- (not accept-errors))
- (throw 'hive-unresumable-coroutine
- "Won't resume coroutine; got an *error* as a reply"
- #:message message))
- (else message)))
-
+(define-inlinable (%<- wants-reply from-actor to action args message-id in-reply-to)
+ ;; Okay, we need to deal with message ids.
+ ;; Could we get rid of them? :\
+ ;; It seems if we can use eq? and have messages be immutable then
+ ;; it should be possible to identify follow-up replies.
+ ;; If we need to track replies across hive boundaries we could
+ ;; register unique ids across the ambassador barrier.
+ (match to
+ (#(_ _ (? channel? channel) dead?)
+ (let ((message (make-message message-id to
+ (and from-actor (actor-id from-actor))
+ action args
+ #:wants-reply wants-reply
+ #:in-reply-to in-reply-to)))
+ (perform-operation
+ (choice-operation
+ (put-operation channel message)
+ (wait-operation dead?)))))
+ ;; TODO: put remote addresses here.
+ (#(actor-id hive-id #f #f)
+ ;; Here we'd make a call to our hive...
+ 'TODO)
+ ;; A message sent to nobody goes nowhere.
+ ;; TODO: Should we display a warning here, probably?
+ (#f #f)))
+
+(define (<- to action . args)
+ (define from-actor (*current-actor*))
+ (%<- #f from-actor to action args
+ (or (and from-actor
+ ((actor-msg-id-generator from-actor)))
+ (big-random-number-string))
+ #f))
+
+;; TODO: this should abort to the prompt, then check for errors
+;; when resuming.
+
+(define (<-wait to action . args)
+ (define prompt (*actor-prompt*))
+ (when (not prompt)
+ (error "Tried to <-wait without being in an actor's context..."))
+
+ (let ((reply (abort-to-prompt prompt to action args)))
+ (cond ((eq? action '*error*)
+ (throw 'hive-unresumable-coroutine
+ "Won't resume coroutine; got an *error* as a reply"
+ #:message reply))
+ (else (apply values (message-body reply))))))
\f
;;; Main actor implementation
(wrap-apply method)) ...)))
(define-class <actor> ()
- ;; An address object
- (id #:init-keyword #:id
+ ;; An address object... a vector of #(actor-id hive-id inbox-channel dead?)
+ ;; - inbox-channel is the receiving channel (as opposed to actor-inbox-deq)
+ ;; - dead? is a fibers condition variable which is set once this actor
+ ;; kicks the bucket
+ (id #:init-keyword #:address
#:getter actor-id)
- ;; The hive we're connected to.
- ;; We need this to be able to send messages.
- (hive #:init-keyword #:hive
- #:accessor actor-hive)
+ ;; The connection to the hive we're connected to.
+ (hive-channel #:init-keyword #:hive-channel
+ #:accessor actor-hive-channel)
+
+ ;; Our queue to send/receive messages on
+ (inbox-deq #:init-thunk make-channel
+ #:accessor actor-inbox-deq)
+
+ (msg-id-generator #:init-thunk simple-message-id-generator
+ #:getter actor-msg-id-generator)
+
;; How we receive and process new messages
(message-handler #:init-value actor-inheritable-message-handler
;; @@: There's no reason not to use #:class instead of
;; - 'wait, as in wait on the init message
;; - #f as in don't bother to init
(should-init #:init-value #t
+ #:getter actor-should-init
#:allocation #:each-subclass)
;; This is the default, "simple" way to inherit and process messages.
- (actions #:init-thunk (build-actions
- ;; Default init method is to do nothing.
- (*init* (const #f))
- ;; Default cleanup method is to do nothing.
- (*cleanup* (const #f)))
+ (actions #:init-thunk (build-actions)
#:allocation #:each-subclass))
+;;; Actors may specify an "init" action that occurs before the actor
+;;; actually begins to run.
+;;; During actor-init!, an actor may send a message to itself or others
+;;; via <- but *may not* use <-wait.
+(define-method (actor-init! (actor <actor>))
+ 'no-op)
+
+(define-method (actor-cleanup! (actor <actor>))
+ 'no-op)
+
;;; Addresses are vectors where the first part is the actor-id and
;;; the second part is the hive-id. This works well enough... they
;;; look decent being pretty-printed.
-(define (make-address actor-id hive-id)
- (vector actor-id hive-id))
+(define (make-address actor-id hive-id channel dead?)
+ (vector actor-id hive-id channel dead?))
(define (address-actor-id address)
(vector-ref address 0))
(define (address-hive-id address)
(vector-ref address 1))
+(define (address-channel address)
+ (vector-ref address 2))
+
+(define (address-dead? address)
+ (vector-ref address 3))
+
(define (address->string address)
(string-append (address-actor-id address) "@"
(address-hive-id address)))
-(define-method (actor-id-actor (actor <actor>))
+(define (address-equal? address1 address2)
+ "Check whether or not the two addresses are equal.
+
+This compares the actor-id and hive-id but ignores the channel and
+dead? condition."
+ (match address1
+ (#(actor-id-1 hive-id-1 _ _)
+ (match address2
+ (#(actor-id-2 hive-id-2)
+ (and (equal? actor-id-1 actor-id-2)
+ (and (equal? hive-id-1 hive-id-2))))
+ (_ #f)))
+ (_ #f)))
+
+(define (actor-id-actor actor)
"Get the actor id component of the actor-id"
(address-actor-id (actor-id actor)))
-(define-method (actor-id-hive (actor <actor>))
+(define (actor-id-hive actor)
"Get the hive id component of the actor-id"
(address-hive-id (actor-id actor)))
-(define-method (actor-id-string (actor <actor>))
+(define (actor-id-string actor)
"Render the full actor id as a human-readable string"
(address->string (actor-id actor)))
-(define %current-actor
+(define (actor-inbox-enq actor)
+ (address-channel (actor-id actor)))
+
+(define *current-actor*
(make-parameter #f))
-(define (actor-alive? actor)
- (hive-resolve-local-actor (actor-hive actor) (actor-id actor)))
+(define *actor-prompt*
+ (make-parameter #f))
+(define (actor-main-loop actor)
+ "Main loop of the actor. Loops around, pulling messages off its queue
+and handling them."
+ ;; @@: Maybe establish some sort of garbage collection routine for these...
+ (define waiting
+ (make-hash-table))
+ (define message-handler
+ (actor-message-handler actor))
+ (define dead?
+ (address-dead? (actor-id actor)))
+ (define prompt (make-prompt-tag (actor-id-actor actor)))
+
+ (define (handle-message message)
+ (catch #t
+ (lambda ()
+ (call-with-values
+ (lambda ()
+ (message-handler actor message))
+ (lambda vals
+ ;; Return reply if necessary
+ (when (message-wants-reply message)
+ (when (message-wants-reply message)
+ (%<- #f actor (message-from message) '*reply*
+ vals ((actor-msg-id-generator actor))
+ (message-id message)))))))
+ (const #t)
+ (let ((err (current-error-port)))
+ (lambda (key . args)
+ (false-if-exception
+ (let ((stack (make-stack #t 4)))
+ (format err "Uncaught exception when handling message ~a:\n"
+ message)
+ (display-backtrace stack err)
+ (print-exception err (stack-ref stack 0)
+ key args)
+ (newline err)
+ ;; If the other actor is waiting on a reply, let's let them
+ ;; know there was an error...
+ (when (message-wants-reply message)
+ (%<- #f actor (message-from message) '*error*
+ (list key) ((actor-msg-id-generator actor))
+ (message-id message)))))))))
+
+ (define (resume-handler message)
+ (define in-reply-to (message-in-reply-to message))
+ (cond
+ ((hash-ref waiting in-reply-to) =>
+ (lambda (kont)
+ (hash-remove! waiting in-reply-to)
+ (kont message)))
+ (else
+ (format (current-error-port)
+ "Tried to resume nonexistant message: ~a\n"
+ (message-id message)))))
+
+ (define halt-or-handle-message
+ ;; It would be nice if we could give priorities to certain operations.
+ ;; halt should always win over getting a message...
+ (choice-operation
+ (wrap-operation (wait-operation dead?)
+ (const #f)) ; halt and return
+ (wrap-operation (get-operation (actor-inbox-deq actor))
+ (lambda (message)
+ (call-with-prompt prompt
+ (lambda ()
+ (if (message-in-reply-to message)
+ ;; resume a continuation which was waiting on a reply
+ (resume-handler message)
+ ;; start handling a new message
+ (handle-message message)))
+ ;; Here's where we abort to if we're doing <-wait
+ ;; @@: maybe use match-lambda if we're going to end up
+ ;; handling multiple ~commands
+ (lambda (kont to action message-args)
+ (define message-id
+ ((actor-msg-id-generator actor)))
+ (hash-set! waiting message-id kont)
+ (%<- #t actor to action message-args message-id #f)))
+ #t)))) ; loop again
+
+ ;; Mutate the parameter; this should be fine since each fiber
+ ;; runs in its own dynamic state with with-dynamic-state.
+ ;; See with-dynamic-state discussion in
+ ;; https://wingolog.org/archives/2017/06/27/growing-fibers
+ (*current-actor* actor)
+ ;; We temporarily set the *actor-prompt* to #f to make sure that
+ ;; actor-init! doesn't try to do a <-wait message (and not accidentally use
+ ;; the parent fiber's *actor-prompt* either.)
+ (*actor-prompt* #f)
+ (actor-init! actor)
+ (*actor-prompt* prompt)
+
+ (let loop ()
+ (and (perform-operation halt-or-handle-message)
+ (loop))))
\f
;;; Actor utilities
\f
;;; The Hive
;;; ========
-;;; Every actor has a hive. The hive is a kind of "meta-actor"
-;;; which routes all the rest of the actors in a system.
-
-(define-generic hive-handle-failed-forward)
+;;; Every actor has a hive, which keeps track of other actors, manages
+;;; cleanup, and performs inter-hive communication.
-(define-class <hive> (<actor>)
+(define-class <hive> ()
+ (id #:init-keyword #:id
+ #:getter hive-id)
(actor-registry #:init-thunk make-hash-table
#:getter hive-actor-registry)
- (msg-id-generator #:init-thunk simple-message-id-generator
- #:getter hive-msg-id-generator)
+ ;; TODO: Rename "ambassadors" to "relays"
;; Ambassadors are used (or will be) for inter-hive communication.
- ;; These are special actors that know how to route messages to other hives.
+ ;; These are special actors that know how to route messages to other
+ ;; hives.
(ambassadors #:init-thunk make-weak-key-hash-table
#:getter hive-ambassadors)
- ;; Waiting coroutines
- ;; This is a map from cons cell of message-id
- ;; to a cons cell of (actor-id . coroutine)
- ;; @@: Should we have a <waiting-coroutine> record type?
- ;; @@: Should there be any way to clear out "old" coroutines?
- (waiting-coroutines #:init-thunk make-hash-table
- #:getter hive-waiting-coroutines)
- ;; Message prompt
- ;; When actors send messages to each other they abort to this prompt
- ;; to send the message, then carry on their way
- (prompt #:init-thunk make-prompt-tag
- #:getter hive-prompt)
- (actions #:allocation #:each-subclass
- #:init-thunk
- (build-actions
- ;; This is in the case of an ambassador failing to forward a
- ;; message... it reports it back to the hive
- (*failed-forward* hive-handle-failed-forward)
- ;; These are called at start and end of run-hive
- (*init-all* hive-handle-init-all)
- (*cleanup-all* hive-handle-cleanup-all))))
-
-(define-method (hive-handle-init-all (hive <hive>) message)
- "Run *init* method on all actors in registry"
- ;; We have to do this hack and run over the list
- ;; twice, because hash-for-each would result in an unrewindable
- ;; continuation, and to avoid the hash-map changing during the
- ;; middle of this.
- (define actor-ids
- (hash-map->list (lambda (actor-id actor) actor-id)
- (hive-actor-registry hive)))
- (for-each (lambda (actor-id)
- (let* ((actor (hash-ref (hive-actor-registry hive)
- actor-id)))
- (match (slot-ref actor 'should-init)
- (#f #f)
- ('wait
- (<-wait actor-id '*init*))
- (_
- (<- actor-id '*init*)))))
- actor-ids))
-
-(define-method (hive-handle-failed-forward (hive <hive>) message)
- "Handle an ambassador failing to forward a message"
- 'TODO)
-
-(define-method (hive-handle-cleanup-all (hive <hive>) message)
- "Send a message to all actors in our registry to clean themselves up."
- ;; We have to do this hack and run over the list
- ;; twice, because hash-for-each would result in an unrewindable
- ;; continuation, and to avoid the hash-map changing during the
- ;; middle of this.
- (define actor-ids
- (hash-map->list (lambda (actor-id actor) actor-id)
- (hive-actor-registry hive)))
- (for-each (lambda (actor-id)
- (<- actor-id '*cleanup*))
- actor-ids))
+ (channel #:init-thunk make-channel
+ #:getter hive-channel)
+ (halt? #:init-thunk make-condition
+ #:getter hive-halt?))
(define* (make-hive #:key hive-id)
- (let ((hive (make <hive>
- #:id (make-address
- "hive" (or hive-id
- (big-random-number-string))))))
- ;; Set the hive's actor reference to itself
- (set! (actor-hive hive) hive)
- ;; Register the actor with itself
- (hive-register-actor! hive hive)
- hive))
-
-(define-method (hive-id (hive <hive>))
- (actor-id-hive hive))
-
-(define-method (hive-gen-actor-id (hive <hive>) cookie)
- (make-address (if cookie
- (string-append cookie ":" (big-random-number-string))
- (big-random-number-string))
- (hive-id hive)))
-
-(define-method (hive-gen-message-id (hive <hive>))
- "Generate a message id using HIVE's message id generator"
- ((hive-msg-id-generator hive)))
-
-(define-method (hive-resolve-local-actor (hive <hive>) actor-address)
- (hash-ref (hive-actor-registry hive) actor-address))
-
-(define-method (hive-resolve-ambassador (hive <hive>) ambassador-address)
- (hash-ref (hive-ambassadors hive) ambassador-address))
-
-(define-method (make-forward-request (hive <hive>) (ambassador <actor>) message)
- (make-message (hive-gen-message-id hive) (actor-id ambassador)
- ;; If we make the hive not an actor, we could either switch this
- ;; to #f or to the original actor...?
- ;; Maybe some more thinking should be done on what should
- ;; happen in case of failure to forward? Handling ambassador failures
- ;; seems like the primary motivation for the hive remaining an actor.
- (actor-id hive)
- '*forward*
- `((original . ,message))))
-
-(define-method (hive-reply-with-error (hive <hive>) original-message
- error-key error-args)
- ;; We only supply the error-args if the original sender is on the same hive
- (define (orig-actor-on-same-hive?)
- (equal? (hive-id hive)
- (address-hive-id (message-from original-message))))
- (set-message-replied! original-message #t)
- (let* ((new-message-body
- (if (orig-actor-on-same-hive?)
- `(#:original-message ,original-message
- #:error-key ,error-key
- #:error-args ,error-args)
- `(#:original-message ,original-message
- #:error-key ,error-key)))
- (new-message (make-message (hive-gen-message-id hive)
- (message-from original-message)
- (actor-id hive) '*error*
- new-message-body
- #:in-reply-to (message-id original-message))))
- ;; We only return a thunk, rather than run 8sync here, because if
- ;; we ran 8sync in the middle of a catch we'd end up with an
- ;; unresumable continuation.
- (lambda () (hive-process-message hive new-message))))
-
-(define-record-type <waiting-on-reply>
- (make-waiting-on-reply actor-id kont send-options)
- waiting-on-reply?
- (actor-id waiting-on-reply-actor-id)
- (kont waiting-on-reply-kont)
- (send-options waiting-on-reply-send-options))
-
-
-(define-method (hive-process-message (hive <hive>) message)
- "Handle one message, or forward it via an ambassador"
- (define (maybe-autoreply actor)
- ;; Possibly autoreply
- (if (message-needs-reply? message)
- (<-auto-reply actor message)))
-
- (define (resolve-actor-to)
- "Get the actor the message was aimed at"
- (let ((actor (hive-resolve-local-actor hive (message-to message))))
- (if (not actor)
- (throw 'actor-not-found
- (format #f "Message ~a from ~a directed to nonexistant actor ~a"
- (message-id message)
- (address->string (message-from message))
- (address->string (message-to message)))
- message))
- actor))
-
- ;; TODO: I'm pretty sure we're building up another stack of prompts here
- ;; with the way we're doing this. That's a real problem.
- (define* (call-catching-coroutine thunk #:optional (catch-errors #t))
- (define queued-error-handling-thunk #f)
- (define (call-catching-errors)
- ;; TODO: maybe parameterize (or attach to hive) and use
- ;; maybe-catch-all from agenda.scm
- ;; @@: Why not just use with-throw-handler and let the catch
- ;; happen at the agenda? That's what we used to do, but
- ;; it ended up with a SIGABRT. See:
- ;; http://lists.gnu.org/archive/html/bug-guile/2016-05/msg00003.html
- (catch #t
- thunk
- ;; In the actor model, we don't totally crash on errors.
- (lambda _ #f)
- ;; If an error happens, we raise it
- (lambda (key . args)
- (if (message-needs-reply? message)
- ;; If the message is waiting on a reply, let them know
- ;; something went wrong.
- ;; However, we have to do it outside of this catch
- ;; routine, or we'll end up in an unrewindable continuation
- ;; situation.
- (set! queued-error-handling-thunk
- (hive-reply-with-error hive message key args)))
- ;; print error message
- (apply print-error-and-continue key args)))
- ;; @@: This is a kludge. See above for why.
- (if queued-error-handling-thunk
- (8sync (queued-error-handling-thunk))))
- (call-with-prompt (hive-prompt hive)
- (if catch-errors
- call-catching-errors
- thunk)
- (lambda (kont actor message send-options)
- ;; Register the coroutine
- (hash-set! (hive-waiting-coroutines hive)
- (message-id message)
- (make-waiting-on-reply
- (actor-id actor) kont send-options))
- ;; Send off the message
- (8sync (hive-process-message hive message)))))
-
- (define (process-local-message)
- (let ((actor (resolve-actor-to)))
- (call-catching-coroutine
- (lambda ()
- (define message-handler (actor-message-handler actor))
- ;; @@: Should a more general error handling happen here?
- (parameterize ((%current-actor actor))
- (let ((result
- (message-handler actor message)))
- (maybe-autoreply actor)
- ;; Returning result allows actors to possibly make a run-request
- ;; at the end of handling a message.
- ;; ... We do want that, right?
- result))))))
-
- (define (resume-waiting-coroutine)
- (case (message-action message)
- ;; standard reply / auto-reply
- ((*reply* *auto-reply* *error*)
- (call-catching-coroutine
- (lambda ()
- (match (hash-remove! (hive-waiting-coroutines hive)
- (message-in-reply-to message))
- ((_ . waiting)
- (if (not (equal? (message-to message)
- (waiting-on-reply-actor-id waiting)))
- (throw 'resuming-to-wrong-actor
- "Attempted to resume a coroutine to the wrong actor!"
- #:expected-actor-id (message-to message)
- #:got-actor-id (waiting-on-reply-actor-id waiting)
- #:message message))
- (let* (;; @@: How should we resolve resuming coroutines to actors who are
- ;; now gone?
- (actor (resolve-actor-to))
- (kont (waiting-on-reply-kont waiting))
- (result (kont message)))
- (maybe-autoreply actor)
- result))
- (#f (throw 'no-waiting-coroutine
- "message in-reply-to tries to resume nonexistent coroutine"
- message))))
- ;; no need to catch errors here, there's already an error handler
- #f))
- ;; Unhandled action for a reply!
- (else
- (throw 'hive-unresumable-coroutine
- "Won't resume coroutine, nonsense action on reply message"
- #:action (message-action message)
- #:message message))))
-
- (define (process-remote-message)
- ;; Find the ambassador
- (let* ((remote-hive-id (hive-id (message-to message)))
- (ambassador (hive-resolve-ambassador remote-hive-id))
- (message-handler (actor-message-handler ambassador))
- (forward-request (make-forward-request hive ambassador message)))
- (message-handler ambassador forward-request)))
-
- (let ((to (message-to message)))
- ;; This seems to be an easy mistake to make, so check that addressing
- ;; is correct here
- (if (not to)
- (throw 'missing-addressee
- "`to' field is missing on message"
- #:message message))
- (if (hive-actor-local? hive to)
- (if (message-in-reply-to message)
- (resume-waiting-coroutine)
- (process-local-message))
- (process-remote-message))))
-
-(define-method (hive-actor-local? (hive <hive>) address)
- (equal? (hive-id hive) (address-hive-id address)))
-
-(define-method (hive-register-actor! (hive <hive>) (actor <actor>))
- (hash-set! (hive-actor-registry hive) (actor-id actor) actor))
-
-(define-method (%hive-create-actor (hive <hive>) actor-class
- init-args id-cookie send-init?)
- "Actual method called by bootstrap-actor / create-actor.
-
-Since this is a define-method it can't accept fancy define* arguments,
-so this gets called from the nicer bootstrap-actor interface. See
-that method for documentation."
- (let* ((actor-id (hive-gen-actor-id hive id-cookie))
+ (make <hive> #:id (or hive-id
+ (big-random-number-string))))
+
+(define (gen-actor-id cookie)
+ (if cookie
+ (string-append cookie ":" (big-random-number-string))
+ (big-random-number-string)))
+
+(define (hive-main-loop hive)
+ "The main loop of the hive. This listens for messages on the hive-channel
+for certain actions to perform.
+
+`messages' here is not the same as a <message> object; these are a list of
+values, the first value being a symbol"
+ (define channel (hive-channel hive))
+ (define halt? (hive-halt? hive))
+ (define registry (hive-actor-registry hive))
+
+ ;; not the same as a <message> ;P
+ (define handle-message
+ (match-lambda
+ (('register-actor actor-id address actor)
+ (hash-set! registry actor-id (vector address actor)))
+ ;; Remove the actor from hive
+ (('remove-actor actor-id)
+ (hash-remove! (hive-actor-registry hive) actor-id))
+ (('register-ambassador hive-id ambassador-actor-id)
+ 'TODO)
+ (('unregister-ambassador hive-id ambassador-actor-id)
+ 'TODO)
+ (('forward-message from-actor-id message)
+ 'TODO)))
+
+ (define halt-or-handle
+ (choice-operation
+ (wrap-operation (get-operation channel)
+ (lambda (msg)
+ (handle-message msg)
+ #t))
+ (wrap-operation (wait-operation halt?)
+ (const #f))))
+
+ (let lp ()
+ (and (perform-operation halt-or-handle)
+ (lp))))
+
+(define *current-hive* (make-parameter #f))
+
+(define* (spawn-hive proc #:key (hive (make-hive)))
+ "Spawn a hive in a fiber running PROC, passing it the fresh hive"
+ (spawn-fiber (lambda () (hive-main-loop hive)))
+ (proc hive))
+
+(define (run-hive proc . args)
+ "Spawn a hive and run it in run-fibers. Takes a PROC as would be passed
+to spawn-hive... all remaining arguments passed to run-fibers."
+ (apply run-fibers
+ (lambda ()
+ (spawn-hive proc))
+ args))
+
+(define (%create-actor hive-channel hive-id
+ actor-class init-args id-cookie send-init?)
+ (let* ((actor-id (gen-actor-id id-cookie))
+ (dead? (make-condition))
+ (inbox-enq (make-channel))
+ (address (make-address actor-id hive-id
+ inbox-enq dead?))
(actor (apply make actor-class
- #:hive hive
- #:id actor-id
+ #:hive-channel hive-channel
+ #:address address
init-args))
- (actor-should-init (slot-ref actor 'should-init)))
- (hive-register-actor! hive actor)
- ;; Maybe run actor init method
- (when (and send-init? actor-should-init)
- (let ((send-method
- (if (eq? actor-should-init 'wait)
- <-wait <-)))
- (send-method actor-id '*init*)))
- ;; return the actor id
- actor-id))
+ (should-init (actor-should-init actor)))
+
+ ;; start the main loop
+ (spawn-fiber (lambda ()
+ ;; start the inbox loop
+ (spawn-fiber
+ (lambda ()
+ (delivery-agent inbox-enq (actor-inbox-deq actor)
+ dead?))
+ ;; this one is decidedly non-parallel, because we want
+ ;; the delivery agent to be in the same thread as its actor
+ #:parallel? #f)
+
+ (actor-main-loop actor))
+ #:parallel? #t)
+
+ (put-message hive-channel (list 'register-actor actor-id address actor))
+
+ ;; return the address
+ address))
(define* (bootstrap-actor hive actor-class #:rest init-args)
"Create an actor on HIVE using ACTOR-CLASS passing in INIT-ARGS args"
- (%hive-create-actor hive actor-class
- init-args (symbol->string (class-name actor-class))
- #f))
+ (%create-actor (hive-channel hive) (hive-id hive) actor-class
+ init-args (symbol->string (class-name actor-class))
+ #f))
(define* (bootstrap-actor* hive actor-class id-cookie #:rest init-args)
"Create an actor, but also allow customizing a 'cookie' added to the id
for debugging"
- (%hive-create-actor hive actor-class
- init-args id-cookie
- #f))
-
-(define (call-with-message message proc)
- "Applies message body arguments into procedure, with message as first
-argument. Similar to call-with-values in concept."
- (apply proc message (message-body message)))
-
-;; (mbody-receive (<- bar baz)
-;; (baz)
-;; basil)
-
-;; Emacs: (put 'mbody-receive 'scheme-indent-function 2)
-
-;; @@: Or receive-msg or receieve-message or??
-(define-syntax-rule (mbody-receive arglist message body ...)
- "Call body with arglist (which can accept arguments like lambda*)
-applied from the message-body of message."
- (call-with-message message
- (lambda* arglist
- body ...)))
-
-(define (mbody-val message)
- "Retrieve the first value from the message-body of message.
-Like single value return from a procedure call. Probably the most
-common case when waiting on a reply from some action invocation."
- (call-with-message message
- (lambda (_ val) val)))
-
-\f
-;;; Various API methods for actors to interact with the system
-;;; ==========================================================
-
-;; TODO: move send-message and friends here...?
+ (%create-actor (hive-channel hive) (hive-id hive) actor-class
+ init-args id-cookie
+ #f))
(define* (create-actor from-actor actor-class #:rest init-args)
"Create an instance of actor-class. Return the new actor's id.
This is the method actors should call directly (unless they want
to supply an id-cookie, in which case they should use
create-actor*)."
- (%hive-create-actor (actor-hive from-actor) actor-class
- init-args #f #t))
+ (%create-actor (actor-hive-channel from-actor) (actor-id-hive from-actor)
+ actor-class init-args #f #t))
(define* (create-actor* from-actor actor-class id-cookie #:rest init-args)
"Create an instance of actor-class. Return the new actor's id.
Like create-actor, but permits supplying an id-cookie."
- (%hive-create-actor (actor-hive from-actor) actor-class
- init-args id-cookie #t))
-
+ (%create-actor (actor-hive-channel from-actor) (actor-id-hive from-actor)
+ actor-class init-args id-cookie #t))
(define* (self-destruct actor #:key (cleanup #t))
"Remove an actor from the hive.
Unless #:cleanup is set to #f, this will first have the actor handle
its '*cleanup* action handler."
- (when cleanup
- (<-wait (actor-id actor) '*cleanup*))
- (hash-remove! (hive-actor-registry (actor-hive actor))
- (actor-id actor)))
-
-
-\f
-;;; 8sync bootstrap utilities
-;;; =========================
-
-(define* (run-hive hive initial-tasks
- #:key (cleanup #t)
- (handle-signals (list SIGINT SIGTERM)))
- "Start up an agenda and run HIVE in it with INITIAL-TASKS.
-
-Keyword arguments:
- - #:cleanup: Whether to run *cleanup* on all actors.
- - #:handle-sigactions: a list of signals to set up interrupt
- handlers for, so cleanup sill still happen as expected.
- Defaults to a list of SIGINT and SIGTERM."
- (dynamic-wind
- (const #f)
- (lambda ()
- (define (run-it escape)
- (define (handle-signal signum)
- (restore-signals)
- (escape signum))
- (for-each (lambda (signum)
- (sigaction signum handle-signal))
- handle-signals)
- (let* ((queue (list->q
- (cons (bootstrap-message hive (actor-id hive) '*init-all*)
- initial-tasks)))
- (agenda (make-agenda #:pre-unwind-handler print-error-and-continue
- #:queue queue)))
- (run-agenda agenda)))
- (call/ec run-it))
- ;; Run cleanup
- (lambda ()
- (when cleanup
- (run-hive-cleanup hive)))))
-
-(define (run-hive-cleanup hive)
- (let ((queue (list->q (list (bootstrap-message hive (actor-id hive)
- '*cleanup-all*)))))
- (run-agenda
- (make-agenda #:queue queue))))
+ (signal-condition! (address-dead? (actor-id actor)))
+ (put-message (actor-hive-channel actor) (list 'remove-actor (actor-id-actor actor)))
+ ;; Set *actor-prompt* to nothing to prevent actor-cleanup! from sending
+ ;; a message with <-wait
+ (*actor-prompt* #f)
+ (actor-cleanup! actor))
+
+;; From a patch I sent to Fibers...
+(define (condition-signalled? cvar)
+ "Return @code{#t} if @var{cvar} has already been signalled.
+
+In general you will want to use @code{wait} or @code{wait-operation} to
+wait on a condition. However, sometimes it is useful to see whether or
+not a condition has already been signalled without blocking."
+ (atomic-box-ref ((@@ (fibers conditions) condition-signalled?) cvar)))
-(define (bootstrap-message hive to-id action . message-body-args)
- (wrap
- (apply <-* `(#:actor ,hive) to-id action message-body-args)))
-
-
-\f
-;;; Basic readers / writers
-;;; =======================
-
-(define (serialize-message message)
- "Serialize a message for read/write"
- (list
- (message-id message)
- (message-to message)
- (message-from message)
- (message-action message)
- (message-body message)
- (message-in-reply-to message)
- (message-wants-reply message)
- (message-replied message)))
-
-(define* (write-message message #:optional (port (current-output-port)))
- "Write out a message to a port for easy reading later.
-
-Note that if a sub-value can't be easily written to something
-Guile's `read' procedure knows how to read, this doesn't do anything
-to improve that. You'll need a better serializer for that.."
- (write (serialize-message message) port))
-
-(define (serialize-message-pretty message)
- "Serialize a message in a way that's easy for humans to read."
- `(*message*
- (id ,(message-id message))
- (to ,(message-to message))
- (from ,(message-from message))
- (action ,(message-action message))
- (body ,(message-body message))
- (in-reply-to ,(message-in-reply-to message))
- (wants-reply ,(message-wants-reply message))
- (replied ,(message-replied message))))
-
-(define (pprint-message message)
- "Pretty print a message."
- (pretty-print (serialize-message-pretty message)))
-
-(define* (read-message #:optional (port (current-input-port)))
- "Read a message serialized via serialize-message from PORT"
- (match (read port)
- ((id to from action body in-reply-to wants-reply replied)
- (make-message-intern
- id to from action body
- in-reply-to wants-reply replied))
- (anything-else
- (throw 'message-read-bad-structure
- "Could not read message from structure"
- anything-else))))
-
-(define (read-message-from-string message-str)
- "Read message from MESSAGE-STR"
- (with-input-from-string message-str
- (lambda ()
- (read-message (current-input-port)))))
+(define (actor-alive? actor)
+ (condition-signalled? (address-dead? (actor-id actor))))