Import/export distributed GC now works too!

parent 55c7297c
......@@ -153,6 +153,8 @@
(struct cmd-send-gc-answer (answer-pos)
(struct cmd-send-gc-export (export-pos)
;; utility for splitting up keyword argument hashtable in a way usable by
;; keyword-apply
......@@ -261,10 +263,8 @@
(define next-question-pos 0)
;; (define next-promise-pos 0)
;; TODO: We need to wrap this in a structure that keeps track of when it
;; can GC
(define exports-val2pos (make-weak-hasheq)) ; exports[val]: chosen by us
(define exports-pos2val (make-hasheqv)) ; exports[pos]: chosen by us
(define exports-val2pos (make-hasheq)) ; exports[val]: chosen by us
(define exports-pos2val (make-hasheqv)) ; exports[pos]: chosen by us
;; TODO: This doesn't make sense if the value isn't wrapped in a weak
;; reference... I think this also needs to go in both directions to work
;; from a GC perspective
......@@ -276,19 +276,67 @@
;; can query, right?
(define running? #t)
;; These are imports that we've processed when we already had allocated
;; a reference. We batch send GC messages about these as appropriate.
;; Note that we say "spare" because there's one more count that's
;; associated with the reference itself.
;; Mapping of slot position -> count
(define spare-import-counts
;; The inverse: tracking how many export numbers we've given so we can
;; know when it hits 0 and is ok to remove
(define export-counts
(define (increment-spare-imports-count! import-pos)
(hash-set! spare-import-counts import-pos
(add1 (hash-ref spare-import-counts import-pos 0))))
;; Go through all the "spare imports" and reset them
(define (handle-spare-imports!)
(for ([(import-pos count) (in-hash spare-import-counts)])
;; Send a gc-export message for this many
(send-to-remote (op:gc-export import-pos count))
;; Reset these
(hash-remove! spare-import-counts import-pos)))
(define (decrement-exports-count-maybe-remove! export-pos delta)
(-> integer? integer? any/c)
(match (hash-ref export-counts export-pos #f)
[(and (? integer?) (? positive? cur-count))
(match (- cur-count delta)
;; time to remove
(hash-remove! export-counts export-pos)
;; Remove this export from both
(let ([val (hash-ref exports-pos2val export-pos)])
(hash-remove! exports-val2pos val)
(hash-remove! exports-pos2val export-pos))]
;; decremented but still positive
[(and (? integer?) (? positive? new-count))
(hash-set! export-counts export-pos new-count)]
(error 'exports-gc-error
"Tried decrementing export-pos ~a by ~a but that's negative: ~a"
(error 'exports-gc-error
"Tried to decrement the exports count for position ~a but its value was ~a"
export-pos other-val)]))
;; Now make the will executor and boot its corresponding thread
;; for cooperative GC.
(define refr-will-executor
;; TODO: Should we move this out from a thread and put it in the
;; main loop and run it after every loop with will-try-execute?
;; That could reduce the chance of some race conditions, though
;; I'm not sure it's strictly necessary.
(lambda ()
(let lp ()
(will-execute refr-will-executor)
#;(define (make-import-will-handler import-id)
(send-to-remote )
(define (make-question-will-handler question-pos)
(lambda _
;; There's (I think?) a possible race condition here if we were to
......@@ -299,6 +347,14 @@
(will-register refr-will-executor question-finder
(make-question-will-handler question-pos)))
(define (make-import-will-handler import-pos)
(lambda _
(hash-remove! imports import-pos)
(async-channel-put internal-ch (cmd-send-gc-export import-pos))))
(define (install-import-will-handler! question-finder import-pos)
(will-register refr-will-executor question-finder
(make-import-will-handler import-pos)))
;; Possibly install an export for this local refr, and return
;; this export id
;; TODO: we maybe need to differentiate between local-live-refr and
......@@ -308,7 +364,18 @@
;; Already have it, no need to increment next-export-pos
[(hash-has-key? exports-val2pos refr)
(hash-ref exports-val2pos refr)]
(define export-pos
(hash-ref exports-val2pos refr))
;; However, we do need to increment our export count
(match (hash-ref export-counts export-pos #f)
;; Uh, we screwed up our bookkeeping at some point
(error 'no-export-count-wtf
"No export count for ~a" export-pos)]
(hash-set! export-counts export-pos (add1 cur-count))])
;; now finally return the export position
;; Nope, let's export this
;; TODO: This doesn't do handoffs for remote refrs yet!!
......@@ -321,6 +388,13 @@
(hash-set! exports-val2pos refr
;; (sanity check:) make sure there's no export count currently
(when (hash-has-key? export-counts export-pos)
(error 'shouldnt-be-export-count-wtf
"Adding a new export but there was already an export count for pos: ~a"
;; and set the export count to 1
(hash-set! export-counts export-pos 1)
(define/contract (marshall-local-refr! local-refr)
......@@ -337,28 +411,37 @@
(define (maybe-install-import! import-desc)
(define import-pos
(desc:import-pos import-desc))
(define (install-new-import!)
;; construct the new reference...
(define new-refr
(match import-desc
[(? desc:import-object?)
(make-remote-object-refr captp-connector
(pos-seal import-pos))]
[(? desc:import-promise?)
(make-remote-promise-refr captp-connector
(pos-seal import-pos))]))
;; Install it...
(hash-set! imports import-pos (make-weak-box new-refr))
;; set up the will handler...
(install-import-will-handler! new-refr import-pos)
;; and return it.
[(hash-has-key? imports import-pos)
;; Oh, we've already got that. Reference and return it.
(hash-ref imports import-pos)]
(match (weak-box-value (hash-ref imports import-pos))
;; Possible race condition: Apparently it was GC'ed
;; mid-operation so now we need to add it back
;; @@: *sweating profusely* but is this all the possible
;; race conditions???
[#f (install-new-import!)]
;; looks like we got the refr, return as-is
(increment-spare-imports-count! import-pos)
;; construct the new reference...
(define new-refr
(match import-desc
[(? desc:import-object?)
(make-remote-object-refr captp-connector
(pos-seal import-pos))]
[(? desc:import-promise?)
(make-remote-promise-refr captp-connector
(pos-seal import-pos))]))
;; TODO: weak boxing goes here
;; install it...
(hash-set! imports import-pos new-refr)
;; and return it.
(define (remote-refr->imported-pos to)
(pos-unseal (remote-refr-sealed-pos to)))
(define (question-finder->question-pos! question-finder)
(if (hash-has-key? questions question-finder)
......@@ -551,10 +634,12 @@
(listen to-refr listener
#:wants-partial? wants-partial?)
[(op:abort reason)
[(op:gc-answer answer-pos)
(hash-remove! answers answer-pos)]
[(op:gc-export (? integer? export-pos) (? integer? wire-delta))
(decrement-exports-count-maybe-remove! export-pos wire-delta)]
[(op:abort reason)
(error 'invalid-message "~a" other-message)]))
......@@ -594,7 +679,9 @@
(send-to-remote listen-msg)]
[(cmd-send-gc-answer (? integer? answer-pos))
(send-to-remote (op:gc-answer answer-pos))]))
(send-to-remote (op:gc-answer answer-pos))]
[(cmd-send-gc-export (? integer? export-pos))
(send-to-remote (op:gc-export export-pos 1))]))
(define (handle-from-machine-representative msg)
(match msg
......@@ -633,6 +720,11 @@
(handle-evt from-machine-representative-ch
;; TODO: Maybe actually add the will executor step right here
;; It'll be a little less performant but if we end up hitting weird
;; race conditions it could eliminate them?
;; But I guess see whether or not we ever hit them.
......@@ -1051,4 +1143,36 @@
;; #:marshallers marshallers)
(define ((^args-counter bcom) . args)
(length args))
(define a-args-counter
(a-vat 'spawn ^args-counter))
(define shares-a-args-counter
(a-vat 'spawn
(lambda (bcom)
(lambda ()
(define shares-a-args-counter-nonce
(a-vat 'call a-nonce-reg 'register shares-a-args-counter))
;; vow on b specifically
(define shares-a-args-counter-vow
(b-vat 'run
(lambda ()
(<- b->a-bootstrap-vow 'fetch shares-a-args-counter-nonce))))
;; check that incremental gc is working
#;(b-vat 'run
(lambda ()
(on (<- shares-a-args-counter-vow)
(lambda (a-args-counter)
(<- a-args-counter
bob-greeter2 bob-greeter2 bob-greeter2
bob-greeter2 bob-greeter2 bob-greeter2
bob-greeter2 bob-greeter2 bob-greeter2)))))
;; ... damn we actually need to test the messages sent at some point
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment