examples/flink: implement task delegation roles in terms of abstract
templates
This commit is contained in:
Sam Caldwell
parent
e16db164df
commit
7815fad415
|
|
@ -5,7 +5,8 @@
|
|||
set-count
|
||||
set-empty?
|
||||
set-first
|
||||
set-remove))
|
||||
set-remove
|
||||
set-add))
|
||||
(require (only-in racket/list
|
||||
partition
|
||||
empty?
|
||||
|
|
@ -38,14 +39,14 @@ of each TaskManager (TM) is contingent on the presence of a job manager.
|
|||
#|
|
||||
In turn, TaskManagers advertise their presence with (task-manager ID slots),
|
||||
where ID is a unique id, and slots is a natural number. The number of slots
|
||||
dictates how many tasks the TM can take on. To reduce contention, we the JM
|
||||
dictates how many tasks the TM can take on. To reduce contention, the JM
|
||||
should only assign a task to a TM if the TM actually has the resources to
|
||||
perform a task.
|
||||
|#
|
||||
(assertion-struct task-manager (id slots))
|
||||
;; an ID is a symbol or a natural number.
|
||||
;; Any -> Bool
|
||||
;; recognize ids
|
||||
;; recognize IDs
|
||||
(define (id? x)
|
||||
(or (symbol? x) (exact-nonnegative-integer? x)))
|
||||
#|
|
||||
|
|
@ -127,6 +128,50 @@ The JobManager then performs the job and, when finished, asserts (job-finished I
|
|||
(define (log fmt . args)
|
||||
(displayln (apply format fmt args)))
|
||||
|
||||
;; ---------------------------------------------------------------------------------------------------
|
||||
;; Generic Implementation of Task Delegation Protocol
|
||||
|
||||
;; a TaskFun is a
|
||||
;; (Task ID (TaskResults -> Void) ((U ACCEPTED OVERLOAD RUNNING) -> Void) -> Void)
|
||||
|
||||
;; ID (-> Bool) TaskFun -> TaskPerformer
|
||||
;; doesn't really account for long-running tasks
|
||||
;; gonna need some effect polymorphism to type uses of this
|
||||
(define (task-performer my-id can-accept? perform-task)
|
||||
(react
|
||||
(during (task-assignment my-id $job-id $task)
|
||||
(field [status #f])
|
||||
(assert (task-state my-id job-id (task-id task) (status)))
|
||||
(cond
|
||||
[(can-accept?)
|
||||
(status RUNNING)
|
||||
(define (on-complete results)
|
||||
(status (finished results)))
|
||||
(perform-task task job-id on-complete status)]
|
||||
[else
|
||||
(status OVERLOAD)]))))
|
||||
|
||||
;; Task
|
||||
;; ID
|
||||
;; ID
|
||||
;; (-> Void)
|
||||
;; (TaskResults -> Void)
|
||||
;; -> TaskAssigner
|
||||
(define (task-assigner tsk job-id performer
|
||||
on-overload!
|
||||
on-complete!)
|
||||
(react
|
||||
(assert (task-assignment performer job-id tsk))
|
||||
(on (asserted (task-state performer job-id (task-id tsk) $status))
|
||||
(match status
|
||||
[(or (== ACCEPTED)
|
||||
(== RUNNING))
|
||||
(void)]
|
||||
[(== OVERLOAD)
|
||||
(stop-current-facet (on-overload!))]
|
||||
[(finished results)
|
||||
(stop-current-facet (on-complete! results))]))))
|
||||
|
||||
;; ---------------------------------------------------------------------------------------------------
|
||||
;; TaskRunner
|
||||
|
||||
|
|
@ -138,25 +183,24 @@ The JobManager then performs the job and, when finished, asserts (job-finished I
|
|||
(assert (task-runner id (status)))
|
||||
(begin/dataflow
|
||||
(log "task-runner ~v state is: ~a" id (status)))
|
||||
(during (task-assignment id $job-id (task $tid $desc))
|
||||
(field [execution-state (if (idle?) RUNNING OVERLOAD)]
|
||||
[word-count (hash)])
|
||||
(assert (task-state id job-id tid (execution-state)))
|
||||
;; we have to avoid asking a non-idle runner to do anything
|
||||
(when (idle?)
|
||||
(on-stop (status IDLE))
|
||||
(on-start
|
||||
(status (executing tid))
|
||||
;; since we currently finish everything in one turn, allow other actors to observe the changes in our
|
||||
;; task-runner state by flushing pending actions.
|
||||
(flush!)
|
||||
(match desc
|
||||
[(map-work data)
|
||||
(word-count (count-new-words (word-count) (string->words data)))
|
||||
(execution-state (finished (word-count)))]
|
||||
[(reduce-work left right)
|
||||
(word-count (hash-union left right #:combine +))
|
||||
(execution-state (finished (word-count)))]))))))
|
||||
;; Task (TaskStateDesc -> Void) -> Void
|
||||
(define (perform-task tsk job-id on-complete! update-status!)
|
||||
(unless (idle?)
|
||||
(error "tried to perform a task when not idle"))
|
||||
;; since we currently finish everything in one turn, these changes to status aren't
|
||||
;; actually visible.
|
||||
(status RUNNING)
|
||||
(match-define (task tid desc) tsk)
|
||||
(match desc
|
||||
[(map-work data)
|
||||
(define wc (count-new-words (hash) (string->words data)))
|
||||
(on-complete! wc)]
|
||||
[(reduce-work left right)
|
||||
(define wc (hash-union left right #:combine +))
|
||||
(on-complete! wc)])
|
||||
(status IDLE))
|
||||
(on-start
|
||||
(task-performer id idle? perform-task))))
|
||||
|
||||
;; (Hash String Nat) String -> (Hash String Nat)
|
||||
(define (word-count-increment h word)
|
||||
|
|
@ -206,40 +250,25 @@ The JobManager then performs the job and, when finished, asserts (job-finished I
|
|||
#:on-remove (log "TM learns that task-runner ~a is NOT IDLE" id))
|
||||
(assert (task-manager id (set-count (idle-runners))))
|
||||
(field [busy-runners (list)])
|
||||
(during (task-assignment id $job-id $t)
|
||||
(match-define (task task-id desc) t)
|
||||
#;(on-start (log "TM receives task ~a" task-id))
|
||||
(log "TM receives task ~a" task-id)
|
||||
(on-stop (log "TM finished with task ~a" task-id))
|
||||
(field [status ACCEPTED])
|
||||
;; TODO - could delegate this assertion, in the non-overloaded case, to TaskRunner
|
||||
;; (also removing the first id from task-state)
|
||||
(assert (task-state id job-id task-id (status)))
|
||||
(cond
|
||||
[(set-empty? (idle-runners))
|
||||
(log "TM can't run ~a right now" task-id)
|
||||
(status OVERLOAD)]
|
||||
[else
|
||||
(define runner (set-first (idle-runners)))
|
||||
;; n.b. modifying a query set is questionable
|
||||
;; but if we wait for the IDLE assertion to be retracted, we might assign multiple tasks to the same runner.
|
||||
;; Could use the busy-runners field to avoid that
|
||||
(idle-runners (set-remove (idle-runners) runner))
|
||||
(log "TM assigns task ~a to runner ~a" task-id runner)
|
||||
(assert (task-assignment runner job-id t))
|
||||
(status RUNNING)
|
||||
(on (asserted (task-state runner job-id task-id $state))
|
||||
(match state
|
||||
[(or (== ACCEPTED)
|
||||
(== RUNNING))
|
||||
;; nothing to do
|
||||
(void)]
|
||||
[(== OVERLOAD)
|
||||
(log "TM overloaded TR with task ~a" task-id)
|
||||
(status OVERLOAD)]
|
||||
[(finished results)
|
||||
(log "TM receives the results of task ~a" task-id)
|
||||
(status state)]))])))))
|
||||
(define (can-accept?)
|
||||
(not (set-empty? (idle-runners))))
|
||||
(define (perform-task tsk job-id on-complete! update-status!)
|
||||
(match-define (task task-id desc) tsk)
|
||||
(define runner (set-first (idle-runners)))
|
||||
;; n.b. modifying a query set is questionable
|
||||
;; but if we wait for the IDLE assertion to be retracted, we might assign multiple tasks to the same runner.
|
||||
;; Could use the busy-runners field to avoid that
|
||||
(idle-runners (set-remove (idle-runners) runner))
|
||||
(log "TM assigns task ~a to runner ~a" task-id runner)
|
||||
;; TODO - since we're both adding and removing from this set I'm not sure TRs
|
||||
;; need to be making assertions about their idleness
|
||||
(on-stop (idle-runners (set-add (idle-runners) runner)))
|
||||
(on-start
|
||||
(task-assigner tsk job-id runner
|
||||
(lambda () (update-status! OVERLOAD))
|
||||
(lambda (results) (on-complete! results)))))
|
||||
(on-start
|
||||
(task-performer id can-accept? perform-task)))))
|
||||
|
||||
;; ---------------------------------------------------------------------------------------------------
|
||||
;; JobManager
|
||||
|
|
@ -297,43 +326,49 @@ The JobManager then performs the job and, when finished, asserts (job-finished I
|
|||
;; Requires (task-ready? t)
|
||||
(define (perform-task t k)
|
||||
(react
|
||||
(define task-facet (current-facet-id))
|
||||
(on-start (tasks-in-progress (add1 (tasks-in-progress))))
|
||||
(on-stop (tasks-in-progress (sub1 (tasks-in-progress))))
|
||||
(match-define (task this-id desc) t)
|
||||
(log "JM begins on task ~a" this-id)
|
||||
|
||||
(field [task-mngr #f])
|
||||
(begin/dataflow
|
||||
;; n.b. cyclic data-flow dependency
|
||||
(unless (task-mngr)
|
||||
(define mngr
|
||||
(for/first ([(id slots) (in-hash (task-managers))]
|
||||
#:when (positive? (- slots (hash-ref (requests-in-flight) id 0))))
|
||||
id))
|
||||
(when mngr
|
||||
(take-slot! mngr)
|
||||
(react (stop-when (asserted (task-state mngr job-id this-id _))
|
||||
(received-answer! mngr)))
|
||||
(task-mngr mngr))))
|
||||
;; TODO - should respond if task manager dies
|
||||
(assert #:when (task-mngr)
|
||||
(task-assignment (task-mngr) job-id t))
|
||||
(on #:when (task-mngr)
|
||||
(asserted (task-state (task-mngr) job-id this-id $state))
|
||||
(match state
|
||||
[(or (== ACCEPTED)
|
||||
(== RUNNING))
|
||||
;; nothing to do
|
||||
(void)]
|
||||
[(== OVERLOAD)
|
||||
;; need to find a new task manager
|
||||
;; don't think we need a release-slot! here, because if we've heard back from a task manager,
|
||||
;; they should have told us a different slot count since we tried to give them work
|
||||
(log "JM overloaded manager ~a with task ~a" (task-mngr) this-id)
|
||||
(task-mngr #f)]
|
||||
[(finished results)
|
||||
(log "JM receives the results of task ~a" this-id)
|
||||
(stop-current-facet (k this-id results))]))))
|
||||
(define (select-a-task-manager)
|
||||
(react
|
||||
(begin/dataflow
|
||||
(define mngr
|
||||
(for/first ([(id slots) (in-hash (task-managers))]
|
||||
#:when (positive? (- slots (hash-ref (requests-in-flight) id 0))))
|
||||
id))
|
||||
(when mngr
|
||||
(take-slot! mngr)
|
||||
(stop-current-facet (assign-task mngr))))))
|
||||
|
||||
;; ID -> ...
|
||||
(define (assign-task mngr)
|
||||
(react
|
||||
(define this-facet (current-facet-id))
|
||||
(on (retracted (task-manager mngr _))
|
||||
;; our task manager has crashed
|
||||
(stop-current-facet (select-a-task-manager)))
|
||||
(on-start
|
||||
;; N.B. when this line was here, and not after `(when mngr ...)` above,
|
||||
;; things didn't work. I think that due to script scheduling, all ready
|
||||
;; tasks were being assigned to the manager
|
||||
#;(take-slot! mngr)
|
||||
(react (stop-when (asserted (task-state mngr job-id this-id _))
|
||||
(received-answer! mngr)))
|
||||
(task-assigner t job-id mngr
|
||||
(lambda ()
|
||||
;; need to find a new task manager
|
||||
;; don't think we need a release-slot! here, because if we've heard back from a task manager,
|
||||
;; they should have told us a different slot count since we tried to give them work
|
||||
(log "JM overloaded manager ~a with task ~a" mngr this-id)
|
||||
(stop-facet this-facet (select-a-task-manager)))
|
||||
(lambda (results)
|
||||
(log "JM receives the results of task ~a" this-id)
|
||||
(stop-facet task-facet (k this-id results)))))))
|
||||
|
||||
(on-start (select-a-task-manager))))
|
||||
|
||||
;; ID Data -> Void
|
||||
;; Update any dependent tasks with the results of the given task, moving
|
||||
|
|
|
|||
Loading…
Reference in New Issue