https://github.com/JuliaLang/julia
Tip revision: e9263b7025725fb7297040318df06d5f86b94721 authored by Viral B. Shah on 14 February 2013, 00:49:00 UTC
Merge branch 'master' of github.com:JuliaLang/julia
Merge branch 'master' of github.com:JuliaLang/julia
Tip revision: e9263b7
multi.jl
## multi.jl - multiprocessing
##
## julia starts with one process, and processors can be added using:
## addprocs_local(n) using exec
## addprocs_ssh({"host1","host2",...}) using remote execution
## addprocs_sge(n) using Sun Grid Engine batch queue
##
## remote_call(w, func, args...) -
## tell a worker to call a function on the given arguments.
## returns a RemoteRef to the result.
##
## remote_do(w, f, args...) - remote function call with no result
##
## wait(rr) - wait for a RemoteRef to be finished computing
##
## fetch(rr) - wait for and get the value of a RemoteRef
##
## remote_call_fetch(w, func, args...) - faster fetch(remote_call(...))
##
## pmap(func, lst) -
## call a function on each element of lst (some 1-d thing), in
## parallel.
##
## RemoteRef() - create an uninitialized RemoteRef on the local processor
##
## RemoteRef(p) - ...or on a particular processor
##
## put(r, val) - store a value to an uninitialized RemoteRef
##
## @spawn expr -
## evaluate expr somewhere. returns a RemoteRef. all variables in expr
## are copied to the remote processor.
##
## @spawnat p expr - @spawn specifying where to run
##
## @async expr -
## run expr as an asynchronous task on the local processor
##
## @parallel (r) for i=1:n ... end -
## parallel loop. the results from each iteration are reduced using (r).
##
## @everywhere expr - run expr everywhere.
# todo:
# - more indexing
# * take() to empty a Ref (full/empty variables)
# * have put() wait on non-empty Refs
# - removing nodes
# - more dynamic scheduling
# * fetch/wait latency seems to be excessive
# * message aggregation
# - timer events
# - send pings at some interval to detect failed/hung machines
# - integrate event loop with other kinds of i/o (non-messages)
# ? method_missing for waiting (ref/assign/localdata seems to cover a lot)
# * serializing closures
# * recover from i/o errors
# * handle remote execution errors
# * all-to-all communication
# * distributed GC
# * call&wait and call&fetch combined messages
# * aggregate GC messages
# * dynamically adding nodes (then always start with 1 and grow)
# * add readline to event loop
# * GOs/darrays on a subset of nodes
## workers and message i/o ##
function send_msg_unknown(s::IO, kind, args)
error("attempt to send to unknown socket")
end
function send_msg(s::IO, kind, args...)
id = worker_id_from_socket(s)
if id > -1
return send_msg(worker_from_id(id), kind, args...)
end
send_msg_unknown(s, kind, args)
end
function send_msg_now(s::IO, kind, args...)
id = worker_id_from_socket(s)
if id > -1
return send_msg_now(worker_from_id(id), kind, args...)
end
send_msg_unknown(s, kind, args)
end
type Worker
host::ByteString
port::Uint16
socket::TcpSocket
sendbuf::IOBuffer
del_msgs::Array{Any,1}
add_msgs::Array{Any,1}
id::Int
gcflag::Bool
Worker(host::String, port::Integer, sock::TcpSocket, id::Int) =
new(bytestring(host), uint16(port), sock, IOBuffer(), {}, {}, id, false)
end
Worker(host::String, port::Integer, sock::TcpSocket) =
Worker(host, port, sock, 0)
Worker(host::String, port::Integer) =
Worker(host, port, connect(host,uint16(port)))
Worker(host::String, port::Integer, tunneluser::String) =
Worker(host, port, connect("localhost",
ssh_tunnel(tunnel_user, host, uint16(port))))
function send_msg_now(w::Worker, kind, args...)
send_msg_(w, kind, args, true)
end
function send_msg(w::Worker, kind, args...)
send_msg_(w, kind, args, false)
end
function flush_gc_msgs(w::Worker)
w.gcflag = false
msgs = w.add_msgs
if !isempty(msgs)
empty!(w.add_msgs)
remote_do(w, add_clients, msgs...)
end
msgs = w.del_msgs
if !isempty(msgs)
empty!(w.del_msgs)
#print("sending delete of $msgs\n")
remote_do(w, del_clients, msgs...)
end
end
#TODO: Move to different Thread
function enq_send_req(sock::TcpSocket,buf,now::Bool)
arr=takebuf_array(buf)
write(sock,arr)
#TODO implement "now"
end
function send_msg_(w::Worker, kind, args, now::Bool)
#println("Sending msg $kind")
buf = w.sendbuf
serialize(buf, kind)
for arg in args
serialize(buf, arg)
end
if !now && w.gcflag
flush_gc_msgs(w)
else
enq_send_req(w.socket,buf,now)
end
end
function flush_gc_msgs()
for w in (PGRP::ProcessGroup).workers
if isa(w,Worker)
k = w::Worker
if k.gcflag
flush_gc_msgs(k)
end
end
end
end
## process group creation ##
type LocalProcess
end
type ProcessGroup
myid::Int
workers::Array{Any,1}
locs::Array{Any,1}
np::Int
# global references
refs::Dict
function ProcessGroup(myid::Integer, w::Array{Any,1}, locs::Array{Any,1})
return new(myid, w, locs, length(w), Dict())
end
end
const PGRP = ProcessGroup(0, {}, {})
function add_workers(PGRP::ProcessGroup, w::Array{Any,1})
n = length(w)
locs = map(x->(x.host,x.port), w)
# NOTE: currently only node 1 can add new nodes, since nobody else
# has the full list of address:port
newlocs = [PGRP.locs, locs]
for i=1:n
push!(PGRP.workers, w[i])
w[i].id = PGRP.np+i
send_msg_now(w[i], w[i].id, newlocs)
create_message_handler_loop(w[i].socket)
end
PGRP.locs = newlocs
PGRP.np += n
:ok
end
myid() = PGRP.myid
nprocs() = PGRP.np
function worker_from_id(i)
pg = PGRP::ProcessGroup
while i > length(pg.workers) || pg.workers[i]===nothing
sleep(0.1)
yield()
end
pg.workers[i]
end
function worker_id_from_socket(s)
for i=1:nprocs()
w = worker_from_id(i)
if isa(w,Worker)
if is(s, w.socket) || is(s, w.sendbuf)
return i
end
end
end
if isa(s,IOStream) && fd(s)==-1
# serializing to a local buffer
return myid()
end
return -1
end
# establish a Worker connection for processes that connected to us
function identify_socket(otherid, sock)
i = otherid
#locs = PGRP.locs
@assert i > PGRP.myid
d = i-length(PGRP.workers)
if d > 0
resize!(PGRP.workers, i)
PGRP.workers[(end-d+1):end] = nothing
PGRP.np += d
end
PGRP.workers[i] = Worker("", 0, sock, i)
#write(STDOUT, "$(PGRP.myid) heard from $i\n")
nothing
end
## remote refs and core messages: do, call, fetch, wait, ref, put ##
const client_refs = WeakKeyDict()
type RemoteRef
where::Int
whence::Int
id::Int
# TODO: cache value if it's fetched, but don't serialize the cached value
function RemoteRef(w, wh, id)
r = new(w,wh,id)
found = getkey(client_refs, r, false)
if !is(found,false)
return found
end
client_refs[r] = true
finalizer(r, send_del_client)
r
end
REQ_ID::Int = 0
function RemoteRef(pid::Integer)
rr = RemoteRef(pid, myid(), REQ_ID)
REQ_ID += 1
if mod(REQ_ID,200) == 0
# force gc after making a lot of refs since they take up
# space on the machine where they're stored, yet the client
# is responsible for freeing them.
gc()
end
rr
end
RemoteRef(w::LocalProcess) = RemoteRef(myid())
RemoteRef(w::Worker) = RemoteRef(w.id)
RemoteRef() = RemoteRef(myid())
global WeakRemoteRef
function WeakRemoteRef(w, wh, id)
return new(w, wh, id)
end
function WeakRemoteRef(pid::Integer)
rr = WeakRemoteRef(pid, myid(), REQ_ID)
REQ_ID += 1
if mod(REQ_ID,200) == 0
gc()
end
rr
end
WeakRemoteRef(w::LocalProcess) = WeakRemoteRef(myid())
WeakRemoteRef(w::Worker) = WeakRemoteRef(w.id)
WeakRemoteRef() = WeakRemoteRef(myid())
end
hash(r::RemoteRef) = hash(r.whence)+3*hash(r.id)
isequal(r::RemoteRef, s::RemoteRef) = (r.whence==s.whence && r.id==s.id)
rr2id(r::RemoteRef) = (r.whence, r.id)
bottom_func() = assert(false)
function lookup_ref(id)
GRP = PGRP::ProcessGroup
wi = get(GRP.refs, id, ())
if is(wi, ())
# first we've heard of this ref
wi = WorkItem(bottom_func)
# this WorkItem is just for storing the result value
GRP.refs[id] = wi
add!(wi.clientset, id[1])
end
wi
end
# is a ref uninitialized? (for locally-owned refs only)
#function ref_uninitialized(id)
# wi = lookup_ref(id)
# !wi.done && is(wi.thunk,bottom_func)
#end
#ref_uninitialized(r::RemoteRef) = (assert(r.where==myid());
# ref_uninitialized(rr2id(r)))
function isready(rr::RemoteRef)
rid = rr2id(rr)
if rr.where == myid()
lookup_ref(rid).done
else
remote_call_fetch(rr.where, id->lookup_ref(id).done, rid)
end
end
function del_client(id, client)
wi = lookup_ref(id)
delete!(wi.clientset, client)
if isempty(wi.clientset)
delete!((PGRP::ProcessGroup).refs, id)
#print("$(myid()) collected $id\n")
end
nothing
end
function del_clients(pairs::(Any,Any)...)
for p in pairs
del_client(p[1], p[2])
end
end
function send_del_client(rr::RemoteRef)
if rr.where == myid()
del_client(rr2id(rr), myid())
else
w = worker_from_id(rr.where)
push!(w.del_msgs, (rr2id(rr), myid()))
w.gcflag = true
end
end
function add_client(id, client)
wi = lookup_ref(id)
add!(wi.clientset, client)
nothing
end
function add_clients(pairs::(Any,Any)...)
for p in pairs
add_client(p[1], p[2])
end
end
function send_add_client(rr::RemoteRef, i)
if rr.where == myid()
add_client(rr2id(rr), i)
elseif i != rr.where
# don't need to send add_client if the message is already going
# to the processor that owns the remote ref. it will add_client
# itself inside deserialize().
w = worker_from_id(rr.where)
push!(w.add_msgs, (rr2id(rr), i))
w.gcflag = true
end
end
function serialize(s, rr::RemoteRef)
i = worker_id_from_socket(s)
if i != -1
send_add_client(rr, i)
end
invoke(serialize, (Any, Any), s, rr)
end
function deserialize(s, t::Type{RemoteRef})
rr = invoke(deserialize, (Any, CompositeKind), s, t)
where = rr.where
if where == myid()
add_client(rr2id(rr), myid())
end
# call ctor to make sure this rr gets added to the client_refs table
RemoteRef(where, rr.whence, rr.id)
end
schedule_call(rid, f_thk, args_thk) =
schedule_call(rid, ()->apply(f_thk, args_thk))
function schedule_call(rid, thunk)
wi = WorkItem(thunk)
(PGRP::ProcessGroup).refs[rid] = wi
add!(wi.clientset, rid[1])
enq_work(wi)
wi
end
#localize_ref(b::Box) = Box(localize_ref(b.contents))
#function localize_ref(r::RemoteRef)
# if r.where == myid()
# fetch(r)
# else
# r
# end
#end
#localize_ref(x) = x
# make a thunk to call f on args in a way that simulates what would happen if
# the function were sent elsewhere
function local_remote_call_thunk(f, args)
if isempty(args)
return f
end
return ()->f(args...)
# TODO: this seems to be capable of causing deadlocks by waiting on
# Refs buried inside the closure that we don't want to wait on yet.
# linfo = ccall(:jl_closure_linfo, Any, (Any,), f)
# if isa(linfo,LambdaStaticData)
# env = ccall(:jl_closure_env, Any, (Any,), f)
# buf = memio()
# serialize(buf, env)
# seek(buf, 0)
# env = deserialize(buf)
# f = ccall(:jl_new_closure, Any, (Ptr{Void}, Any, Any),
# C_NULL, env, linfo)::Function
# end
# f(map(localize_ref,args)...)
end
function remote_call(w::LocalProcess, f, args...)
rr = RemoteRef(w)
schedule_call(rr2id(rr), local_remote_call_thunk(f,args))
rr
end
function remote_call(w::Worker, f, args...)
rr = RemoteRef(w)
#println("$(myid()) asking for $rr")
send_msg(w, :call, rr2id(rr), f, args)
rr
end
remote_call(id::Integer, f, args...) = remote_call(worker_from_id(id), f, args...)
# faster version of fetch(remote_call(...))
function remote_call_fetch(w::LocalProcess, f, args...)
rr = WeakRemoteRef(w)
oid = rr2id(rr)
wi = schedule_call(oid, local_remote_call_thunk(f,args))
wi.notify = ((), :call_fetch, oid, wi.notify)
yield(WaitFor(:call_fetch, rr))
end
function remote_call_fetch(w::Worker, f, args...)
# can be weak, because the program will have no way to refer to the Ref
# itself, it only gets the result.
rr = WeakRemoteRef(w)
oid = rr2id(rr)
send_msg(w, :call_fetch, oid, f, args)
yield(WaitFor(:call_fetch, rr))
end
remote_call_fetch(id::Integer, f, args...) =
remote_call_fetch(worker_from_id(id), f, args...)
# faster version of wait(remote_call(...))
remote_call_wait(w::LocalProcess, f, args...) = wait(remote_call(w,f,args...))
function remote_call_wait(w::Worker, f, args...)
rr = RemoteRef(w)
oid = rr2id(rr)
send_msg(w, :call_wait, oid, f, args)
yield(WaitFor(:wait, rr))
end
remote_call_wait(id::Integer, f, args...) =
remote_call_wait(worker_from_id(id), f, args...)
function remote_do(w::LocalProcess, f, args...)
# the LocalProcess version just performs in local memory what a worker
# does when it gets a :do message.
# same for other messages on LocalProcess.
enq_work(WorkItem(local_remote_call_thunk(f, args)))
nothing
end
function remote_do(w::Worker, f, args...)
send_msg(w, :do, f, args)
nothing
end
remote_do(id::Integer, f, args...) = remote_do(worker_from_id(id), f, args...)
function sync_msg(verb::Symbol, r::RemoteRef)
pg = (PGRP::ProcessGroup)
oid = rr2id(r)
if r.where==myid() || isa(pg.workers[r.where], LocalProcess)
wi = lookup_ref(oid)
if wi.done
if is(verb,:fetch)
return work_result(wi)
else
return r
end
else
# add to WorkItem's notify list
wi.notify = ((), verb, oid, wi.notify)
end
else
send_msg(worker_from_id(r.where), verb, oid)
end
# yield to event loop, return here when answer arrives
v = yield(WaitFor(verb, r))
return is(verb,:fetch) ? v : r
end
wait(r::RemoteRef) = sync_msg(:wait, r)
fetch(r::RemoteRef) = sync_msg(:fetch, r)
fetch(x::ANY) = x
# writing to an uninitialized ref
function put_ref(rid, val::ANY)
wi = lookup_ref(rid)
while wi.done
wi.notify = ((), :take, rid, wi.notify)
yield(WaitFor(:take, RemoteRef(myid(), rid[1], rid[2])))
end
wi.result = val
wi.done = true
notify_done(wi)
end
function put(rr::RemoteRef, val::ANY)
rid = rr2id(rr)
if rr.where == myid()
put_ref(rid, val)
else
remote_call_fetch(rr.where, put_ref, rid, val)
end
val
end
function take_ref(rid)
wi = lookup_ref(rid)
while !wi.done
wait(RemoteRef(myid(), rid[1], rid[2]))
end
val = wi.result
wi.done = false
notify_empty(wi)
val
end
function take(rr::RemoteRef)
rid = rr2id(rr)
if rr.where == myid()
take_ref(rid)
else
remote_call_fetch(rr.where, take_ref, rid)
end
end
## work queue ##
type WorkItem
thunk::Function
task # the Task working on this item, or ()
done::Bool
result
notify::Tuple
argument # value to pass task next time it is restarted
clientset::IntSet
WorkItem(thunk::Function) = new(thunk, (), false, (), (), (), IntSet())
WorkItem(task::Task) = new(()->(), task, false, (), (), (), IntSet())
end
function work_result(w::WorkItem)
v = w.result
if isa(v,WeakRef)
v = v.value
end
v
end
type WaitFor
msg::Symbol
rr
end
global work_cb, fgcm_cb
function enq_work(wi::WorkItem)
unshift!(Workqueue, wi)
queueAsync(work_cb::SingleAsyncWork)
end
enq_work(f::Function) = enq_work(WorkItem(f))
enq_work(t::Task) = enq_work(WorkItem(t))
function perform_work()
job = pop!(Workqueue)
perform_work(job)
end
function perform_work(job::WorkItem)
local result
try
if isa(job.task,Task)
# continuing interrupted work item
arg = job.argument
job.argument = ()
job.task.runnable = true
result = is(arg,()) ? yieldto(job.task) : yieldto(job.task, arg)
else
job.task = Task(job.thunk)
job.task.storage = nothing
result = yieldto(job.task)
end
catch err
print("exception on ", myid(), ": ")
show(add_backtrace(err,backtrace()))
println()
result = err
end
# restart job by yielding back to whatever task just switched to us
job.task = current_task().last
if istaskdone(job.task)
# job done
job.done = true
job.result = result
end
if job.done
job.task = ()
# do notifications
notify_done(job)
job.thunk = bottom_func # avoid reference retention
elseif isa(result,WaitFor)
job.task.runnable = false
wf::WaitFor = result
if wf.msg === :consume
P = wf.rr::Task
# queue consumers waiting for producer P. note that in order
# to avoid scheduling overhead for a typical produce/consume,
# there is no fairness unless consumers explicitly yield().
if P.consumers === nothing
P.consumers = {job}
else
unshift!(P.consumers, job)
end
else
# add to waiting set to wait on a sync event
rr = wf.rr
#println("$(myid()) waiting for $rr")
oid = rr2id(rr)
waitinfo = (wf.msg, job, rr)
waiters = get(Waiting, oid, false)
if isequal(waiters,false)
Waiting[oid] = {waitinfo}
else
push!(waiters, waitinfo)
end
end
elseif job.task.runnable
# otherwise return to queue
enq_work(job)
end
end
function deliver_result(sock::IO, msg, oid, value)
#print("$(myid()) sending result $oid\n")
if is(msg,:fetch) || is(msg,:call_fetch)
val = value
else
@assert is(msg, :wait)
val = oid
end
try
send_msg_now(sock, :result, msg, oid, val)
catch err
# send exception in case of serialization error; otherwise
# request side would hang.
send_msg_now(sock, :result, msg, oid, err)
end
end
const empty_cell_ = {}
function deliver_result(sock::(), msg, oid, value)
# restart task that's waiting on oid
jobs = get(Waiting, oid, empty_cell_)
for i = 1:length(jobs)
j = jobs[i]
if j[1]==msg
job = j[2]
job.argument = value
enq_work(job)
delete!(jobs, i)
break
end
end
if isempty(jobs) && !is(jobs,empty_cell_)
delete!(Waiting, oid)
end
nothing
end
notify_done (job::WorkItem) = notify_done(job, false)
notify_empty(job::WorkItem) = notify_done(job, true)
# notify waiters that a certain job has finished or Ref has been emptied
function notify_done(job::WorkItem, take)
newnot = ()
while !is(job.notify,())
(sock, msg, oid, job.notify) = job.notify
if take == is(msg,:take)
let wr = work_result(job)
deliver_result(sock, msg, oid, wr)
if is(msg,:call_fetch)
# can delete the ref right away since we know it is
# unreferenced by the client
delete!((PGRP::ProcessGroup).refs, oid)
end
end
else
newnot = (sock, msg, oid, newnot)
end
end
job.notify = newnot
nothing
end
## message event handlers ##
# activity on accept fd
function accept_handler(server::TcpSocket, status::Int32)
if(status == -1)
error("An error occured during the creation of the server")
end
client = accept(server)
create_message_handler_loop(client)
end
type DisconnectException <: Exception end
function create_message_handler_loop(sock::AsyncStream) #returns immediately
enq_work(@task begin
global PGRP
#println("message_handler_loop")
start_reading(sock)
wait_connected(sock)
if PGRP.np == 0
# first connection; get process group info from client
PGRP.myid = deserialize(sock)
PGRP.locs = locs = deserialize(sock)
#print("\nLocation: ",locs,"\nId:",PGRP.myid,"\n")
# joining existing process group
PGRP.np = length(PGRP.locs)
PGRP.workers = w = cell(PGRP.np)
w[1] = Worker("", 0, sock, 1)
for i = 2:(PGRP.myid-1)
w[i] = Worker(locs[i][1], locs[i][2])
w[i].id = i
create_message_handler_loop(w[i].socket)
send_msg_now(w[i], :identify_socket, PGRP.myid)
end
w[PGRP.myid] = LocalProcess()
for i = (PGRP.myid+1):PGRP.np
w[i] = nothing
end
end
#println("loop")
while true
#try
msg = deserialize(sock)
#println("got msg: ",msg)
# handle message
if is(msg, :call) || is(msg, :call_fetch) || is(msg, :call_wait)
id = deserialize(sock)
f = deserialize(sock)
args = deserialize(sock)
#print("$(myid()) got call $id\n")
wi = schedule_call(id, f, args)
if is(msg, :call_fetch)
wi.notify = (sock, :call_fetch, id, wi.notify)
elseif is(msg, :call_wait)
wi.notify = (sock, :wait, id, wi.notify)
end
elseif is(msg, :do)
f = deserialize(sock)
args = deserialize(sock)
#print("got args: $args\n")
let func=f, ar=args
enq_work(WorkItem(()->apply(func, ar)))
end
elseif is(msg, :result)
# used to deliver result of wait or fetch
mkind = deserialize(sock)
oid = deserialize(sock)
val = deserialize(sock)
deliver_result((), mkind, oid, val)
elseif is(msg, :identify_socket)
otherid = deserialize(sock)
identify_socket(otherid, sock)
else
# the synchronization messages
oid = deserialize(sock)::(Int,Int)
wi = lookup_ref(oid)
if wi.done
deliver_result(sock, msg, oid, work_result(wi))
else
# add to WorkItem's notify list
# TODO: should store the worker here, not the socket,
# so we don't need to look up the worker later
wi.notify = (sock, msg, oid, wi.notify)
end
end
#catch e
# if isa(e,EOFError)
#print("eof. $(myid()) exiting\n")
# stop_reading(sock)
# # TODO: remove machine from group
# throw(DisconnectException())
# else
# print("deserialization: ", e, "\n")
# #while nb_available(sock) > 0 #|| select(sock)
# # read(sock, Uint8)
# #end
#end
#end
end
end)
end
## worker creation and setup ##
# the entry point for julia worker processes. does not return.
# argument is descriptor to write listening port # to.
start_worker() = start_worker(STDOUT)
function start_worker(out::IO)
default_port = uint16(9009)
(actual_port,sock) = open_any_tcp_port(accept_handler,default_port)
write(out, "julia_worker:") # print header
write(out, "$(dec(actual_port))#") # print port
write(out, bind_addr) #TODO: print hostname
write(out, '\n')
# close STDIN; workers will not use it
#close(STDIN)
ccall(:jl_install_sigint_handler, Void, ())
global const Scheduler = current_task()
try
event_loop(false)
catch err
print("unhandled exception on $(myid()): $(err)\nexiting.\n")
end
close(sock)
exit(0)
end
start_remote_workers(m, c) = start_remote_workers(m, c, false)
function start_remote_workers(machines, cmds, tunnel)
n = length(cmds)
outs = cell(n)
w = cell(n)
for i=1:n
outs[i],_ = readsfrom(cmds[i])
outs[i].line_buffered = true
end
for i=1:n
local hostname::String, port::Int16
stream = outs[i]
stream.line_buffered = true
while true
wait_readline(stream)
conninfo = readline(stream.buffer)
hostname, port = parse_connection_info(conninfo)
if hostname != ""
break
end
end
if tunnel
s = split(machines[i],'@')
if length(s) > 1
user = s[1]
else
user = ENV["USER"]
end
w[i] = Worker(hostname, port, user)
else
w[i] = Worker(hostname, port)
end
let wrker = w[i]
# redirect console output from workers to the client's stdout:
start_reading(stream,function(stream::AsyncStream,nread::Int)
if(nread>0)
try
line = readbytes(stream.buffer, nread)
print("\tFrom worker $(wrker.id):\t",line)
catch err
println("\tError parsing reply from worker $(wrker.id):\t",err)
return false
end
end
true
end)
end
end
w
end
function parse_connection_info(str)
m = match(r"^julia_worker:(\d+)#(.*)", str)
if m != nothing
(m.captures[2], parse_int(Int16, m.captures[1]))
else
("", int16(-1))
end
end
tunnel_port = 9201
# establish an SSH tunnel to a remote worker
# returns P such that localhost:P connects to host:port
ssh_tunnel(host, port) = ssh_tunnel(ENV["USER"], host, port)
function ssh_tunnel(user, host, port)
global tunnel_port
localp = tunnel_port::Int
while !success(`ssh -f -o ExitOnForwardFailure=yes $(user)@$host -L $localp:$host:$port -N`)
localp += 1
end
tunnel_port = localp+1
localp
end
function worker_ssh_cmd(host)
`ssh -n $host "bash -l -c \"cd $JULIA_HOME && ./julia-release-basic --worker\""`
end
#function worker_ssh_cmd(host, key)
# `ssh -i $key -n $host "bash -l -c \"cd $JULIA_HOME && ./julia-release-basic --worker\""`
#end
function addprocs_ssh(machines)
add_workers(PGRP, start_remote_workers(machines, map(worker_ssh_cmd, machines)))
end
# start processes via SSH, then connect to them through an SSH tunnel.
# the tunnel is only used from the head (process 1); the nodes are assumed
# to be mutually reachable without a tunnel, as is often the case in a cluster.
function addprocs_ssh_tunnel(machines)
add_workers(PGRP, start_remote_workers(machines, map(worker_ssh_cmd, machines), true))
end
#function addprocs_ssh(machines, keys)
# if !(isa(keys, Array)) && isa(machines,Array)
# key = keys
# keys = [ key for x = 1:length(machines)]
# cmdargs = { {machines[x],keys[x]} for x = 1:length(machines)}
# else
# cmdargs = {{machines,keys}}
# end #if/else
# add_workers(PGRP, start_remote_workers(machines, map(x->worker_ssh_cmd(x[1],x[2]), cmdargs)))
#end
worker_local_cmd() = `$JULIA_HOME/julia-release-basic --bind-to $bind_addr --worker`
addprocs_local(np::Integer) =
add_workers(PGRP, start_remote_workers({ "localhost" for i=1:np },
{ worker_local_cmd() for i=1:np }))
function start_sge_workers(n)
home = JULIA_HOME
sgedir = "$home/../../SGE"
run(`mkdir -p $sgedir`)
qsub_cmd = `echo $home/julia-release-basic --worker` | `qsub -N JULIA -terse -e $sgedir -o $sgedir -t 1:$n`
out,_ = readsfrom(qsub_cmd)
if !success(qsub_cmd)
error("batch queue not available (could not run qsub)")
end
id = chomp(split(readline(out),'.')[1])
println("job id is $id")
print("waiting for job to start");
workers = cell(n)
for i=1:n
# wait for each output stream file to get created
fname = "$sgedir/JULIA.o$(id).$(i)"
local fl, hostname, port
fexists = false
sleep(0.5)
while !fexists
try
fl = open(fname)
try
conninfo = readline(fl)
hostname, port = parse_connection_info(conninfo)
finally
close(fl)
end
fexists = (hostname != "")
catch
print(".");
sleep(0.5)
end
end
#print("hostname=$hostname, port=$port\n")
workers[i] = Worker(hostname, port)
end
print("\n")
workers
end
addprocs_sge(n) = add_workers(PGRP, start_sge_workers(n))
## higher-level functions: spawn, pmap, pfor, etc. ##
sync_begin() = task_local_storage(:SPAWNS, ({}, get(task_local_storage(), :SPAWNS, ())))
function sync_end()
spawns = get(task_local_storage(), :SPAWNS, ())
if is(spawns,())
error("sync_end() without sync_begin()")
end
refs = spawns[1]
task_local_storage(:SPAWNS, spawns[2])
for r in refs
wait(r)
end
end
macro sync(block)
quote
sync_begin()
v = $(esc(block))
sync_end()
v
end
end
function sync_add(r)
spawns = get(task_local_storage(), :SPAWNS, ())
if !is(spawns,())
push!(spawns[1], r)
end
r
end
spawnat(p, thunk) = sync_add(remote_call(p, thunk))
let lastp = 1
global chooseproc
function chooseproc(thunk::Function)
p = -1
env = thunk.env
if isa(env,Tuple)
for v in env
if isa(v,Box)
v = v.contents
end
if isa(v,RemoteRef)
p = v.where; break
end
end
end
if p == -1
p = lastp; lastp += 1
if lastp > nprocs()
lastp = 1
end
end
p
end
end
spawn_somewhere(thunk) = spawnat(chooseproc(thunk),thunk)
find_vars(e) = find_vars(e, {})
function find_vars(e, lst)
if isa(e,Symbol)
if !isdefined(e) || isconst(e)
# exclude global constants
else
push!(lst, e)
end
elseif isa(e,Expr)
for x in e.args
find_vars(x,lst)
end
end
lst
end
# wrap an expression in "let a=a,b=b,..." for each var it references
function localize_vars(expr)
v = find_vars(expr)
# requires a special feature of the front end that knows how to insert
# the correct variables. the list of free variables cannot be computed
# from a macro.
Expr(:localize, {:(()->($expr)), v...}, Any)
end
macro spawn(expr)
expr = localize_vars(:(()->($expr)))
:(spawn_somewhere($(esc(expr))))
end
function spawnlocal(thunk)
rr = RemoteRef(myid())
sync_add(rr)
rid = rr2id(rr)
wi = WorkItem(thunk)
(PGRP::ProcessGroup).refs[rid] = wi
add!(wi.clientset, rid[1])
push!(Workqueue, wi) # add to the *front* of the queue, work first
queueAsync(work_cb::SingleAsyncWork)
yield()
rr
end
macro async(expr)
expr = localize_vars(:(()->($expr)))
:(spawnlocal($(esc(expr))))
end
macro spawnlocal(expr)
warn_once("@spawnlocal is deprecated, use @async instead.")
:(@async $(esc(expr)))
end
macro spawnat(p, expr)
expr = localize_vars(:(()->($expr)))
:(spawnat($(esc(p)), $(esc(expr))))
end
function at_each(f, args...)
for i=1:nprocs()
sync_add(remote_call(i, f, args...))
end
end
macro everywhere(ex)
quote
@sync begin
at_each(()->eval(Main,$(expr(:quote,ex))))
end
end
end
function pmap_static(f, lsts...)
np = nprocs()
n = length(lsts[1])
{ remote_call((i-1)%np+1, f, map(L->L[i], lsts)...) for i = 1:n }
end
pmap(f) = f()
# dynamic scheduling by creating a local task to feed work to each processor
# as it finishes.
# example unbalanced workload:
# rsym(n) = (a=rand(n,n);a*a')
# L = {rsym(200),rsym(1000),rsym(200),rsym(1000),rsym(200),rsym(1000),rsym(200),rsym(1000)};
# pmap(eig, L);
function pmap(f, lsts...)
np = nprocs()
n = length(lsts[1])
results = cell(n)
i = 1
# function to produce the next work item from the queue.
# in this case it's just an index.
next_idx() = (idx=i; i+=1; idx)
@sync begin
for p=1:np
if p != myid() || np == 1
@spawnat myid() begin
while true
idx = next_idx()
if idx > n
break
end
results[idx] = remote_call_fetch(p, f,
map(L->L[idx], lsts)...)
end
end
end
end
end
results
end
function preduce(reducer, f, r::Range1{Int})
np = nprocs()
N = length(r)
each = div(N,np)
rest = rem(N,np)
if each < 1
return fetch(@spawn f(first(r), first(r)+N-1))
end
results = cell(np)
for i=1:np
lo = first(r) + (i-1)*each
hi = lo + each-1
if i==np
hi += rest
end
results[i] = @spawn f(lo, hi)
end
mapreduce(fetch, reducer, results)
end
function pfor(f, r::Range1{Int})
np = nprocs()
N = length(r)
each = div(N,np)
rest = rem(N,np)
if each < 1
@spawn f(first(r), first(r)+N-1)
return
end
for i=1:np
lo = first(r) + (i-1)*each
hi = lo + each-1
if i==np
hi += rest
end
@spawn f(lo,hi)
end
nothing
end
function make_preduce_body(reducer, var, body)
localize_vars(
quote
function (lo::Int, hi::Int)
$(esc(var)) = lo
ac = $(esc(body))
for $(esc(var)) = (lo+1):hi
ac = ($(esc(reducer)))(ac, $(esc(body)))
end
ac
end
end
)
end
function make_pfor_body(var, body)
localize_vars(
quote
function (lo::Int, hi::Int)
for $(esc(var)) = lo:hi
$(esc(body))
end
end
end
)
end
macro parallel(args...)
na = length(args)
if na==1
loop = args[1]
if isa(loop,Expr) && loop.head === :comprehension
ex = loop.args[1]
loop.args[1] = esc(ex)
nd = length(loop.args)-1
ranges = map(e->esc(e.args[2]), loop.args[2:])
for i=1:nd
var = loop.args[1+i].args[1]
loop.args[1+i] = :( $(esc(var)) = ($(ranges[i]))[I[$i]] )
end
return :( DArray((I::(Range1{Int}...))->($loop),
tuple($(map(r->:(length($r)),ranges)...))) )
end
elseif na==2
reducer = args[1]
loop = args[2]
else
throw(ArgumentError("wrong number of arguments to @parallel"))
end
if !isa(loop,Expr) || !is(loop.head,:for)
error("malformed @parallel loop")
end
var = loop.args[1].args[1]
r = loop.args[1].args[2]
body = loop.args[2]
if na==1
quote
pfor($(make_pfor_body(var, body)), $(esc(r)))
end
else
quote
preduce($(esc(reducer)),
$(make_preduce_body(reducer, var, body)), $(esc(r)))
end
end
end
## demos ##
# fv(a)=eig(a)[2][2]
# A=randn(800,800);A=A*A';
# pmap(fv, {A,A,A})
#all2all() = at_each(hello_from, myid())
#hello_from(i) = print("message from $i to $(myid())\n")
# monte carlo estimate of pi
# function buffon(niter)
# nc =
# @parallel (+) for i=1:niter
# rand() <= sin(rand()*pi/2) ? 1 : 0
# end
# 2/(nc/niter)
# end
## event processing, I/O and work scheduling ##
function yield(args...)
ct = current_task()
# preserve Task.last across calls to the scheduler
prev = ct.last
v = yieldto(Scheduler, args...)
ct.last = prev
return v
end
function _jl_work_cb(args...)
if !isempty(Workqueue)
perform_work()
else
queueAsync(fgcm_cb::SingleAsyncWork)
end
if !isempty(Workqueue)
# really this should just make process_event be non-blocking
queueAsync(work_cb::SingleAsyncWork)
end
end
function event_loop(isclient)
global work_cb = SingleAsyncWork(eventloop(), _jl_work_cb)
global fgcm_cb = SingleAsyncWork(eventloop(), (args...)->flush_gc_msgs());
queueAsync(work_cb::SingleAsyncWork)
iserr, lasterr = false, ()
while true
try
if iserr
show(lasterr)
iserr, lasterr = false, ()
else
run_event_loop()
end
catch err
bt = backtrace()
if isa(err,DisconnectException)
# TODO: wake up tasks waiting for failed process
if !isclient
return
end
elseif isclient && isa(err,InterruptException)
# root task is waiting for something on client. allow C-C
# to interrupt.
interrupt_waiting_task(roottask_wi,err)
end
iserr, lasterr = true, add_backtrace(err,bt)
end
end
end
# force a task to stop waiting, providing with_value as the value of
# whatever it's waiting for.
function interrupt_waiting_task(wi::WorkItem, with_value)
for (oid, jobs) in Waiting
for j in jobs
if is(j[2], wi)
deliver_result((), j[1], oid, with_value)
return
end
end
end
wi.argument = with_value
enq_work(wi)
end