Add a watcher mechanism to detect when Distributed might be in use

This should help users figure out if one of their packages is using Distributed and
another is using DistributedNext.

Project.toml

@@ -7,10 +7,14 @@ Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 Serialization = "9e88b42a-f829-5b0c-bbe9-9e923198166b"
 Sockets = "6462fe0b-24de-5631-8697-dd941f90decc"
 
+[compat]
+Distributed = "1"
+
 [extras]
+Distributed = "8ba89e20-285c-5b6f-9357-94700520ee1b"
 LibSSH = "00483490-30f8-4353-8aba-35b82f51f4d0"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
 [targets]
-test = ["LinearAlgebra", "Test", "LibSSH"]
+test = ["LinearAlgebra", "Test", "LibSSH", "Distributed"]

docs/src/_changelog.md

@@ -9,6 +9,11 @@ This documents notable changes in DistributedNext.jl. The format is based on
 
 ## Unreleased
 
+### Added
+- A watcher mechanism has been added to detect when both the Distributed stdlib
+  and DistributedNext may be active and adding workers. This should help prevent
+  incompatibilities from both libraries being used simultaneously ([#10]).
+
 ### Fixed
 - Fixed behaviour of `isempty(::RemoteChannel)`, which previously had the
   side-effect of taking an element from the channel ([#3]).

src/DistributedNext.jl

@@ -72,7 +72,17 @@ export
 # Used only by shared arrays.
     check_same_host
 
+distributed_module::Union{Module, Nothing} = nothing
+
+function _find_distributed_stdlib()
+    # Find the Distributed module if it's been loaded
+    distributed_pkgid = Base.PkgId(Base.UUID("8ba89e20-285c-5b6f-9357-94700520ee1b"), "Distributed")
+    global distributed_module = get(Base.loaded_modules, distributed_pkgid, nothing)
+end
+
 function _require_callback(mod::Base.PkgId)
+    _find_distributed_stdlib()
+
     if Base.toplevel_load[] && myid() == 1 && nprocs() > 1
         # broadcast top-level (e.g. from Main) import/using from node 1 (only)
         @sync for p in procs()
@@ -116,6 +126,24 @@ include("precompile.jl")
 
 function __init__()
     init_parallel()
+
+    # Start a task to watch for the Distributed stdlib being loaded and
+    # initialized to support multiple workers. We do this by checking if the
+    # cluster cookie has been set, which is most likely to have been done
+    # through Distributed.init_multi() being called by Distributed.addprocs() or
+    # something.
+    _find_distributed_stdlib()
+    watcher_task = Threads.@spawn while true
+        if !isnothing(distributed_module)
+            if isdefined(distributed_module.LPROC, :cookie) && inited[]
+                @warn "DistributedNext has detected that the Distributed stdlib may be in use. Be aware that these libraries are not compatible, you should use either one or the other."
+                return
+            end
+        end
+
+        sleep(0.5)
+    end
+    errormonitor(watcher_task)
 end
 
 end

test/distributed_exec.jl

@@ -1,6 +1,6 @@
 # This file is a part of Julia. License is MIT: https://julialang.org/license
 
-using Test, DistributedNext, Random, Serialization, Sockets
+using DistributedNext, Random, Serialization, Sockets
 import DistributedNext: launch, manage
 
 

test/distributed_stdlib_detection.jl

@@ -0,0 +1,24 @@
+using DistributedNext
+
+@testset "Distributed.jl detection" begin
+    # Just loading Distributed should do nothing
+    cmd = `$test_exename $test_exeflags -e 'using Distributed, DistributedNext; sleep(1)'`
+    stderr_buf = IOBuffer()
+    run(pipeline(cmd; stderr=stderr_buf))
+    stderr_str = String(take!(stderr_buf))
+    @test isempty(stderr_str)
+
+    # Only one of the two being active should also do nothing
+    cmd = `$test_exename $test_exeflags -e 'using Distributed, DistributedNext; Distributed.init_multi(); sleep(1)'`
+    stderr_buf = IOBuffer()
+    run(pipeline(cmd; stderr=stderr_buf))
+    stderr_str = String(take!(stderr_buf))
+    @test isempty(stderr_str)
+
+    # But both being active at the same time should trigger a warning
+    cmd = `$test_exename $test_exeflags -e 'using Distributed, DistributedNext; Distributed.init_multi(); DistributedNext.init_multi(); sleep(1)'`
+    stderr_buf = IOBuffer()
+    run(pipeline(cmd; stderr=stderr_buf))
+    stderr_str = String(take!(stderr_buf))
+    @test contains(stderr_str, "DistributedNext has detected that the Distributed stdlib may be in use")
+end

test/runtests.jl

@@ -1,5 +1,7 @@
 # This file is a part of Julia. License is MIT: https://julialang.org/license
 
+using Test
+
 # Run the distributed test outside of the main driver since it needs its own
 # set of dedicated workers.
 include(joinpath(Sys.BINDIR, "..", "share", "julia", "test", "testenv.jl"))
@@ -18,3 +20,5 @@ end
 include("distributed_exec.jl")
 
 include("managers.jl")
+
+include("distributed_stdlib_detection.jl")