Dagger.workers() do not include DistributedNext workers ?

[ederag]

Trying to delegate work to a remote workstation through ssh tunneling,
Run a task on a specific Distributed worker seemed appropriate, but the first line on the example fails:

julia> using DistributedNext

julia> addprocs([("proxybasalt", :auto)], tunnel=true);

julia> using Dagger

julia> sc_workers = Dagger.scope(worker = 3)
ERROR: no process with id 3 exists
Stacktrace:
  [1] error(s::String)
    @ Base ./error.jl:35
  [2] worker_from_id(pg::Distributed.ProcessGroup, i::Int64)
    @ Distributed /usr/local/build/julia/custom/1.10.9/share/julia/stdlib/v1.10/Distributed/src/cluster.jl:1098
  [3] worker_from_id
    @ /usr/local/build/julia/custom/1.10.9/share/julia/stdlib/v1.10/Distributed/src/cluster.jl:1090 [inlined]
  [4] remotecall_fetch
    @ /usr/local/build/julia/custom/1.10.9/share/julia/stdlib/v1.10/Distributed/src/remotecall.jl:492 [inlined]
  [5] #228
    @ /usr/local/build/julia/depot/packages/Dagger/0XSAN/src/utils/system_uuid.jl:38 [inlined]
  [6] get!(default::Dagger.var"#228#229"{Int64}, h::Dict{Int64, Base.UUID}, key::Int64)
    @ Base ./dict.jl:479
  [7] system_uuid
    @ /usr/local/build/julia/depot/packages/Dagger/0XSAN/src/utils/system_uuid.jl:37 [inlined]
  [8] ProcessScope
    @ /usr/local/build/julia/depot/packages/Dagger/0XSAN/src/scopes.jl:76 [inlined]
  [9] to_scope(sc::@NamedTuple{worker::Int64})
    @ Dagger /usr/local/build/julia/depot/packages/Dagger/0XSAN/src/scopes.jl:344
 [10] #scope#116
    @ /usr/local/build/julia/depot/packages/Dagger/0XSAN/src/scopes.jl:283 [inlined]
 [11] top-level scope
    @ REPL[4]:1

julia> 3 in workers()
true

julia> Dagger.workers()
1-element Vector{Int64}:
 1

I'd expect Dagger to make use of the DistributedNext workers.

Versions:

julia-1.10.9
  [fab6aee4] DistributedNext v1.0.0
  [d58978e5] Dagger v0.18.15

[m-fila]

Did you set Dagger preferences to use DistributedNext instead of Distributed?

julia --project -e "using Dagger; Dagger.set_distributed_package!(\"DistributedNext\")"

[ederag]

No, since it's not mentioned in the cited stable docs (nor in the dev docs),
and searches did not show this.

For other users looking for the same error messages,
if the command is ran in the same session:

julia> Dagger.set_distributed_package!(DistributedNext)
ERROR: type `Module` is not a valid TOML type, pass a conversion function to `TOML.print`
...

One really has to issue the given command from the command line,
in the project directory, both on the local and the remote.

julia --project -e "using Dagger; Dagger.set_distributed_package!(\"DistributedNext\")"
[ Info: MemPool.jl preference has been set, restart your Julia session for this change to take effect!
[ Info: TimespanLogging.jl preference has been set, restart your Julia session for this change to take effect!
[ Info: Dagger.jl preference has been set, restart your Julia session for this change to take effect!

Thanks !