Failed to precompile `DistributedNext`

[lkdvos]

Currenntly, on a fresh install of DistributedNext in a temporary environment, I get the following error during precompilation:

PkgPrecompileError: The following 1 direct dependency failed to precompile:

DistributedNext

Failed to precompile DistributedNext [fab6aee4-877b-4bac-a744-3eca44acbb6f] to "/Users/ldevos/.julia/compiled/v1.12/DistributedNext/jl_AwqKvj".
Worker 2 terminated.
BaseUnhandled Task ERROR: IOError: read: connection reset by peer (ECONNRESET)
Stacktrace:
 [1] wait_readnb(x::Sockets.TCPSocket, nb::Int64)
   @ Base ./stream.jl:410
 [2] (::Base.var"#wait_locked#readbytes!##0")(s::Sockets.TCPSocket, buf::IOBuffer, nb::Int64)
   @ Base ./stream.jl:934
 [3] readbytes!(s::Sockets.TCPSocket, a::Vector{UInt8}, nb::Int64)
   @ Base ./stream.jl:940
 [4] read
   @ ./io.jl:1177 [inlined]
 [5] process_hdr(s::Sockets.TCPSocket, validate_cookie::Bool)
   @ DistributedNext ~/.julia/packages/DistributedNext/3cHVQ/src/process_messages.jl:276
 [6] message_handler_loop(r_stream::Sockets.TCPSocket, w_stream::Sockets.TCPSocket, incoming::Bool)
   @ DistributedNext ~/.julia/packages/DistributedNext/3cHVQ/src/process_messages.jl:159
 [7] process_tcp_streams(r_stream::Sockets.TCPSocket, w_stream::Sockets.TCPSocket, incoming::Bool)
   @ DistributedNext ~/.julia/packages/DistributedNext/3cHVQ/src/process_messages.jl:134
 [8] (::DistributedNext.var"#process_messages##0#process_messages##1"{Sockets.TCPSocket, Sockets.TCPSocket, Bool})()
   @ DistributedNext ~/.julia/packages/DistributedNext/3cHVQ/src/process_messages.jl:122
.IOError("read: socket is not connected (ENOTCONN)", -57)CapturedException(Base.IOError("read: socket is not connected (ENOTCONN)", -57), Any[(wait_readnb(x::Sockets.TCPSocket, nb::Int64) at stream.jl:410, 1), ((::Base.var"#wait_locked#readbytes!##0")(s::Sockets.TCPSocket, buf::IOBuffer, nb::Int64) at stream.jl:934, 1), (readbytes!(s::Sockets.TCPSocket, a::Vector{UInt8}, nb::Int64) at stream.jl:940, 1), (read at io.jl:1177 [inlined], 1), (process_hdr(s::Sockets.TCPSocket, validate_cookie::Bool) at process_messages.jl:260, 1), (message_handler_loop(r_stream::Sockets.TCPSocket, w_stream::Sockets.TCPSocket, incoming::Bool) at process_messages.jl:159, 1), (process_tcp_streams(r_stream::Sockets.TCPSocket, w_stream::Sockets.TCPSocket, incoming::Bool) at process_messages.jl:134, 1), ((::DistributedNext.var"#process_messages##0#process_messages##1"{Sockets.TCPSocket, Sockets.TCPSocket, Bool})() at process_messages.jl:122, 1)])
Process(1) - Unknown remote, closing connection.
Master process (id 1) could not connect within 60.0 seconds.
exiting

Version info

(jl_2J0WQF) pkg> st
Status `/private/var/folders/8x/gsbvn1cs0nx_014zjk3twt3m0000gp/T/jl_2J0WQF/Project.toml`
  [fab6aee4] DistributedNext v1.3.1

julia> versioninfo()
Julia Version 1.12.6
Commit 15346901f00 (2026-04-09 19:20 UTC)
Build Info:
  Official https://julialang.org release
Platform Info:
  OS: macOS (arm64-apple-darwin24.0.0)
  CPU: 16 × Apple M3 Max
  WORD_SIZE: 64
  LLVM: libLLVM-18.1.7 (ORCJIT, apple-m3)
  GC: Built with stock GC
Threads: 1 default, 1 interactive, 1 GC (on 12 virtual cores)

@jack-dunham

[JamesWrigley]

Just to confirm, is 127.0.0.1 a working interface on your machine? That's the interface the precompile workload will bind to.

[giordano]

Maybe the precompilation workload should be robust to this situation (run in a try block?)

[JamesWrigley]

Actually it could just be a regular race condition, the in-process worker is started in a separate task and it looks like we don't wait for the socket to be bound before telling addprocs() that the worker has been launched:

DistributedNext.jl/src/managers.jl

Line 542 in 50d15ab

start_worker(pipe.in, cookie; close_stdin=false, stderr_to_stdout=false, exit_on_close=false)

Same kind of race as #5 (comment).
(I'm going out soon so probably won't fix it today, but will have a look tonight/tomorrow unless someone else gets to it first)

[lkdvos]

Just to confirm, is 127.0.0.1 a working interface on your machine? That's the interface the precompile workload will bind to.

ping 127.0.0.1                                                                                                                                                                                                           
PING 127.0.0.1 (127.0.0.1): 56 data bytes
64 bytes from 127.0.0.1: icmp_seq=0 ttl=64 time=1.300 ms
64 bytes from 127.0.0.1: icmp_seq=1 ttl=64 time=0.720 ms
64 bytes from 127.0.0.1: icmp_seq=2 ttl=64 time=3.017 ms
64 bytes from 127.0.0.1: icmp_seq=3 ttl=64 time=0.927 ms
64 bytes from 127.0.0.1: icmp_seq=4 ttl=64 time=0.843 ms

I think so?

[JamesWrigley]

Yeah that's fine, it's probably the race condition. I just asked because once I got an issue from someone using ZMQ.jl who had localhost bound to a remote host on their tailscale network and that was breaking the precompilation workload 😛

[JamesWrigley]

Could you try with #74?