Can't do multiple mapCube computations in the same script on SLURM

When I am trying to do multiple mapCube runs on the same data on a SLURM cluster I am running into this error directly after the first run

 Error: Fatal error on process 1
│   exception =
│    IOError: stream is closed or unusable
│    Stacktrace:
│      [1] check_open
│        @ ./stream.jl:386 [inlined]
│      [2] uv_write_async(s::Sockets.TCPSocket, p::Ptr{UInt8}, n::UInt64)
│        @ Base ./stream.jl:1018
│      [3] uv_write(s::Sockets.TCPSocket, p::Ptr{UInt8}, n::UInt64)
│        @ Base ./stream.jl:981
│      [4] uv_write
│        @ ./stream.jl:977 [inlined]
│      [5] flush(s::Sockets.TCPSocket)
│        @ Base ./stream.jl:1073
│      [6] send_msg_(w::Distributed.Worker, header::Distributed.MsgHeader, msg::Distributed.ResultMsg, now::Bool)
│        @ Distributed /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.6/Distributed/src/messages.jl:180
│      [7] send_msg_now
│        @ /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.6/Distributed/src/messages.jl:118 [inlined]
│      [8] send_msg_now(s::Sockets.TCPSocket, header::Distributed.MsgHeader, msg::Distributed.ResultMsg)
│        @ Distributed /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.6/Distributed/src/messages.jl:113
│      [9] deliver_result(sock::Sockets.TCPSocket, msg::Symbol, oid::Distributed.RRID, value::RemoteException)
│        @ Distributed /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.6/Distributed/src/process_messages.jl:95
│     [10] macro expansion
│        @ /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.6/Distributed/src/process_messages.jl:286 [inlined]
│     [11] (::Distributed.var"#105#107"{Distributed.CallMsg{:call_fetch}, Distributed.MsgHeader, Sockets.TCPSocket})()
│        @ Distributed ./task.jl:406
└ @ Distributed /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.6/Distributed/src/process_messages.jl:99

This seems to be https://github.com/JuliaLang/julia/issues/35442 but maybe we can work around this somehow.

Reproduce multiple mapCube runs on the same data under SLURM and start by tracing the worker and connection lifecycle around the mapCube entry point. Compare the behavior with Julia issue 35442; done means sequential mapCube computations complete without the closed-stream error.