Add warning about new workers not sharing prior global state (JuliaLa…

…ng#14)

Originally proposed in JuliaLang/julia#50843

src/cluster.jl

@@ -92,7 +92,7 @@ mutable struct WorkerConfig
     end
 end
 
-@enum WorkerState W_CREATED W_CONNECTED W_TERMINATING W_TERMINATED
+@enum WorkerState W_CREATED W_CONNECTED W_TERMINATING W_TERMINATED W_UNKNOWN_STATE
 mutable struct Worker
     id::Int
     msg_lock::Threads.ReentrantLock # Lock for del_msgs, add_msgs, and gcflag

src/managers.jl

@@ -450,6 +450,11 @@ Launch `np` workers on the local host using the in-built `LocalManager`.
 Local workers inherit the current package environment (i.e., active project,
 [`LOAD_PATH`](@ref), and [`DEPOT_PATH`](@ref)) from the main process.
 
+!!! warning
+    Note that workers do not run a `~/.julia/config/startup.jl` startup script, nor do they synchronize
+    their global state (such as command-line switches, global variables, new method definitions, and loaded modules) with any
+    of the other running processes.
+
 **Keyword arguments**:
  - `restrict::Bool`: if `true` (default) binding is restricted to `127.0.0.1`.
  - `dir`, `exename`, `exeflags`, `env`, `topology`, `lazy`, `enable_threaded_blas`: same effect

src/process_messages.jl

@@ -210,8 +210,7 @@ function message_handler_loop(r_stream::IO, w_stream::IO, incoming::Bool)
             handle_msg(msg, header, r_stream, w_stream, version)
         end
     catch e
-        werr = worker_from_id(wpid)
-        oldstate = werr.state
+        oldstate = W_UNKNOWN_STATE
 
         # Check again as it may have been set in a message handler but not propagated to the calling block above
         if wpid < 1
@@ -222,6 +221,8 @@ function message_handler_loop(r_stream::IO, w_stream::IO, incoming::Bool)
             println(stderr, e, CapturedException(e, catch_backtrace()))
             println(stderr, "Process($(myid())) - Unknown remote, closing connection.")
         elseif !(wpid in map_del_wrkr)
+            werr = worker_from_id(wpid)
+            oldstate = werr.state    
             set_worker_state(werr, W_TERMINATED)
 
             # If unhandleable error occurred talking to pid 1, exit