sig profile hanging workers before SIGKILL

Author: IanButterworth

src/managers.jl

@@ -126,7 +126,7 @@ addprocs([
 
 * `exeflags`: additional flags passed to the worker processes. It can either be a `Cmd`, a `String`
   holding one flag, or a collection of strings, with one element per flag.
-  E.g. `\`--threads=auto project=.\``, `"--compile-trace=stderr"` or `["--threads=auto", "--compile=all"]`. 
+  E.g. `\`--threads=auto project=.\``, `"--compile-trace=stderr"` or `["--threads=auto", "--compile=all"]`.
 
 * `topology`: Specifies how the workers connect to each other. Sending a message between
   unconnected workers results in an error.
@@ -767,7 +767,8 @@ function kill(manager::SSHManager, pid::Int, config::WorkerConfig)
     nothing
 end
 
-function kill(manager::LocalManager, pid::Int, config::WorkerConfig; exit_timeout = 15, term_timeout = 15)
+function kill(manager::LocalManager, pid::Int, config::WorkerConfig; profile_wait = 6, exit_timeout = 15, term_timeout = 15)
+    # profile_wait = 6 is 1s for profile, 5s for the report to show
     # First, try sending `exit()` to the remote over the usual control channels
     remote_do(exit, pid)
 
@@ -776,7 +777,14 @@ function kill(manager::LocalManager, pid::Int, config::WorkerConfig; exit_timeou
 
         # Check to see if our child exited, and if not, send an actual kill signal
         if !process_exited(config.process)
-            @warn("Failed to gracefully kill worker $(pid), sending SIGQUIT")
+            @warn "Failed to gracefully kill worker $(pid)"
+            profile_sig = Sys.iswindows() ? nothing : Sys.isbsd() ? ("SIGINFO", 29) : ("SIGUSR1" , 10)
+            if profile_sig !== nothing
+                @warn("Sending profile $(profile_sig[1]) to worker $(pid)")
+                kill(config.process, profile_sig[2])
+                sleep(profile_wait)
+            end
+            @warn("Sending SIGQUIT to worker $(pid)")
             kill(config.process, Base.SIGQUIT)
 
             sleep(term_timeout)

test/distributed_exec.jl

@@ -1934,7 +1934,7 @@ include("splitrange.jl")
 
         # Next, ensure we get a log message when a worker does not cleanly exit
         w = only(addprocs(1))
-        @test_logs (:warn, r"sending SIGQUIT") begin
+        @test_logs (:warn, r"Sending SIGQUIT") match_mode=:any begin
             remote_do(w) do
                 # Cause the 'exit()' message that `rmprocs()` sends to do nothing
                 Core.eval(Base, :(exit() = nothing))