5 strategies (#6)

* Implement #5

* Bumped version to 0.5.0

CHANGELOG.md

@@ -1,10 +1,14 @@
 # Change Log
 
-## 0.4.0
+## 0.5.0 (29 February, 2024)
+
+- Fix `one_for_all` and `rest_for_all` strategies [#5](https://github.com/cardo-org/Visor.jl/issues/5)
+
+## 0.4.0 (1 February, 2024)
 
 - Update supervisor status at process termination [#3](https://github.com/cardo-org/Visor.jl/issues/3)
 
-## 0.3.0
+## 0.3.0 (24 January, 2024)
 
 - Setup supervisors settings with `setsupervisor` and `setroot` functions.
 

Project.toml

@@ -1,7 +1,7 @@
 name = "Visor"
 uuid = "cf786855-3531-4b86-ba6e-3e33dce7dcdb"
 authors = ["Attilio Donà"]
-version = "0.4.0"
+version = "0.5.0"
 
 [deps]
 DataStructures = "864edb3b-99cc-5e75-8d2d-829cb0a9cfe8"

src/Visor.jl

@@ -508,6 +508,18 @@ function restart_policy(supervisor, process)
     @debug "[$process] process retry startstamps: $(process.startstamps)"
     if length(process.startstamps) > supervisor.intensity
         @warn "[$process]: reached max of $(supervisor.intensity) restarts in $(supervisor.period) secs period"
+
+        # removing failed processes make possible to manually startup again
+        stopped = [process]
+        if supervisor.strategy === :one_for_all
+            stopped = supervisor_shutdown(supervisor)
+        elseif supervisor.strategy === :rest_for_one
+            stopped = supervisor_shutdown(supervisor, process)
+        end
+
+        for p in stopped
+            delete!(supervisor.processes, p.id)
+        end
         put!(supervisor.inbox, ProcessFatal(process))
     else
         process.isrestart = true
@@ -899,8 +911,10 @@ end
 function shutdown(sv::Supervisor, reset::Bool=true)
     @debug "[$sv] supervisor: shutdown request (reset=$reset)"
     if isdefined(sv, :task) && !istaskdone(sv.task)
-        put!(sv.inbox, Shutdown(; reset=reset))
-        close(sv.inbox)
+        if isopen(sv.inbox)
+            put!(sv.inbox, Shutdown(; reset=reset))
+            close(sv.inbox)
+        end
         wait(sv.task)
         sv.status = idle
     end

test/runtests.jl

@@ -35,9 +35,15 @@ logging(; debug=DEBUG == "0" ? [] : [Visor])
     @time @safetestset "chain_restart" begin
         include("test_chain_restart.jl")
     end
+    @time @safetestset "one_for_all_terminate" begin
+        include("test_one_for_all_terminate.jl")
+    end
     @time @safetestset "one_for_all" begin
         include("test_one_for_all.jl")
     end
+    @time @safetestset "rest_for_one_terminate" begin
+        include("test_rest_for_one_terminate.jl")
+    end
     @time @safetestset "rest_for_one" begin
         include("test_rest_for_one.jl")
     end
@@ -55,4 +61,4 @@ logging(; debug=DEBUG == "0" ? [] : [Visor])
     end
 end
 
-@info "expected tests: 43" 
+@info "expected tests: 49"

test/test_chain_restart.jl

@@ -33,3 +33,5 @@ p1_specs = [process("good", myworker), process("bad", faulty)]
 handle = supervise([supervisor("sv", p1_specs; intensity=2)])
 
 @test restarts === 2
+
+shutdown()

test/test_one_for_all_terminate.jl

@@ -0,0 +1,75 @@
+using Visor
+using Test
+
+#
+# 
+#       root
+#        /
+#       /  
+#      s1   
+#     /  \
+#   s11  w3
+#   /  \
+#  w1  w2
+#
+#  startup order: w1, w2, w3  
+#  if strategy is :one_for_all and the failed process does not restart terminate 
+#  all sibling processes.   
+#
+
+#ENV["JULIA_DEBUG"] = Visor
+
+starts = 0
+
+stopped = []
+
+function myworker(self)
+    global starts
+
+    @info "[$self]: starting"
+    starts += 1
+    if self.id === "w3"
+        sleep(2)
+        #if starts < 4
+        @info "[$self]: THROW EXCEPTION"
+        error("bang")
+        #end
+    else
+        sleep(20)
+    end
+    @info "[$self]: terminate"
+    #    for msg in self.inbox
+    #        @debug "[$(self.id)] recv: $msg"
+    #        if isshutdown(msg)
+    #            push!(stopped, self.id)
+    #            break
+    #        end
+    #    end
+end
+
+s11_specs = [
+    process("w1", myworker; thread=true), process("w2", myworker; force_interrupt_after=1)
+]
+
+s1_specs = [
+    supervisor("s11", s11_specs; intensity=1, terminateif=:shutdown),
+    process("w3", myworker; force_interrupt_after=1),
+]
+
+specs = [supervisor("s1", s1_specs; strategy=:one_for_all)]
+
+handle = Visor.supervise(specs; wait=false)
+
+timer_not_triggered = true
+timer = Timer((tim) -> begin
+    global timer_not_triggered = false
+    shutdown(handle)
+end, 10)
+
+@test wait(handle) === nothing
+
+# if all processes terminate the timer that shutdown thw system
+# is not triggered 
+@test timer_not_triggered
+@test starts === 6
+close(timer)

test/test_rest_for_one_terminate.jl

@@ -0,0 +1,66 @@
+include("utils.jl")
+
+NAME = "Alfred"
+
+EXPECTED_RESTARTS = 7
+restart_count = 0
+
+function p2(self)
+    global restart_count += 1
+    sleep(1)
+    return error("simulate a fault")
+end
+
+function myworker(self)
+    global restart_count
+
+    @info "(re)starting [$(self.id)]"
+    restart_count += 1
+
+    try
+        for msg in self.inbox
+            @info "[$(self.id)] recv: $msg"
+            if isshutdown(msg)
+                break
+            elseif msg.request === :get_name
+                put!(msg.inbox, NAME)
+            end
+        end
+    catch e
+        @error "worker: $e"
+        rethrow()
+    end
+end
+
+workerspecs = [
+    process("p1", myworker)
+    process("p2", p2)
+    process("p3", myworker)
+    process("p4", myworker)
+]
+
+specs = [supervisor("boss", workerspecs; strategy=:rest_for_one)]
+
+@info "starting ..."
+handle = Visor.supervise(specs; wait=false)
+
+only_p1_running = false
+timer = Timer((tim) -> begin
+
+    # check that only p1 is running
+    boss = from("boss")
+    procs = boss.processes
+    if (length(procs) === 1)
+        global only_p1_running = true
+    end
+    @info "boss running processes: $(boss.processes)"
+    shutdown(handle)
+end, 8)
+
+@test wait(handle) === nothing
+
+# if all processes terminate the timer that shutdown thw system
+# is not triggered 
+@test only_p1_running
+@test EXPECTED_RESTARTS === restart_count
+close(timer)