update

check_router_reboot.jl

@@ -19,14 +19,15 @@ const TIMEOUT_SECS = 30                # request timeout
 const ATTEMPTS_PER_CHECK = 1          # number of ping attempts per check
 const BACKOFF_BETWEEN_ATTEMPTS = 1    # seconds between ping attempts
 const FAILS_TO_REBOOT = 3             # consecutive failed checks required to trigger reboot
-const COOLDOWN_AFTER_REBOOT_SECS = 600  # do not reboot again within this many seconds
-const DRY_RUN = false                  # set false to actually reboot
+const COOLDOWN_AFTER_REBOOT_SECS = 120  # do not reboot again within this many seconds
+const DRY_RUN = true                  # set false to actually reboot
 const CHECK_INTERVAL_SECS = 60      # run a check every CHECK_INTERVAL_SECS seconds
 
 const thisFolderPath = @__DIR__
+const thisFilePath = @__FILE__
 const LogFilePath = "$thisFolderPath/check_router_reboot_log.txt"  # write logs here and also broadcast
 const StateFilePath = "$thisFolderPath/check_and_reboot_state.json"
-
+# println(0)
 # Simple broadcast helper
 # Simple broadcast helper (safe Cmd construction)
 function broadcast_msg(msg::AbstractString)
@@ -226,12 +227,12 @@ end
 
 # Single check iteration
 function perform_check!(st::State)
-
+  # println(1)
   success = false
   last_result = nothing
   for i in 1:ATTEMPTS_PER_CHECK
-      ok, result = check_router_once(ROUTER_IP) 
-      # ok, result = values(JSON.parsefile("test_ping_result.json"))  # for testing without actual ping
+      # ok, result = check_router_once(ROUTER_IP) 
+      ok, result = values(JSON.parsefile("/home/ton/docker-programs/check_and_reboot/test_ping_result.json"))  # for testing without actual ping
       if ok
           success = true
           break
@@ -244,13 +245,14 @@ function perform_check!(st::State)
   in_cooldown = false
   if st.last_reboot_datetime !== nothing
     timepass = ((Dates.now() - st.last_reboot_datetime).value / 1000) |> floor |> Int
-
     if timepass < COOLDOWN_AFTER_REBOOT_SECS
         in_cooldown = true
     end
   end
-
+  # @show in_cooldown
+  # println(2)
   if in_cooldown
+    # println("2-1")
     # During cooldown, track failures but don't trigger reboot yet
     if success
       broadcast_msg("Broadcasting from file: $thisFilePath")
@@ -258,6 +260,7 @@ function perform_check!(st::State)
       st.consecutive_fails = 0
       save_state(st, StateFilePath)
     else
+      # println("2-2")
       st.consecutive_fails += 1
       broadcast_msg("Broadcasting from file: $thisFilePath")
       logmsg("$ROUTER_IP is unreachable during cooldown. Consecutive fails: $(st.consecutive_fails)/$FAILS_TO_REBOOT.")
@@ -265,37 +268,44 @@ function perform_check!(st::State)
     end
     return
   end
-
+  # println(3)
   # Outside cooldown - full check with potential reboot
   if success
-      if st.consecutive_fails > 0
-          logmsg("$ROUTER_IP is reachable; resetting consecutive failure counter.")
-      else
-          # logmsg("$ROUTER_IP is reachable.")
-      end
-      st.consecutive_fails = 0
-      save_state(st, StateFilePath)
-      return
+    # println("3-1")
+    if st.consecutive_fails > 0
+      # println("3-2")
+        logmsg("$ROUTER_IP is reachable; resetting consecutive failure counter.")
+    else
+        # logmsg("$ROUTER_IP is reachable.")
+    end
+    st.consecutive_fails = 0
+    save_state(st, StateFilePath)
+    return
   else
-      st.consecutive_fails += 1
-      broadcast_msg("Broadcasting from file: $thisFilePath")
-      logmsg("$ROUTER_IP is unreachable (last result: $routerresult). Consecutive fails: $(st.consecutive_fails)/$FAILS_TO_REBOOT.")
-      save_state(st, StateFilePath)
+    st.consecutive_fails += 1
+    broadcast_msg("Broadcasting from file: $thisFilePath")
+    logmsg("$ROUTER_IP is unreachable (last result: $routerresult). Consecutive fails: $(st.consecutive_fails)/$FAILS_TO_REBOOT.")
+    save_state(st, StateFilePath)
   end
+  # println(4)
   if st.consecutive_fails >= FAILS_TO_REBOOT
+    # println("4-1")
+    save_state(st, StateFilePath)
+    ok = do_reboot()
+    if ok
+      # println("4-2")
+      broadcast_msg("Broadcasting from file: $thisFilePath")
+      logmsg("Reboot executed (or simulated). Resetting failure counter.")
+      st.consecutive_fails = 0
+      st.last_reboot_datetime = Dates.now()
+      @show st
+      @show StateFilePath
       save_state(st, StateFilePath)
-      ok = do_reboot()
-      if ok
-          thisFilePath = @__FILE__
-          broadcast_msg("Broadcasting from file: $thisFilePath")
-          logmsg("Reboot executed (or simulated). Resetting failure counter.")
-          st.consecutive_fails = 0
-          st.last_reboot_datetime = Dates.now()
-          save_state(st, StateFilePath)
-      else
-          logmsg("Reboot attempt failed; will retry after next interval.")
-      end
+    else
+      logmsg("Reboot attempt failed; will retry after next interval.")
+    end
   end
+  # println(5)
 end