From 89e7a06deb6d353f6e4d46ebfcd919a06f6e933d Mon Sep 17 00:00:00 2001
From: Cloud Wu <cloudwu@gmail.com>
Date: Thu, 13 Sep 2018 18:27:05 +0800
Subject: [PATCH] add more info when service init blocked

---
 lualib/skynet.lua       | 40 +++++++++++++++++++++++++++++++---------
 service/launcher.lua    | 14 ++++++++++++++
 service/service_mgr.lua | 28 ++++++++++++++++++++++------
 3 files changed, 67 insertions(+), 15 deletions(-)

diff --git a/lualib/skynet.lua b/lualib/skynet.lua
index 68aa5d9c..789cb77c 100644
--- a/lualib/skynet.lua
+++ b/lualib/skynet.lua
@@ -12,6 +12,7 @@ local profile = require "skynet.profile"
 
 local cresume = profile.resume
 local running_thread = nil
+local init_thread = nil
 
 local function coroutine_resume(co, ...)
 	running_thread = co
@@ -219,6 +220,7 @@ function skynet.timeout(ti, func)
 	local co = co_create(func)
 	assert(session_id_coroutine[session] == nil)
 	session_id_coroutine[session] = co
+	return co	-- for debug
 end
 
 local function suspend_sleep(session, token)
@@ -770,8 +772,9 @@ end
 
 function skynet.start(start_func)
 	c.callback(skynet.dispatch_message)
-	skynet.timeout(0, function()
+	init_thread = skynet.timeout(0, function()
 		skynet.init_service(start_func)
+		init_thread = nil
 	end)
 end
 
@@ -788,22 +791,41 @@ function skynet.stat(what)
 end
 
 function skynet.task(ret)
-	if type(ret) == "number" then
+	if ret == nil then
+		local t = 0
+		for session,co in pairs(session_id_coroutine) do
+			t = t + 1
+		end
+		return t
+	end
+	if ret == "init" then
+		if init_thread then
+			return debug.traceback(init_thread)
+		else
+			return
+		end
+	end
+	local tt = type(ret)
+	if tt == "table" then
+		for session,co in pairs(session_id_coroutine) do
+			ret[session] = debug.traceback(co)
+		end
+		return
+	elseif tt == "number" then
 		local co = session_id_coroutine[ret]
 		if co then
 			return debug.traceback(co)
 		else
 			return "No session"
 		end
-	end
-	local t = 0
-	for session,co in pairs(session_id_coroutine) do
-		if ret then
-			ret[session] = debug.traceback(co)
+	elseif tt == "thread" then
+		for session, co in pairs(session_id_coroutine) do
+			if co == ret then
+				return session
+			end
 		end
-		t = t + 1
+		return
 	end
-	return t
 end
 
 function skynet.term(service)
diff --git a/service/launcher.lua b/service/launcher.lua
index 272a459a..d458d13e 100644
--- a/service/launcher.lua
+++ b/service/launcher.lua
@@ -6,6 +6,7 @@ local string = string
 local services = {}
 local command = {}
 local instance = {} -- for confirm (function command.LAUNCH / command.ERROR / command.LAUNCHOK)
+local launch_session = {} -- for command.QUERY, service_address -> session
 
 local function handle_to_address(handle)
 	return tonumber("0x" .. string.sub(handle , 2))
@@ -68,6 +69,7 @@ function command.REMOVE(_, handle, kill)
 		-- instance is dead
 		response(not kill)	-- return nil to caller of newservice, when kill == false
 		instance[handle] = nil
+		launch_session[handle] = nil
 	end
 
 	-- don't return (skynet.ret) because the handle may exit
@@ -77,10 +79,12 @@ end
 local function launch_service(service, ...)
 	local param = table.concat({...}, " ")
 	local inst = skynet.launch(service, param)
+	local session = skynet.context()
 	local response = skynet.response()
 	if inst then
 		services[inst] = service .. " " .. param
 		instance[inst] = response
+		launch_session[inst] = session
 	else
 		response(false)
 		return
@@ -107,6 +111,7 @@ function command.ERROR(address)
 	local response = instance[address]
 	if response then
 		response(false)
+		launch_session[address] = nil
 		instance[address] = nil
 	end
 	services[address] = nil
@@ -119,11 +124,20 @@ function command.LAUNCHOK(address)
 	if response then
 		response(true, address)
 		instance[address] = nil
+		launch_session[address] = nil
 	end
 
 	return NORET
 end
 
+function command.QUERY(_, request_session)
+	for address, session in pairs(launch_session) do
+		if session == request_session then
+			return address
+		end
+	end
+end
+
 -- for historical reasons, launcher support text command (for C service)
 
 skynet.register_protocol {
diff --git a/service/service_mgr.lua b/service/service_mgr.lua
index e1efeb41..c4d6273c 100644
--- a/service/service_mgr.lua
+++ b/service/service_mgr.lua
@@ -42,12 +42,17 @@ local function waitfor(name , func, ...)
 
 	assert(type(s) == "table")
 
-	if not s.launch and func then
-		s.launch = true
+	local session, source = skynet.context()
+
+	if s.launch == nil and func then
+		s.launch = {
+			session = session,
+			source = source,
+			co = co,
+		}
 		return request(name, func, ...)
 	end
 
-	local session, source = skynet.context()
 	table.insert(s, {
 		co = co,
 		session = session,
@@ -96,9 +101,20 @@ local function list_service()
 		if type(v) == "string" then
 			v = "Error: " .. v
 		elseif type(v) == "table" then
-			local querying = { "Querying:" }
-			for _, detail in ipairs(v) do
-				table.insert(querying, skynet.address(detail.source) .. " " .. tostring(skynet.call(detail.source, "debug", "TASK", detail.session)))
+			local querying = {}
+			if v.launch then
+				local session = skynet.task(v.launch.co)
+				local launching_address = skynet.call(".launcher", "lua", "QUERY", session)
+				table.insert(querying, "Init as " .. skynet.address(launching_address))
+				table.insert(querying,  tostring(skynet.call(launching_address, "debug", "TASK", "init")))
+				table.insert(querying, "Launching from " .. skynet.address(v.launch.source))
+				table.insert(querying, tostring(skynet.call(v.launch.source, "debug", "TASK", v.launch.session)))
+			end
+			if #v > 0 then
+				table.insert(querying , "Querying:" )
+				for _, detail in ipairs(v) do
+					table.insert(querying, skynet.address(detail.source) .. " " .. tostring(skynet.call(detail.source, "debug", "TASK", detail.session)))
+				end
 			end
 			v = table.concat(querying, "\n")
 		else