SSmetrics - Elasticsearch powered metrics viewing

code/__DEFINES/MC.dm

@@ -66,12 +66,14 @@
 /datum/controller/subsystem/##X/New(){\
     NEW_SS_GLOBAL(SS##X);\
     PreInit();\
+	ss_id=#X;\
 }\
 /datum/controller/subsystem/##X
 
 #define PROCESSING_SUBSYSTEM_DEF(X) GLOBAL_REAL(SS##X, /datum/controller/subsystem/processing/##X);\
 /datum/controller/subsystem/processing/##X/New(){\
     NEW_SS_GLOBAL(SS##X);\
     PreInit();\
+	ss_id="processing_[#X]";\
 }\
 /datum/controller/subsystem/processing/##X

code/controllers/subsystem.dm

@@ -2,6 +2,8 @@
 /datum/controller/subsystem
 	// Metadata; you should define these.
 	name = "fire codertrain" //name of the subsystem
+	/// Subsystem ID. Used for when we need a technical name for the SS
+	var/ss_id = "fire_codertrain_again"
 	var/init_order = INIT_ORDER_DEFAULT		//order of initialization. Higher numbers are initialized first, lower numbers later. Use defines in __DEFINES/subsystems.dm for easy understanding of order.
 	var/wait = 20			//time to wait (in deciseconds) between each call to fire(). Must be a positive integer.
 	var/priority = FIRE_PRIORITY_DEFAULT	//When mutiple subsystems need to run in the same tick, higher priority subsystems will run first and be given a higher share of the tick before MC_TICK_CHECK triggers a sleep
@@ -227,3 +229,14 @@
 		if("queued_priority") //editing this breaks things.
 			return 0
 	. = ..()
+
+// Returns the metrics for the subsystem.
+// This can be overriden on subtypes for variables that could affect tick usage
+// Example: ATs on SSair
+/datum/controller/subsystem/proc/get_metrics()
+	SHOULD_CALL_PARENT(TRUE)
+	var/list/out = list()
+	out["cost"] = cost
+	out["tick_usage"] = tick_usage
+	out["custom"] = list() // Override as needed on child
+	return out

code/controllers/subsystem/air.dm

@@ -66,6 +66,11 @@ SUBSYSTEM_DEF(air)
 	msg += "AT/MS:[round((cost ? active_turfs.len/cost : 0),0.1)]"
 	..(msg)
 
+/datum/controller/subsystem/air/get_metrics()
+	. = ..()
+	var/list/cust = list()
+	cust["active_turfs"] = length(active_turfs)
+	.["custom"] = cust
 
 /datum/controller/subsystem/air/Initialize(timeofday)
 	setup_overlays() // Assign icons and such for gas-turf-overlays

code/controllers/subsystem/metrics.dm

@@ -0,0 +1,30 @@
+SUBSYSTEM_DEF(metrics)
+	name = "Metrics"
+	flags = SS_NO_INIT
+	wait = 1 MINUTES
+	offline_implications = "Server metrics will no longer be ingested into monitoring systems. No immediate action is needed."
+	runlevels = RUNLEVEL_LOBBY | RUNLEVEL_SETUP | RUNLEVEL_GAME | RUNLEVEL_POSTGAME // ALL THE LEVELS
+	/// The real time of day the server started. Used to calculate time drift
+	var/world_init_time = 0 // Not set in here. Set in world/New()
+
+/datum/controller/subsystem/metrics/fire(resumed)
+	// AA TODO
+	return
+
+/datum/controller/subsystem/metrics/proc/get_metrics_json()
+	var/list/out = list()
+	out["cpu"] = world.cpu
+	// out["maptick"] = world.map_cpu // TODO: 514
+	out["elapsed_processed"] = world.time
+	out["elapsed_real"] = (REALTIMEOFDAY - world_init_time)
+	out["client_count"] = length(GLOB.clients)
+	out["round_id"] = text2num(GLOB.round_id) // This is so we can filter the metrics by a single round ID
+
+	// Funnel in all SS metrics
+	var/list/ss_data = list()
+	for(var/datum/controller/subsystem/SS in Master.subsystems)
+		ss_data[SS.ss_id] = SS.get_metrics()
+
+	out["subsystems"] = ss_data
+	// And send it all
+	return json_encode(out)

code/controllers/subsystem/mobs.dm

@@ -12,6 +12,12 @@ SUBSYSTEM_DEF(mobs)
 	/// The amount of giant spiders that exist in the world. Used for mob capping.
 	var/giant_spiders = 0
 
+/datum/controller/subsystem/mobs/get_metrics()
+	. = ..()
+	var/list/cust = list()
+	cust["processing"] = length(GLOB.mob_living_list)
+	.["custom"] = cust
+
 /datum/controller/subsystem/mobs/stat_entry()
 	..("P:[GLOB.mob_living_list.len]")
 

code/controllers/subsystem/processing/processing.dm

@@ -14,6 +14,12 @@ SUBSYSTEM_DEF(processing)
 /datum/controller/subsystem/processing/stat_entry()
 	..("[stat_tag]:[processing.len]")
 
+/datum/controller/subsystem/processing/get_metrics()
+	. = ..()
+	var/list/cust = list()
+	cust["processing"] = length(processing)
+	.["custom"] = cust
+
 /datum/controller/subsystem/processing/fire(resumed = 0)
 	if(!resumed)
 		currentrun = processing.Copy()

code/controllers/subsystem/tickets/tickets.dm

@@ -41,6 +41,12 @@ SUBSYSTEM_DEF(tickets)
 
 	var/ticketCounter = 1
 
+/datum/controller/subsystem/tickets/get_metrics()
+	. = ..()
+	var/list/cust = list()
+	cust["tickets"] = length(allTickets) // Not a perf metric but I want to see a graph where SSair usage spikes and 20 tickets come in
+	.["custom"] = cust
+
 /datum/controller/subsystem/tickets/Initialize()
 	if(!close_messages)
 		close_messages = list("<font color='red' size='4'><b>- [ticket_name] Rejected! -</b></font>",

code/controllers/subsystem/timer.dm

@@ -37,6 +37,12 @@ SUBSYSTEM_DEF(timer)
 /datum/controller/subsystem/timer/stat_entry(msg)
 	..("B:[bucket_count] P:[length(second_queue)] H:[length(hashes)] C:[length(clienttime_timers)] S:[length(timer_id_dict)]")
 
+/datum/controller/subsystem/timer/get_metrics()
+	. = ..()
+	var/list/cust = list()
+	cust["bucket_count"] = bucket_count
+	.["custom"] = cust
+
 /datum/controller/subsystem/timer/fire(resumed = FALSE)
 	var/lit = last_invoke_tick
 	var/last_check = world.time - TICKS2DS(BUCKET_LEN*1.5)

code/game/world.dm

@@ -8,6 +8,8 @@ GLOBAL_LIST_INIT(map_transition_config, list(CC_TRANSITION_CONFIG))
 	// Right off the bat
 	enable_auxtools_debugger()
 
+	SSmetrics.world_init_time = REALTIMEOFDAY
+
 	// Do sanity checks to ensure RUST actually exists
 	if(!fexists(RUST_G))
 		DIRECT_OUTPUT(world.log, "ERROR: RUSTG was not found and is required for the game to function. Server will now exit.")

code/modules/unit_tests/_unit_tests.dm

@@ -12,6 +12,7 @@
 #include "spawn_humans.dm"
 #include "sql.dm"
 #include "subsystem_init.dm"
+#include "subsystem_metric_sanity.dm"
 #include "timer_sanity.dm"
 #include "unit_test.dm"
 #endif

code/modules/unit_tests/subsystem_metric_sanity.dm

@@ -0,0 +1,19 @@
+// Unit test to ensure SS metrics are valid
+/datum/unit_test/subsystem_metric_sanity/Run()
+	for(var/datum/controller/subsystem/SS in Master.subsystems)
+		var/list/data = SS.get_metrics()
+		if(length(data) != 3)
+			Fail("SS[SS.ss_id] has invalid metrics data!")
+			continue
+		if(isnull(data["cost"]))
+			Fail("SS[SS.ss_id] has invalid metrics data! No 'cost' found in [json_encode(data)]")
+			continue
+		if(isnull(data["tick_usage"]))
+			Fail("SS[SS.ss_id] has invalid metrics data! No 'tick_usage' found in [json_encode(data)]")
+			continue
+		if(isnull(data["custom"]))
+			Fail("SS[SS.ss_id] has invalid metrics data! No 'custom' found in [json_encode(data)]")
+			continue
+		if(!islist(data["custom"]))
+			Fail("SS[SS.ss_id] has invalid metrics data! 'custom' is not a list in [json_encode(data)]")
+			continue

paradise.dme

@@ -235,6 +235,7 @@
 #include "code\controllers\subsystem\machinery.dm"
 #include "code\controllers\subsystem\mapping.dm"
 #include "code\controllers\subsystem\medals.dm"
+#include "code\controllers\subsystem\metrics.dm"
 #include "code\controllers\subsystem\mobs.dm"
 #include "code\controllers\subsystem\nano_mob_hunter.dm"
 #include "code\controllers\subsystem\nightshift.dm"