mirror of
https://github.com/ParadiseSS13/Paradise.git
synced 2026-07-16 17:43:35 +01:00
Further process tweaks
- Integrates Volundr's btime library and associated process scheduler changes. - btime is implemented separately from the process scheduler, as precise time measurement is also useful elsewhere. - `TimeOfHour` is no longer internally throttled; throttling is instead done by `SCHECK`. - If btime's `gettime` cannot be called at world startup, an error will be output and the world will stop. - Retains the change to schedule processes according to game time, rather than real time. - Removes the (now unused) update queue files. - Removes the process scheduler testing files. - These are standalone tests for the process scheduler, completely unrelated to its use in the full codebase. We never used them. - Moves the process scheduler defines into __DEFINES. - Makes the lighting process run once before the round starts. - Renames `scheck` to `sleepCheck`, to ensure any code that tries to use `scheck` will fail to compile. - Adds `SCHECK` and `SCHECK_EVERY` macros that skip calling `sleepCheck` entirely until a specified number of `SCHECK`s (50 by default) have been called. - Makes most processes iterate using their `last_object` variable, allowing hang recovery to show the type that caused the hang. - Makes processes output an error when they filter out a type they refuse to process. - Rolls the recently-added alarm subsystem into the alarm process. - Removes the now unused subsystems code.
This commit is contained in:
Krausus
@@ -48,7 +48,7 @@
|
||||
// This controls how often the process will yield (call sleep(0)) while it is running.
|
||||
// Every concurrent process should sleep periodically while running in order to allow other
|
||||
// processes to execute concurrently.
|
||||
var/tmp/sleep_interval = PROCESS_DEFAULT_SLEEP_INTERVAL
|
||||
var/tmp/sleep_interval
|
||||
|
||||
// hang_warning_time - this is the time (in 1/10 seconds) after which the server will begin to show "maybe hung" in the context window
|
||||
var/tmp/hang_warning_time = PROCESS_DEFAULT_HANG_WARNING_TIME
|
||||
@@ -59,20 +59,20 @@
|
||||
// hang_restart_time - After this much time(in 1/10 seconds), the server will automatically kill and restart the process.
|
||||
var/tmp/hang_restart_time = PROCESS_DEFAULT_HANG_RESTART_TIME
|
||||
|
||||
// cpu_threshold - if world.cpu >= cpu_threshold, scheck() will call sleep(1) to defer further work until the next tick. This keeps a process from driving a tick into overtime (causing perceptible lag)
|
||||
var/tmp/cpu_threshold = PROCESS_DEFAULT_CPU_THRESHOLD
|
||||
|
||||
// How many times in the current run has the process deferred work till the next tick?
|
||||
var/tmp/cpu_defer_count = 0
|
||||
|
||||
// How many SCHECKs have been skipped (to limit btime calls)
|
||||
var/tmp/calls_since_last_scheck = 0
|
||||
|
||||
/**
|
||||
* recordkeeping vars
|
||||
*/
|
||||
|
||||
// Records the time (server ticks) at which the process last finished sleeping
|
||||
// Records the time (1/10s timeofday) at which the process last finished sleeping
|
||||
var/tmp/last_slept = 0
|
||||
|
||||
// Records the time (s-ticks) at which the process last began running
|
||||
// Records the time (1/10s timeofday) at which the process last began running
|
||||
var/tmp/run_start = 0
|
||||
|
||||
// Records the number of times this process has been killed and restarted
|
||||
@@ -98,7 +98,7 @@
|
||||
idle()
|
||||
name = "process"
|
||||
schedule_interval = 50
|
||||
sleep_interval = 2
|
||||
sleep_interval = world.tick_lag / PROCESS_DEFAULT_SLEEP_INTERVAL
|
||||
last_slept = 0
|
||||
run_start = 0
|
||||
ticks = 0
|
||||
@@ -106,11 +106,12 @@
|
||||
last_object = null
|
||||
|
||||
/datum/controller/process/proc/started()
|
||||
var/timeofhour = TimeOfHour
|
||||
// Initialize last_slept so we can know when to sleep
|
||||
last_slept = world.timeofday
|
||||
last_slept = timeofhour
|
||||
|
||||
// Initialize run_start so we can detect hung processes.
|
||||
run_start = world.timeofday
|
||||
run_start = timeofhour
|
||||
|
||||
// Initialize defer count
|
||||
cpu_defer_count = 0
|
||||
@@ -162,16 +163,16 @@
|
||||
setStatus(PROCESS_STATUS_HUNG)
|
||||
|
||||
/datum/controller/process/proc/handleHung()
|
||||
var/timeofhour = TimeOfHour
|
||||
var/datum/lastObj = last_object
|
||||
var/lastObjType = "null"
|
||||
if(istype(lastObj))
|
||||
lastObjType = lastObj.type
|
||||
|
||||
// If world.timeofday has rolled over, then we need to adjust.
|
||||
if (world.timeofday < run_start)
|
||||
run_start -= 864000
|
||||
|
||||
var/msg = "[name] process hung at tick #[ticks]. Process was unresponsive for [(world.timeofday - run_start) / 10] seconds and was restarted. Last task: [last_task]. Last Object Type: [lastObjType]"
|
||||
// If timeofhour has rolled over, then we need to adjust.
|
||||
if (timeofhour < run_start)
|
||||
run_start -= 36000
|
||||
var/msg = "[name] process hung at tick #[ticks]. Process was unresponsive for [(timeofhour - run_start) / 10] seconds and was restarted. Last task: [last_task]. Last Object Type: [lastObjType]"
|
||||
logTheThing("debug", null, null, msg)
|
||||
logTheThing("diary", null, null, msg, "debug")
|
||||
message_admins(msg)
|
||||
@@ -188,33 +189,36 @@
|
||||
// Allow inheritors to clean up if needed
|
||||
onKill()
|
||||
|
||||
killed = TRUE
|
||||
|
||||
// This should del
|
||||
del(src)
|
||||
|
||||
/datum/controller/process/proc/scheck(var/tickId = 0)
|
||||
// Do not call this directly - use SHECK or SCHECK_EVERY
|
||||
/datum/controller/process/proc/sleepCheck(var/tickId = 0)
|
||||
calls_since_last_scheck = 0
|
||||
if (killed)
|
||||
// The kill proc is the only place where killed is set.
|
||||
// The kill proc should have deleted this datum, and all sleeping procs that are
|
||||
// owned by it.
|
||||
CRASH("A killed process is still running somehow...")
|
||||
if (hung)
|
||||
// This will only really help if the doWork proc ends up in an infinite loop.
|
||||
handleHung()
|
||||
CRASH("Process [name] hung and was restarted.")
|
||||
|
||||
// For each tick the process defers, it increments the cpu_defer_count so we don't
|
||||
// defer indefinitely
|
||||
if (world.cpu >= (cpu_threshold + cpu_defer_count * world.tick_lag * 10))
|
||||
sleep(world.tick_lag)
|
||||
if (main.getCurrentTickElapsedTime() > main.timeAllowance)
|
||||
sleep(world.tick_lag*1)
|
||||
cpu_defer_count++
|
||||
last_slept = world.timeofday
|
||||
last_slept = TimeOfHour
|
||||
else
|
||||
// If world.timeofday has rolled over, then we need to adjust.
|
||||
if (world.timeofday < last_slept)
|
||||
last_slept -= 864000
|
||||
var/timeofhour = TimeOfHour
|
||||
// If timeofhour has rolled over, then we need to adjust.
|
||||
if (timeofhour < last_slept)
|
||||
last_slept -= 36000
|
||||
|
||||
if (world.timeofday > last_slept + sleep_interval)
|
||||
// If we haven't slept in sleep_interval ticks, sleep to allow other work to proceed.
|
||||
if (timeofhour > last_slept + sleep_interval)
|
||||
// If we haven't slept in sleep_interval deciseconds, sleep to allow other work to proceed.
|
||||
sleep(0)
|
||||
last_slept = world.timeofday
|
||||
last_slept = TimeOfHour
|
||||
|
||||
/datum/controller/process/proc/update()
|
||||
// Clear delta
|
||||
@@ -223,17 +227,22 @@
|
||||
|
||||
var/elapsedTime = getElapsedTime()
|
||||
|
||||
if (elapsedTime > hang_restart_time)
|
||||
if (hung)
|
||||
handleHung()
|
||||
return
|
||||
else if (elapsedTime > hang_restart_time)
|
||||
hung()
|
||||
else if (elapsedTime > hang_alert_time)
|
||||
setStatus(PROCESS_STATUS_PROBABLY_HUNG)
|
||||
else if (elapsedTime > hang_warning_time)
|
||||
setStatus(PROCESS_STATUS_MAYBE_HUNG)
|
||||
|
||||
|
||||
/datum/controller/process/proc/getElapsedTime()
|
||||
if (world.timeofday < run_start)
|
||||
return world.timeofday - (run_start - 864000)
|
||||
return world.timeofday - run_start
|
||||
var/timeofhour = TimeOfHour
|
||||
if (timeofhour < run_start)
|
||||
return timeofhour - (run_start - 36000)
|
||||
return timeofhour - run_start
|
||||
|
||||
/datum/controller/process/proc/tickDetail()
|
||||
return
|
||||
@@ -316,6 +325,8 @@
|
||||
/datum/controller/process/proc/enable()
|
||||
disabled = 0
|
||||
|
||||
/datum/controller/process/proc/getAverageRunTime()
|
||||
return main.averageRunTime(src)
|
||||
/datum/controller/process/proc/getLastRunTime()
|
||||
return main.getProcessLastRunTime(src)
|
||||
|
||||
@@ -326,7 +337,10 @@
|
||||
return ticks
|
||||
|
||||
/datum/controller/process/proc/statProcess()
|
||||
stat("[name]", "T#[getTicks()]|LR [getLastRunTime()]|HR [getHighestRunTime()]|D [cpu_defer_count]")
|
||||
var/averageRunTime = round(getAverageRunTime(), 0.1)/10
|
||||
var/lastRunTime = round(getLastRunTime(), 0.1)/10
|
||||
var/highestRunTime = round(getHighestRunTime(), 0.1)/10
|
||||
stat("[name]", "T#[getTicks()]|AR [averageRunTime]|LR [lastRunTime]|HR [highestRunTime]|D [cpu_defer_count]")
|
||||
|
||||
/datum/controller/process/proc/catchException(var/exception/e, var/thrower)
|
||||
var/etext = "[e]"
|
||||
@@ -351,3 +365,8 @@
|
||||
world.log << "This exception will now be ignored for ten minutes."
|
||||
spawn(6000)
|
||||
exceptions[eid] = 0
|
||||
|
||||
/datum/controller/process/proc/catchBadType(var/datum/caught)
|
||||
if(isnull(caught) || !istype(caught) || !isnull(caught.gcDestroyed))
|
||||
return // Only bother with types we can identify and that don't belong
|
||||
catchException("Type [caught.type] does not belong in process' queue")
|
||||
Reference in New Issue
Block a user