timetout, max_upload, max_download and max_websocket_message may now be
configured on a domain-by-domain basis. This uses the same system that
we use for the block/allow-list from before:
Example:
[[http.rules]]
host = "*"
action = "allow"
max_upload = 4194304
max_download = 16777216
timeout = 30000
OK, so let's get this out of the way, there's some actual changes mixed
in here too. I'm really sorry:
- Turtles can now not be renamed with unnamed item tags (previously it
would clear the name, this seemed a little unideal).
- commands.getBlock(s)Data will also include NBT.
Now, onto the horror story which is these inspection changes:
- Make a lot of methods static
- Typo fixes
- Make utility classes final + private constructor
- Lots of reformatting (ifs -> ternary, invert control flow, etc...)
- ???
- Profit!
I'm so going to regret this - can pretty much guarantee this is going to
break something.
Unlike ComputerThread, we do not have a single source of tasks, and so
need a smarter way to handle scheduling and rate limiting. This
introduces a cooldown system, which works on both a global and
per-computer level:
Each computer is allowed to do some work for 5ms. If they go over that
budget, then they are marked as "hot", and will not execute work on the
next tick, until they have cooled down. This ensures that _on average_
computers perform at most 5ms of work per tick.
Obviously this is a rather large time span, so we also apply a global
10ms to all computers. This uses the same cooldown principle, meaning we
keep to an average of 10ms, even if we go over budget.
We attempted to simplify this 0bfb7049b0,
but that change now means that minimumVirtualRuntime is not updated. As
a result, new tasks will have a runtime of 0 when the queue is empty.
- Only update all runtimes and the minimum runtime when queuing new
exectors. We only need to update the current executor's runtime.
- Fix overflows when comparing times within TimeoutState.
System.nanotime() may (though probably won't) return negative values.
- Hopefully explain how the scheduler works a little bit.
- Runners would set their active executor before starting resetting the
time, meaning it would be judged as running and terminated.
- Similarly, the cumulative time start was reset to 0, meaning the
computer had been judged to run for an impossibly long time.
- If a computer hit the terminate threshold, but not the hard abort
one, then we'd print the stack trace of the terminated thread - we
now do it before interrupting.
There's still race conditions here when terminating a computer, but
hopefully these changes will mean they never occur under normal
operations (only when a computer has run for far too long).
- Fix the timeout error message displaying utter rot.
- Don't resize the runner array. We don't handle this correctly, so
we shouldn't handle it at all.
- Increment virtualRuntime after a task has executed.
- The computer queue is a priority queue sorted by "virtual runtime".
- Virtual runtime is based on the time this task has executed, divided
by the number of pending tasks.
- We try to execute every task within a given period. Each computer is
allocated a fair share of that period, depending how many tasks are
in the queue. Once a computer has used more than that period, the
computer is paused and the next one resumed.
TimeoutState now introduces a TIMESLICE, which is the maximum period of
time a computer can run before we will look into pausing it.
When we have executed a task for more than this period, and if there are
other computers waiting to execute work, then we will suspend the
machine.
Suspending the machine sets a flag on the ComputerExecutor, and pauses
the "cumulative" time - the time spent handling this particular event.
When resuming the machine, we restart our timer and resume the machine.
- TimeoutState uses nanoseconds rather than milliseconds. While this is
slightly less efficient on Windows, it's a) not the bottleneck of Lua
execution and b) we need a monotonic counter, otherwise we could
fail to terminate computers if the time changes.
- Add an exception handler to all threads.
- Document several classes a little better - I'm not sure how useful
all of these are, but _hopefully_ it'll make the internals a little
more accessible.
- Move state management (turnOn, shutdown, etc...) event handling and
the command queue into a ComputerExecutor
- This means the computer thread now just handles running "work" on
computer executors, rather than managing a separate command queue +
requeuing it.
- Instead of setting soft/hard timeouts on the ILuaMachine, we instead
provide it with a TimeoutState instance. This holds the current abort
flags, which can then be polled within debug hooks.
This means the Lua machine has to do less state management, but also
allows a more flexible implementation of aborts.
- Soft aborts are now handled by the TimeoutState - we track when the
task was started, and now only need to check we're more than 7s since
then.
Note, these timers work with millisecond granularity, rather than
nano, as this invokes substantially less overhead.
- Instead of having n runners being observed with n managers, we now
have n runners and 1 manager (or Monitor).
The runners are now responsible for pulling work from the queue. When
the start to execute a task, they set the time execution commenced.
The monitor then just checks each runner every 0.1s and handles hard
aborts (or killing the thread if need be).
- Rename unload -> close to be a little more consistent
- Make pollAndResetChanged be atomic, so we don't need to aquire a lock
- Get the computer queue from the task owner, rather than a separate
argument.
The latest version of Cobalt has several major changes, which I'm
looking forward to taking advantage of in the coming months:
- The Lua interpreter has been split up from the actual LuaClosure
instance. It now runs multiple functions within one loop, handling
pushing/popping and resuming method calls correctly.
This means we have a theoretically infinite call depth, as we're no
longer bounded by Java's stack size. In reality, this is limited to
32767 (Short.MAX_VALUE), as that's a mostly equivalent to the limits
PUC Lua exposes.
- The stack is no longer unwound in the event of errors. This both
simplifies error handling (not that CC:T needs to care about that)
but also means one can call debug.traceback on a now-dead coroutine
(which is more useful for debugging than using xpcall).
- Most significantly, coroutines are no longer each run on a dedicated
thread. Instead, yielding or resuming throws an exception to unwind
the Java stack and switches to a different coroutine.
In order to preserve compatability with CC's assumption about LuaJ's
threading model (namely that yielding blocks the thread), we also
provide a yieldBlock method (which CC:T consumes). This suspends the
current thread and switches execution to a new thread (see
SquidDev/Cobalt@b5ddf164f1 for more
details). While this does mean we need to use more than 1 thread,
it's still /substantially/ less than would otherwise be needed.
We've been running these changes on SwitchCraft for a few days now and
haven't seen any issues. One nice thing to observe is that the number of
CC thread has gone down from ~1.9k to ~100 (of those, ~70 are dedicated
to running coroutines). Similarly, the server has gone from generating
~15k threads over its lifetime, to ~3k. While this is still a lot, it's
a substantial improvement.
This allows us to track how much work various peripherals are doing.
This will not work with all systems, such as Plethora, as that has its
own execution system.
- Trackers are created per-user, meaning multiple people can run
/computercraft track at once.
- Allow sorting the tracking information by arbitrary fields.
- Add support for tracking arbitrary fields (though none are currently
implemented).
- ComputerThread constructs multiple threads instead of just one,
depending on a config options.
- The synchronized blocks of PeripheralAPI.PeripheralWrapper have been
shifted a little to ensure no deadlocks occur.
This makes a couple of significant changes to the original system, to
reduce the number of threads created and allow for multiple threads in
the future. There are several notable changes from the original
implementation:
- A blocking queue is used for the main task queue queue. This removes
the need for the "monitor" variable and allows for multiple threads
polling this queue in the future.
- The thread used to execute tasks is "cached" between tasks,
significantly reducing the number of threads which need to be
created. If a task needs to be stopped then the thread is then
terminated and a new one constructed, though this rarely happens.
This will hopefully make it easier to track down various issues which
might otherwise go unnoticed or provide little information.
The main areas modified are those that external APIs may provide values
for or interact with: various providers and ILuaObject/IPeripheral
implementations. However, we do also log in a couple of other places
which indicate a problem with this, or another, mod.
Threads that aren't daemon threads can keep the JVM from shutting down.
I'm certain that this doesn't happen very often but if one of these
threads hangs it can cause the rest of the JVM to not shut down
when the main thread exits.
By making all threads daemon threads if the main thread terminates
the rest of these threads will shut down.