Mirrors/Purpur
1
0
Fork 1
mirror of https://github.com/PurpurMC/Purpur.git synced 2026-02-18 17:07:43 +01:00
Code Issues Packages Projects Releases Wiki Activity

Updated Upstream (Paper & Airplane)

Browse Source 
Upstream has released updates that appear to apply and compile correctly

Paper Changes:
7259c32ba [CI-SKIP] [Auto] Rebuild Patches
0e06e912e Fix annotations so Inventory#getContents returns non-null array with nullable type (#5350)
b90b4f8ad Properly apply Sign#isEditable to TEs when calling BlockState#update
0641de2d4 [CI-SKIP] Change some fields to textara (#5405)
98a34ea47 [CI-SKIP] 🚀 Upgrade to new form-based templates (#5278)

Airplane Changes:
87d4a1705 Track queue debug command
This commit is contained in:
BillyGalbreath
2021-03-24 03:22:28 -05:00
parent 55244ddb2f
commit f31bcc632d
5 changed files with 1568 additions and 579 deletions
Download Patch File Download Diff File Expand all files Collapse all files

739
patches/server/0001-Tuinity-Server-Changes.patch
View File

@@ -3,587 +3,170 @@ From: Spottedleaf <Spottedleaf@users.noreply.github.com>
Date: Fri, 14 Dec 2018 21:53:58 -0800
Subject: [PATCH] Tuinity Server Changes
Brand changes
GNU LESSER GENERAL PUBLIC LICENSE
Version 3, 29 June 2007
MC-Dev fixes
Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
Everyone is permitted to copy and distribute verbatim copies
of this license document, but changing it is not allowed.
Util patch
This version of the GNU Lesser General Public License incorporates
the terms and conditions of version 3 of the GNU General Public
License, supplemented by the additional permissions listed below.
Tuinity Server Config
Multi-Threaded Server Ticking Vanilla
This patch is the vanilla server changes
Currently a placeholder patch.
Multi-Threaded ticking CraftBukkit
These are the changes to CB
Currently a placeholder patch.
Add soft async catcher
Must be enabled via -Dtuinity.strict-thread-checks=true
Delay chunk unloads
Chunk unloads are now delayed by 1s. Specifically, ticket level
reduction is delayed by 1s. This is done to allow players to
teleport and have their pets follow them, as the chunks will no longer
unload or have entity ticking status removed.
It's also targetted to reduce performance regressions when
plugins or edge cases in code do not spam sync loads since chunks
without tickets get unloaded immediately.
Configurable under `delay-chunkunloads-by` in config.
This patch replaces the paper patch as the paper patch only
affects player loaded chunks, when we want to target all
loads.
Attempt to recalculate regionfile header if it is corrupt
Instead of trying to relocate the chunk, which is seems to never
be the correct choice, so we end up duplicating or swapping chunks,
we instead drop the current regionfile header and recalculate -
hoping that at least then we don't swap chunks, and maybe recover
them all.
Lag compensate block breaking
Use time instead of ticks if ticks fall behind
Update version fetcher repo
Sets the target github repo to Tuinity in the version checker. Also disables the jenkins build lookups.
This patch is licensed under the MIT license. See /licenses/MIT.md.
Per World Spawn Limits
This patch is licensed under the MIT license. See /licenses/MIT.md.
Detail more information in watchdog dumps
- Dump position, world, velocity, and uuid for currently ticking entities
- Dump player name, player uuid, position, and world for packet handling
Execute chunk tasks mid-tick
This will help the server load chunks if tick times are high.
Change writes to use NORMAL priority rather than LOW
Should limit build up of I/O tasks, or at least properly
indicate to server owners that I/O is falling behind
Allow controlled flushing for network manager
Only make one flush call when emptying the packet queue too
This patch will be used to optimise out flush calls in later
patches.
Consolidate flush calls for entity tracker packets
Most server packets seem to be sent from here, so try to avoid
expensive flush calls from them.
This change was motivated due to local testing:
- My server spawn has 130 cows in it (for testing a prev. patch)
- Try to let 200 players join spawn
Without this change, I could only get 20 players on before they
all started timing out due to the load put on the Netty I/O threads.
With this change I could get all 200 on at 0ms ping.
(one of the primary issues is that my CPU is kinda trash, and having
4 extra threads at 100% is just too much for it).
So in general this patch should reduce Netty I/O thread load.
Time scoreboard search
Plugins leaking scoreboards will make this very expensive,
let server owners debug it easily
Make CallbackExecutor strict again
The correct fix for double scheduling is to avoid it. The reason
this class is used is because double scheduling causes issues
elsewhere, and it acts as an explicit detector of what double
schedules. Effectively, use the callback executor as a tool of
finding issues rather than hiding these issues.
This patch also reverts incorrect use(s) of the class by paper.
- getChunkFutureAsynchronously
There is no risk at all of recursion. The future is executed on
the chunk provider's thread queue, the same place general plugin
load callbacks are executed on. Forcing the task execution into
the callback executor also prevents the future from catching
any exception thrown from it.
Optimise entity hard collision checking
Very few entities actually hard collide, so store them in their own
entity slices and provide a special getEntites type call just for them.
This reduces entity collision checking impact (in my testing) by 25%
for crammed entities (shove 130 cows into an 8x6 area in one chunk).
Less crammed entities are likely to show significantly less benefit.
Effectively, this patch optimises crammed entity situations.
Improved oversized chunk data packet handling
Now target all TE data, except for TE's that do not have
update packets.
This patch relies upon the improve extra packet handling
patch, as we now use PacketPlayOutMapChunk as an extra packet.
See its patch notes for further details.
Reduce iterator allocation from chunk gen
Replace via iterating over an array
Prevent long map entry creation in light engine
Use fastiterator to prevent it
Highly optimise single and multi-AABB VoxelShapes and collisions
Reduce allocation rate from crammed entities
Optimise chunk tick iteration
Use a dedicated list of entity ticking chunks to reduce the cost
Use entity ticking chunk map for entity tracker
Should bring us back in-line with tracker performance
before the loaded entity list reversion.
Improve paper prevent moving into unloaded chunk check
Check the AABB of the move
Improve async tp to not load chunks when crossing worlds
Fixes an issue where a getCubes call would load neighbouring chunks.
Loads less chunks than paper's implementation
Revert getChunkAt(Async) retaining chunks for long periods of time
Rework PlayerChunk main thread checks
These need to fail instead of continuing, as hiding these errors
the way paper has is just going to allow unexpected reordering
of callbacks.
For example, thanks to this patch incorrect future
completion (completion of the world gen future,
PlayerChunkMap#b(PlayerChunk, ChunkStatus)) was detected and fixed.
Allow Entities to be removed from a world while ticking
Fixes issues like disconnecting players while ticking them, or
issues where teleporting players across worlds while ticking.
Also allows us to run mid tick while ticking entities.
Prevent unload() calls removing tickets for sync loads
Optimise collision checking in player move packet handling
Move collision logic to just the hasNewCollision call instead of getCubes + hasNewCollision
Manually inline methods in BlockPosition
Separate lookup locking from state access in UserCache
Prevent lookups from stalling simple state access/write calls
Distance manager tick timings
Recently this has been taking up more time, so add a timings to
really figure out how much.
Name craft scheduler threads according to the plugin using them
Provides quick access to culprits running far more threads than
they should be
Fix swamp hut cat generation deadlock
The worldgen thread will attempt to get structure references
via the world's getChunkAt method, which is fine if the gen is
not cancelled - but if the chunk was unloaded, the call will block
indefinitely. Instead of using the world state, we use the already
supplied generatoraccess which will always have the chunk available.
Range check flag dirty calls in PlayerChunk
Simply return.
Optimise tab complete
Some of the toLowerCase calls can be expensive.
Do not allow ticket level changes while unloading playerchunks
Sync loading the chunk at this stage would cause it to load
older data, as well as screwing our region state.
Make sure inlined getChunkAt has inlined logic for loaded chunks
Tux did some profiling some time ago and showed that the
previous getChunkAt method which had inlined logic for loaded
chunks did get inlined, but the standard CPS.getChunkAt
method was not inlined.
Paper recently reverted this optimisation, so it's been reintroduced
here.
Add packet limiter config
Example config:
packet-limiter:
kick-message: '&cSent too many packets'
limits:
all:
interval: 7.0
max-packet-rate: 500.0
PacketPlayInAutoRecipe:
interval: 4.0
max-packet-rate: 5.0
action: DROP
all section refers to all incoming packets, the action for all is
hard coded to KICK.
For specific limits, the section name is the class's name,
and an action can be defined: DROP or KICK
If interval or rate are less-than 0, the limit is ignored
Optimise closest entity lookup
Rewrites the entity slice storage so that entity by
class lookups look through less entities in total.
Also optimise the nearest entity by class method
used by entity AI as well.
As a sidenote, this entity slice implementation
removes the async catchers because it has been
designed to be MT-Safe for reads off of other
threads.
Optimise nearby player lookups
Use a distance map to map out close players.
Note that it's important that we cache the distance map value per chunk
since the penalty of a map lookup could outweigh the benefits of
searching less players (as it basically did in the outside range patch).
Remove streams for villager AI
Don't lookup fluid state when raytracing
Just use the iblockdata already retrieved, removes a getType call.
Reduce pathfinder branches
Reduce static path type detection to simple lazy-init fields
Add Velocity natives for encryption and compression
This patch is licensed under the MIT license. See /licenses/MIT.md.
Optimise non-flush packet sending
Places like entity tracking make heavy use of packet sending,
and internally netty will use some very expensive thread wakeup
calls when scheduling.
Thanks to various hacks in ProtocolLib as well as other
plugins, we cannot simply use a queue of packets to group
send on execute. We have to call execute for each packet.
Tux's suggestion here is exactly what was needed - tag
the Runnable indicating it should not make a wakeup call.
Big thanks to Tux for making this possible as I had given
up on this optimisation before he came along.
Locally this patch drops the entity tracker tick by a full 1.5x.
Do not retain playerchunkmap instance in light thread factory
The executor returned is finalizable and of course
that causes issues.
Do not load chunks during a crash report
This causes deadlocks in some cases when generating
crash reports.
Fixes https://github.com/Spottedleaf/Tuinity/issues/215
Improve abnormal server shutdown process
- When we're trying to kill the main thread from watchdog,
step up the stop() spamming after 15s to really kill the main thread.
- Do not wait for window disposing when disposing of the server
gui. It looks like during sigint shutdown there can be some
deadlock between the server thread and awt shutdown thread here.
Copy passenger list in enderTeleportTo
Fixes https://github.com/Spottedleaf/Tuinity/issues/208
Revert MC-4 fix
When messing around with collisions, I ran into problems where
entity position was off by ULP and that caused clipping problems.
Now, the collision epsilon is 1.0e-7 to account for those errors.
But this patch is going to cause problems on the order of 1.0e-4.
I do not want to deal with clipping problems. The very fact it works
shows it's causing the clipping to occur serverside.
Prevent light queue overfill when no players are online
block changes don't queue light updates (and they shouldn't)
Don't allow StructureLocateEvent to change worlds
Callers and even the function itself aren't expecting
this to happen
Properly handle cancellation of projectile hit event
Subclasses override and run logic. Some of this logic destroys
the projectile as well, which wouldn't be fitting for cancellation.
Rewrite the light engine
The standard vanilla light engine is plagued by
awful performance. Paper's changes to the light engine
help a bit, however they appear to cause some lighting
errors - most easily noticed in coral generation.
The vanilla light engine's is too abstract to be modified -
so an entirely new implementation is required to fix the
performance and lighting errors.
The new implementation is designed primarily to optimise
light level propagations (increase and decrease). Unlike
the vanilla light engine, this implementation tracks more
information per queued value when performing a
breadth first search. Vanilla just tracks coordinate, which
means every time they handle a queued value, they must
also determine the coordinate's target light level
from its neighbours - very wasteful, especially considering
these checks read neighbour block data.
The new light engine tracks both position and target level,
as well as whether the target block needs to be read at all
(for checking sided propagation). So, the work done per coordinate
is significantly reduced because no work is done for calculating
the target level.
In my testing, the block get calls were reduced by approximately
an order of magnitude. However, the light read checks were only
reduced by approximately 2x - but this is fine, light read checks
are extremely cheap compared to block gets.
Generation testing showed that the new light engine improved
total generation (not lighting itself, but the whole generation process)
by 2x. According to cpu time, the light engine itself spent 10x less time
lighting chunks for generation.
Optimise WorldServer#notify
Iterating over all of the navigators in the world is pretty expensive.
Instead, only iterate over navigators in the current region that are
eligible for repathing.
Actually unload POI data
While it's not likely for a poi data leak to be meaningful,
sometimes it is.
This patch also prevents the saving/unloading of POI data when
world saving is disabled.
Send full pos packets for hard colliding entities
Prevent collision problems due to desync (i.e boats)
Configurable under
`send-full-pos-for-hard-colliding-entities`
Fix chunks refusing to unload at low TPS
The full chunk future is appended to the chunk save future, but
when moving to unloaded ticket level it is not being completed with
the empty chunk access, so the chunk save must wait for the full
chunk future to complete. We can simply schedule to the immediate
executor to get this effect, rather than the main mailbox.
Fix incorrect isRealPlayer init
Some plugins, namely ProtocolSupport, don't route to where
paper placed their logic. So it wont correctly set in this case.
Fix by moving it to a different place.
Do not run raytrace logic for AIR
Saves approx. 5% for the raytrace call, as most (expensive)
raytracing tends to go through air and returning early is an
easy win. The remaining problems with this function
are mostly with the block getting itself.
Make entity tracker use highest range of passengers
This should prevent people from having to up their animal range
just so players riding horses or whatever can be seen at the
configured player range.
Do not allow the server to unload chunks at request of plugins
In general the chunk system is not well suited for this behavior,
especially if it is called during a chunk load. The chunks pushed
to be unloaded will simply be unloaded next tick, rather than
immediately.
Do not run close logic for inventories on chunk unload
Still call the event and change the active container though. We
want to avoid close logic because it's possible to load the
chunk through it. This should also be OK from a leak prevention/
state desync POV because the TE is getting unloaded anyways.
Fix NPE in pickup logic for arrow
Apparently the shooter can be null for tridents returning to players.
Remove chunk lookup & lambda allocation from counting mobs
The chunk lookup doesn't look cheap.
Optimise snow & ice in chunk ticking
Avoid the biome lookups most of the time by
hiding them behind the snow/air checks. The biome
checks take up the most here, followed by the block gets.
Soften some of the block gets by passing the chunk to
the biome methods, removing the need to invoke a chunk
lookup.
Oprimise map impl for tracked players
Reference2BooleanOpenHashMap is going to have
better lookups than HashMap.
Use hash table for maintaing changed block set
When a lot of block changes occur the iteration for checking can
add up a bit and cause a small performance impact.
Replace player chunk loader system
The old one has undebuggable problems. Rewriting seems
the most sensible option.
This new player chunk manager will also strictly rate limit
chunk sends so that netty threads do not get overloaded, whether
it be from the anti-xray logic or the compression itself.
Chunk loading is also rate limited in the same manner, so this
will result in a maximum responsiveness for change.
Config:
```
player-chunks:
autoconfig-send-distance: true
min-load-radius: 3
max-concurrent-sends: 12.0
max-concurrent-loads: 5.0
```
autoconfig-send-distance - Whether to try to use the client's
view distance for the send view distance in the server. In the
case that no plugin has explicitly set the send distance and
the client view distance is less-than the server's send distance,
the client's view distance will be used. This will not affect
tick view distance or no-tick view distance.
min-load-radius - The radius of chunks around a player that
are not throttled for loading. The number of chunks
affected is actually the configured value plus one as this
config controls the chunks the client will be able to render.
max-concurrent-sends - The maximum number of chunks that
can be queued to send at any given time. Low values
are generally going to solve server-sided networking
bottlenecks like anti-xray and chunk compression. Client
side networking is unlikely to be helped (i.e this wont help
people running off McDonald's wifi). Setting this
value to negative will make the server dynamically scale it
with players. i.e -5 will use 5 * online players for the max sends.
max-concurrent-loads - The maxmium number of chunks
that can be queued to be loaded at any given time. Lower
values help the responsitivity to player movement and
higher values help loading when the server is at a low TPS.
Optimise general POI access
There are a couple of problems with mojang's POI code.
Firstly, it's all streams. Unsurprisingly, stacking
streams on top of each other is horrible for performance
and ultimately took up half of a villager's tick!
Secondly, sometime's the search radius is large and there are
a significant number of poi entries per chunk section. Even
removing streams at this point doesn't help much. The only solution
is to start at the search point and iterate outwards. This
type of approach shows massive gains for portals, simply because
we can avoid sync loading a large area of chunks. I also tested
a massive farm I found in JellySquid's discord, which showed
to benefit significantly simply because the farm had so many
portal blocks that searching through them all was very slow.
Great care has been taken so that behavior remains identical to
vanilla, however I cannot account for oddball Stream API
implementations, if they even exist (streams can technically
be loose with iteration order in a sorted stream given its
source stream is not tagged with ordered, and mojang does not
tag the source stream as ordered). However in my testing on openjdk
there showed no difference, as expected.
This patch also specifically optimises other areas of code to
use PoiAccess. For example, some villager AI and portaling code
had to be specifically modified.
Be aware of entity teleports when chunk checking entities
If an entity were to be teleported via the sync load
that occured from the getChunkAt call, we would either
add to the wrong chunk or add to both chunks! not good.
Custom table implementation for blockstate state lookups
Testing some redstone intensive machines showed to bring about a 10%
improvement.
0. Additional Definitions.
As used herein, "this License" refers to version 3 of the GNU Lesser
General Public License, and the "GNU GPL" refers to version 3 of the GNU
General Public License.
"The Library" refers to a covered work governed by this License,
other than an Application or a Combined Work as defined below.
An "Application" is any work that makes use of an interface provided
by the Library, but which is not otherwise based on the Library.
Defining a subclass of a class defined by the Library is deemed a mode
of using an interface provided by the Library.
A "Combined Work" is a work produced by combining or linking an
Application with the Library. The particular version of the Library
with which the Combined Work was made is also called the "Linked
Version".
The "Minimal Corresponding Source" for a Combined Work means the
Corresponding Source for the Combined Work, excluding any source code
for portions of the Combined Work that, considered in isolation, are
based on the Application, and not on the Linked Version.
The "Corresponding Application Code" for a Combined Work means the
object code and/or source code for the Application, including any data
and utility programs needed for reproducing the Combined Work from the
Application, but excluding the System Libraries of the Combined Work.
1. Exception to Section 3 of the GNU GPL.
You may convey a covered work under sections 3 and 4 of this License
without being bound by section 3 of the GNU GPL.
2. Conveying Modified Versions.
If you modify a copy of the Library, and, in your modifications, a
facility refers to a function or data to be supplied by an Application
that uses the facility (other than as an argument passed when the
facility is invoked), then you may convey a copy of the modified
version:
a) under this License, provided that you make a good faith effort to
ensure that, in the event an Application does not supply the
function or data, the facility still operates, and performs
whatever part of its purpose remains meaningful, or
b) under the GNU GPL, with none of the additional permissions of
this License applicable to that copy.
3. Object Code Incorporating Material from Library Header Files.
The object code form of an Application may incorporate material from
a header file that is part of the Library. You may convey such object
code under terms of your choice, provided that, if the incorporated
material is not limited to numerical parameters, data structure
layouts and accessors, or small macros, inline functions and templates
(ten or fewer lines in length), you do both of the following:
a) Give prominent notice with each copy of the object code that the
Library is used in it and that the Library and its use are
covered by this License.
b) Accompany the object code with a copy of the GNU GPL and this license
document.
4. Combined Works.
You may convey a Combined Work under terms of your choice that,
taken together, effectively do not restrict modification of the
portions of the Library contained in the Combined Work and reverse
engineering for debugging such modifications, if you also do each of
the following:
a) Give prominent notice with each copy of the Combined Work that
the Library is used in it and that the Library and its use are
covered by this License.
b) Accompany the Combined Work with a copy of the GNU GPL and this license
document.
c) For a Combined Work that displays copyright notices during
execution, include the copyright notice for the Library among
these notices, as well as a reference directing the user to the
copies of the GNU GPL and this license document.
d) Do one of the following:
0) Convey the Minimal Corresponding Source under the terms of this
License, and the Corresponding Application Code in a form
suitable for, and under terms that permit, the user to
recombine or relink the Application with a modified version of
the Linked Version to produce a modified Combined Work, in the
manner specified by section 6 of the GNU GPL for conveying
Corresponding Source.
1) Use a suitable shared library mechanism for linking with the
Library. A suitable mechanism is one that (a) uses at run time
a copy of the Library already present on the user's computer
system, and (b) will operate properly with a modified version
of the Library that is interface-compatible with the Linked
Version.
e) Provide Installation Information, but only if you would otherwise
be required to provide such information under section 6 of the
GNU GPL, and only to the extent that such information is
necessary to install and execute a modified version of the
Combined Work produced by recombining or relinking the
Application with a modified version of the Linked Version. (If
you use option 4d0, the Installation Information must accompany
the Minimal Corresponding Source and Corresponding Application
Code. If you use option 4d1, you must provide the Installation
Information in the manner specified by section 6 of the GNU GPL
for conveying Corresponding Source.)
5. Combined Libraries.
You may place library facilities that are a work based on the
Library side by side in a single library together with other library
facilities that are not Applications and are not covered by this
License, and convey such a combined library under terms of your
choice, if you do both of the following:
a) Accompany the combined library with a copy of the same work based
on the Library, uncombined with any other library facilities,
conveyed under the terms of this License.
b) Give prominent notice with the combined library that part of it
is a work based on the Library, and explaining where to find the
accompanying uncombined form of the same work.
6. Revised Versions of the GNU Lesser General Public License.
The Free Software Foundation may publish revised and/or new versions
of the GNU Lesser General Public License from time to time. Such new
versions will be similar in spirit to the present version, but may
differ in detail to address new problems or concerns.
Each version is given a distinguishing version number. If the
Library as you received it specifies that a certain numbered version
of the GNU Lesser General Public License "or any later version"
applies to it, you have the option of following the terms and
conditions either of that published version or of any later version
published by the Free Software Foundation. If the Library as you
received it does not specify a version number of the GNU Lesser
General Public License, you may choose any version of the GNU Lesser
General Public License ever published by the Free Software Foundation.
If the Library as you received it specifies that a proxy can decide
whether future versions of the GNU Lesser General Public License shall
apply, that proxy's public statement of acceptance of any version is
permanent authorization for you to choose that version for the
Library.
diff --git a/pom.xml b/pom.xml
index 25ec9c4f8c08551fd80a597a1b39854e70e4f895..dad9981b1cd1ca8b27a45972188fd2b92f5146e7 100644