bff5ac5d37
Async chunk compression needs a little bit more work still, disable for now
2014-02-02 12:26:23 +11:00
77fbb062d9
Refactor chunk compression to use its own thread pool in order to eliminate disconnect race conditions caused by Mojang's nasty disconnect routines.
2014-02-02 09:41:27 +11:00
fd83772302
Cap Entity Collisions
...
Limit a single entity to colliding a max of configurable times per tick.
This will alleviate issues where living entities are hoarded in 1x1 pens.
2014-02-01 18:14:48 +11:00
be31d62f43
Cap Entity Collisions
...
Limit a single entity to colliding a max of configurable times per tick.
This will alleviate issues where living entities are hoarded in 1x1 pens.
2014-02-01 18:13:02 +11:00
b4530237fb
Implement Locale Getter for Players
2014-02-01 18:12:22 +11:00
16afc3b31c
Add wiki link to /timings on command
2014-02-01 18:08:42 +11:00
61a227c8c3
Add warning when null chunk buffers are encountered - most likely due to the orebfuscator plugin.
2014-02-01 18:07:22 +11:00
161e853450
In short, there isn't actually anything wrong with the async chunk compressor, it just accidentally caused console logging of errors which were previously ignored.
...
This commit restores that behaviour
You may be asking yourself why we are completely ignoring any errors which come this far down the pipeline.
The answer is quite simple:
Mojang did it
The default Mojang pipeline doesn't have any ChannelOutboundHandlerAdapter or similar instances, and thus nothing to handle exceptionCaught
So when a channel.write() or channel.flush() fails, the error message is actually just passed straight to the future provided.
It is then subsequently discarded, the channel closed, and no one except the user was any the wiser it actually happened!
Unfortunately for us, the default exceptionCaught in this class sends a blaring warning to the server admins indicating that it couldn't send a packet to a disconnected user!
We don't care about these warnings, if we did something wrong to disconnect the user, it is already logged in the proper location, as are broken sockets
tl;dr no need to blare warnings on each write to a broken socket
2014-02-01 17:53:32 +11:00
12358efad3
Use correct bitmask variable for orebfuscator
2014-02-01 10:44:25 +11:00
31f53fbbbe
Upstream merge
2014-01-31 20:42:29 +00:00
16ea6a1c25
Didn't update fully for some reason...
2014-01-31 21:22:40 +11:00
cf8a0fe75f
Automatically update upstream versions. Bukkit-1.7.2-R0.2-16-ga621d16, CraftBukkit-1.7.2-R0.2-30-g88d138f
2014-01-31 05:19:28 -05:00
8decebe96e
Don't allow respawning offline players
2014-01-31 18:15:33 +11:00
4864fd1c26
Sacrifice oreobfuscator performance for safety and reliability.
2014-01-31 16:07:28 +11:00
42a026634b
Correctly set arrows' current block
2014-01-30 20:42:01 +00:00
564ebde3ad
Move exception logging to method body so we don't end up logging all outbound exceptions, which other plugins (cough ProtocolLib) may or may not be causing.
...
Also rumours this patch causes the server to break with ProtocolLib installed.
@aadnk
2014-01-30 16:02:25 +11:00
7bd5666ee2
Show full stack trace for errors during chunk caching/compression
2014-01-30 06:58:37 +11:00
a780739071
Use a crc32 of all the blocks in the chunk as our hash key. This is much more unique.
2014-01-29 12:51:20 +11:00
1bf13129a5
Merge pull request #108 from Dmck2b/PermGenLink
...
Add more information about changing PermGen size
2014-01-28 16:04:00 -08:00
911b500845
Fix unit tests
2014-01-29 09:08:29 +11:00
feacfb991d
[Experimental] - Enable compressed bulk chunk cache.
...
Sort chunks a final time before sending. This gives the advantage that chunks will load as close to the player as they possibly can, and then move out, as well as increase the cache hit rate when using smaller bulk chunk limits such as the default of 5.
2014-01-29 09:04:51 +11:00
23615e1c9f
Use simpler iterator
2014-01-29 08:47:29 +11:00
202b1fd11d
Implement a compressed chunk cache. This still needs a lot of work to figure out a good mechanism for looking up previous chunks, however at the moment when coupled with a high bulk chunk limit, it can be effective for solving lag issues.
2014-01-29 08:47:29 +11:00
cd7bb0da14
Only 1 class is compressible, don't bother interfacing.
2014-01-29 08:38:37 +11:00
638b1724d0
Only create a single executor / compressor instance.
2014-01-29 08:38:37 +11:00
e8933c1890
Allow Configuring Chunks per Packet. This enables server owners to cram more chunks per packet, potentially leading to higher compression ratios, thus saving bandwidth and load time.
...
This is currently set at the Vanilla default value of 5, however initial testing seems to suggest there could be good performance / usability / bandwidth gains by setting it to its maximum value of 32768. Testers are welcome to experiment with this option, beware the placebo though!
2014-01-29 08:38:37 +11:00
82cf01b53f
Implement Threaded Bulk Chunk Compression.
...
This offloads chunk compression to a new thread, which should help with the pauses often seen when teleporting large amounts of players.
2014-01-29 08:38:37 +11:00
73d3519fca
Add more information about changing PermGen size
2014-01-28 17:51:26 +00:00
72b48c77af
Merge pull request #107 from brajo/master
...
Fix typo on /timings on message
2014-01-27 07:08:00 -08:00
9557f49051
Fix typo on /timings on message
2014-01-27 14:47:37 +00:00
6bf69c66f3
Merge tick shuffling branch into master.
2014-01-27 08:40:12 +11:00
d3f7cce233
Revert Spigot branding.
2014-01-26 21:54:02 +11:00
6eefe4e38e
Allow configuring of player sample size, and default the sample to 12, the same as Vanilla. This has some performance overhead, as we have to shuffle the list each time, but this is better than the server displaying as offline!
2014-01-26 21:50:53 +11:00
ab3c5a42f7
Revert "Experimentally brand as "Spigot". This will be reverted if *any* plugin compatibility issues arise."
...
This reverts commit 71255cf140
2014-01-26 21:33:46 +11:00
8e785b132f
Always use real player count for displayed online players.
2014-01-26 12:48:32 +11:00
aec53a59da
Initialise ping array on startup. The code to decide when to regenerate the array is a bit iffy (Mojang!), as it relies on System.nanoTime() which is not an absolute time source.
2014-01-26 12:30:38 +11:00
fa94a8a99a
Adjust TPS command to start at 20tps, show * for catchup TPS, and round to 2 decimal places.
2014-01-26 12:17:55 +11:00
8832052c84
Fix Ping Player Sample to bring it back in line with Vanilla behaviour. The behaviour which Bukkit introduced recently is broken on larger servers as it introduces too many players to the list.
2014-01-26 12:14:05 +11:00
14c0d5cba9
We want to use Math.min, not max to cap displayed TPS to 20
2014-01-26 11:04:55 +11:00
71255cf140
Experimentally brand as "Spigot". This will be reverted if *any* plugin compatibility issues arise.
2014-01-26 09:17:40 +11:00
3d6f241c12
Fix double comparison issues.
2014-01-25 16:18:32 +11:00
53f2f7499c
Don't allow plugins to create scoreboards async - this can crash the server.
2014-01-25 15:23:21 +11:00
3420b081e9
Merge pull request #99 from brutuscat2/master
...
TPS isn't capitalized
2014-01-24 19:45:22 -08:00
2f9b4a8573
TPS isn't capitalized
...
It makes me want to hurt someone when its written like "Tps"
2014-01-24 22:36:32 -05:00
a50a1d4c83
Implement a new TPS command which will show the average TPS over the last 1,5 and 15 minutes, using the same algorithm as the linux kernel (exponentiallialy damped moving average).
2014-01-25 14:09:21 +11:00
9a280ecec9
Disable /timings on as it produces inaccurate results.
2014-01-25 08:45:00 +11:00
49f9f20d1c
Add Spigot links.
2014-01-23 13:17:47 +11:00
526b404545
Move watchdog and timings code to tick method, not tick loop itself.
2014-01-22 19:23:59 +11:00
29c6c774dd
Pop tick loop patch to top
2014-01-22 19:19:39 +11:00
2367cf0a3d
Reapply tick loop patch.
2014-01-22 19:16:21 +11:00
md_5
Aikar
Smove
Thinkofdeath
SpigotMC
md-5
Dmck2b
thinkofdeath
brajo
brutuscat2