ShadowOptimizer

The name of this server in the TPS history.

Only matters when several servers write to the same MySQL: it is the column that keeps the graphs apart. "auto" uses the machine name and the port, which two servers on one machine cannot share.

How many samples between counting entities and chunks too.

This is the only read that costs anything: it means asking every world how many entities it holds. A count every fifteen seconds answers the same question as one every second and costs a fifteenth. Raise it if you have many worlds; there is no reason to lower it.

How many hours of history to keep in the database.

One row per minute, so 1440 rows a day: 72 hours is a little over four thousand rows, which is nothing. Raise it if you want to answer "since when has it been getting worse?" over weeks instead of days.

How many minutes between background scans.

It exists only so that whoever opens the GUI finds a report ready instead of a few seconds of waiting. Set 0 for none: on-demand scans keep working.

Note: cleanup.items.world-limit needs this above zero, because the per-world item count is refreshed by this scan and by nothing else.

How many chunks to look at per tick.

The scan takes a few seconds instead of one tick, and that is on purpose. Raising it makes the scan faster and more noticeable: on a server already in trouble that is exactly the wrong trade.

The hard ceiling of chunks to look at in one pass.

A network of ten worlds can have a hundred thousand chunks loaded. Past this number the scan stops: the first twenty thousand rows of a ranking contain the culprit anyway.

The budget is shared FAIRLY between worlds: each gets a slice, and whatever is left over goes to the ones that were cut. Otherwise the main world - always first in the list - would take the whole budget and the others would never be looked at at all.

How many rows each ranking has. 25 is more than enough to find a culprit.

How many milliseconds between looks.

100 means ten samples a second, so 36,000 in an hour: enough for the percentages to be stable. Lowering it makes the estimate finer and starts to cost; raising it makes it noisy.

The names of the threads that run the game.

Only worth touching on an exotic server build that names them differently. The comparison is "the name starts with".

How many minutes between passes.

Whether to announce the warning and the result in chat.

Leave it on. Half the staff reports on a server running an anti-lag come from things vanishing without notice, and a warning nobody sees is not a warning.

How many seconds ahead to warn. The pass starts after the last one.

How many chunks to handle per tick.

Worlds cleanup never enters.

This is the ONLY thing that switches cleanup off for a world: leave it empty and every loaded world is covered, the ones other plugins create at runtime included.

ShadowGames arenas exclude themselves while a match is running: there is no need to list them here.

Other plugins' metadata that marks an entity as "theirs".

The PersistentDataContainer and the scoreboard tags can be enumerated and are always checked. Bukkit metadata cannot: the interface only lets you ask for a key by name. Hence this list, which grows without recompiling. It is the least elegant part of the plugin and it is the best the API allows.

Other plugins' data that does NOT make an entity "theirs".

Some plugins write a note on ordinary mobs just to remember how they were born (a spawn egg, a spawner) - ShadowBattlePass and ShadowQuests do it so that kills of farm mobs don't count. Those mobs are still ordinary mobs and the cleanup removes them: the two keys of the suite are always in this list, add here the ones of other plugins that do the same. Format: namespace:key, as /data get entity shows it.

Merge before removing. This is the step that makes the plugin different from every other one: switch it off and it becomes a ClearLag.

Remove the items on the ground: ALL of them, however fresh, except

- the materials in never-remove below (plus the factory list),
- items that carry another plugin's data: crate keys, vouchers,
  custom items (they are recognised by that data, nothing to list).

A name written on an item does not save it. The countdown before the pass (warning-seconds) is the warning to pick things up.

How many ground items to tolerate in ONE world before stepping in right away. Zero switches the threshold off.

Materials never removed, however old.

This list is ADDED to the factory one, which already holds netherite, elytra, totems, the dragon egg, beacons, enchanted golden apples, nether stars and every shulker box.

Remove the mobs: ALL of them - natural, from spawn eggs, from spawners - wherever they are and whoever is near, except

- the types in never-remove below (plus the factory list: animals
  people breed, villagers, golems, bosses...),
- mobs with a name tag, tamed, on a leash, in a vehicle,
- other plugins' mobs (their data, or foreign-metadata-keys above).

Limit removal to hostile mobs.

OFF by default, because there is now something better: the list below. "Hostiles only" is wrong in two directions. It leaves out bats and squid, which are not hostile and which nobody breeds: on a server with an ocean those are thousands of untouchable entities. And it does not tell a cow from a villager, which are both passive and have nothing else in common.

Turn it on only if you prefer the blunt rule to the precise one.

Mob types NEVER removed, however old.

Set true to NOT start from the factory list.

For anyone who wants to take something OFF that list - for example to clean cows too, on a server where nobody breeds. It has to be said on purpose: deleting somebody's livestock by accident is exactly the defect this plugin exists in order not to have.

How many mobs of ONE precise type to tolerate in a world.

max-per-type:
  COW: 60
  SHEEP: 60
  CHICKEN: 40
  IRON_GOLEM: 20

Arrows and the like with no owner left.

HIGH | NORMAL | MEDIUM | LOW, or the name of a preset you write below.

One entity type, its own cap, whatever the preset says (-1 = none).

How long a chunk's count is trusted before it is counted again.

-1 on any entry means "no limit".

Hoppers.

They are the most expensive thing a player can build without noticing: each one checks every tick whether there is something to move above it, even when it is empty and even when nobody is around.

Over the limit they are NOT blocked: they are slowed down. Blocking them would break somebody's farm without touching a single block; slowing them leaves it working, just slower.

One transfer in how many gets through, over the limit.

How many milliseconds to reuse a chunk's count for.

Counting a chunk's entities on every spawn means counting them twenty times a second for the same chunk. One second is short enough for the limit to work and enormous compared to how often spawns happen.

The same, for hoppers. Nobody places forty a second.

At most how often to compact the items of a chunk that went past max-items.

The item limit never cancels anything: when a chunk is over the threshold, the identical stacks already there are merged and the new item is born anyway. On a farm producing without pause, without this ceiling it would compact dozens of times a second - which would make the anti-lag plugin the thing producing the lag.

Whether to tell anyone holding shadowoptimizer.notify in chat when a chunk reaches its limit. The message has a cooldown of its own, declared in the language file.

How many minutes it lasts when you do not say.

Chain explosions.

Redstone. This is a big thing to do to a server, which is why panic is switched on by hand, lasts a few minutes and is announced in chat.

Natural spawning and spawner spawning. Eggs, commands, breeding and plugins keep working: those are things a player asked for, and blocking them would make the server look broken.

Up to how many mobs in one stack.

The ceiling is not for the server: it is for the player. Killing a stack of a thousand means a thousand hits on the same spot, and at that point the farm is not playable any more.

Within how many blocks to look for a partner.

Small on purpose. A large radius merges mobs the player sees as separate - the one in front and the one round the corner - and what it feels like is monsters "disappearing".

From how many upwards to write the counter above the head.

How the counter is written. MiniMessage format. Players read this, so it may stay in the server language.

A baby does not join a stack of adults.

Two mobs with different names never merge. A name is how a player says "this one is mine".

Types that never stack.

Set true to NOT start from the factory list.

This is the switch that makes the list above the only truth, and it is the only way to REMOVE a type from it - for example to stack cows too, on a server where nobody breeds. It has to stay an explicit choice: whoever flips it without knowing finds out from the players, not from the log.

Worlds where mob stacking applies. Empty means all of them.

Spawns that were "wanted" and stack anyway.

Up to how many units in one pile.

High, because here there is nobody to keep entertained: 4096 cobble in a single entity is 64 fewer entities and nobody notices.

Do not raise it to a million: if you ever uninstall the plugin, the piles stay as ordinary items with their VISIBLE amount, and the excess is lost. At this value what gets lost is cobble.

From how many units the counter is written above the pile. 2 means every pile that is more than one thing says how many it is - a dropped stack of 32 too, not only a pile that outgrew 64.

Materials that never merge.

There is no need to list items with different enchantments or names: the comparison is the game's own and looks at every component. Two swords with different enchantments never merge, and there is no option here that could change that.

How many drops to deliver per tick.

A stack of fifty skeletons dying all at once produces two hundred and fifty entities. Spawning them in the same tick would be the problem this plugin solves, caused by the plugin itself.

The queue ceiling. If something kills ten thousand mobs in one command, losing drops beats filling memory. It is written in the console when it happens, and it has never happened for a legitimate reason.

The maximum loot multiplier for a stack that dies all at once. That only happens when something bypasses damage: /kill, the void, a plugin removing entities.

Whether to multiply experience too.

Only in these worlds. Empty = everywhere.

Materials that always fall on the ground, even with this on.

The case this exists for is a block whose drop is the point of the animation - something you want to see land.

Who gets it. Empty = everybody.

Fill it in and it becomes a rank perk: "shadowoptimizer.pickup".

In creative you already have everything, so by default this does not apply there: it only fills the bag with rubbish.

On for everybody. A player who does not want it turns it off for themselves with /safedrop off (remembered in safe-drop-players.yml).

Anything that wears out: tools, weapons, armour, bows, elytra.

Anything enchanted, even when it is not a tool. An enchanted book is worth as much as the sword it will end up on.

Anything a player has renamed. Off by default: a renamed stack of dirt would start asking questions too.

Always protected, whatever the three switches above say.

Never protected, even when they say so.

How long the second Q has to arrive.

Only in these worlds. Empty = everywhere.

How big a batch is for a spawner on its own.

The ceiling, even with many spawners grouped together.

Treat the spawners of one chunk as a single one.

This is what turns sixty spawners across two chunks into two events instead of sixty.

How many mobs of the same type around a spawner before stopping it.

The count adds up stack sizes: a stack of fifty counts as fifty, otherwise the limit would defeat itself precisely thanks to the optimisation.

Within how many blocks to count.

false = local H2 database inside plugins/ShadowOptimizer/data/. Nothing to install: the plugin works the moment it is copied in.

MySQL is good for one thing: seeing the TPS of every server on a network in a single graph.

Table prefix.

MySQL configured but unreachable: with true the plugin says so in the console and falls back to the local file, so the server still starts.

Small on purpose: this plugin writes one row a minute. A pool of ten connections for two inserts is memory thrown away.

Lets another process (a web panel, say) open the same H2 file while the server is running. Rarely needed.

The language used when no better one is found.

Ships with en_US and it_IT. Set "it_IT" here for an Italian server.

Follow the language each player set in their client.

With two languages shipped this means an English speaker reads English without doing anything. A missing key falls back to the default language, key by key: an incomplete translation still works.

Write health.yml in our folder, so the panel can show the server's health next to the updates.

How many seconds between rewrites.

Write every SQL query to the console. That is for us, not for you.

The title comes from the language file (gui.chunks.title) because it carries the page number: only what is not text is changed here.

The window title.

How many rows. From 1 to 6.