LagXpert Free

LagXpert v2.7 — Autonomous Optimizer & Lag Diagnostics

This release does two things: it makes the automatic optimization actually run, and it adds a diagnostics system that tells you why your server is lagging and where. The honest framing matters here. Parts of v2.6 shipped configuration that was read, displayed in commands, and then never acted upon. This release closes those gaps and adds verification so they cannot silently reopen.

Lag Diagnostics (new)

The most common admin question is not "what is my TPS" — it is "which build is doing this to me". v2.7 answers that directly. /lagxpert diagnose scans every loaded chunk and produces a ranked report: - Scoring is relative to limits, not raw counts. A chunk with 40 mobs where the limit is 200 is fine. A chunk with 12 hoppers where the limit is 8 is a problem. Ranking by raw count surfaces the wrong chunks, so every metric contributes its percentage of the limit that actually applies to that world. - Intervention history counts. A chunk LagXpert has had to clean up twenty times in the last hour ranks above one that is merely full right now. - Limits come from the live adaptive engine, so the report reflects what is being enforced at that moment, matching what players experience. - Plain-language conclusions, not just numbers. The report tells you whether load is concentrated in a few chunks or spread everywhere, which violation type dominates, whether memory pressure is the real cause, which world is carrying the load, and — importantly — when the cause is outside LagXpert's scope so you stop looking in the wrong place. Scanning is two-phase: chunk snapshots are gathered on the owning thread, then all scoring happens asynchronously. Diagnosing lag must not cause lag. Reports are cached for 30 seconds so a command and a GUI opened together share one scan, and concurrent scans are rejected rather than queued.

Diagnostics GUI

/lagxpert diagnose in-game opens an interactive interface with five screens: - Overview — server state, TPS across three windows, memory, entity totals per world, the plain-language diagnosis, and the limits currently in force. - Hotspots — paginated ranked chunk list, colour-coded by severity, each showing which metrics are over limit. - Chunk detail — every measured contributor for one chunk, plus a teleport button to go look at it. - Corrective actions — the audit trail, with timestamps, triggers and counts. - Trends — worst hour of the day, peak players, entity growth direction and 24h projection.

Bedrock compatibility

This was built Bedrock-first rather than retrofitted: - Screen sizes come from BedrockPlayerUtils.getSafeInventorySize, and every slot position is computed from the actual size. Nothing is placed in a slot a Bedrock client will never render. - Materials that Geyser translates badly (spawn eggs, command blocks, structure blocks, knowledge books) are substituted automatically. - Lore is capped at 10 lines, past which Bedrock truncates tooltips unpredictably. - Legacy & colour codes only. - If the inventory cannot be opened at all, the full report is sent as chat text. A Bedrock admin is never left without a diagnosis. The Bedrock rules now live in one shared helper (BedrockUI) that both the config GUI and the diagnostics GUI route through, so they cannot drift apart. The previous BedrockCompatibleGUI class — 574 lines that registered no listener and had no callers — has been removed rather than left as a maintenance trap.

Emergency Controller

A four-level state machine (NORMAL → WARNING → CRITICAL → EMERGENCY) with hysteresis: 3 consecutive bad readings to escalate, 5 good to de-escalate, minimum 10 seconds per state. The EMERGENCY tier is now reachable. In the pre-release code computeTargetState() could only ever return NORMAL, WARNING or CRITICAL, which made the entire responses.emergency block — the 25% mob cap, the 16-block AI distance, the 1-minute unload threshold, and the custom command list — permanently dead. It now escalates via dedicated thresholds.tps.emergency / thresholds.ram.emergency values, and also on sustained CRITICAL: a server pinned at 12 TPS for two minutes is an emergency in practice even if it never crosses the raw threshold. Threshold ordering is validated at startup. Misconfiguring the tiers so one becomes unreachable now logs a warning instead of failing silently.

Responses that now actually execute

| Response | Status | |---|---| | Mob cap multiplier (75% / 50% / 25%) | working | | Chunk preloader pause | working | | Aggressive chunk unloading | working | | AI distance reduction | working | | Redstone clock shutdown | working | | Block natural spawns | now implemented | | Forced item cleanup | now implemented | | Server-wide AI freeze | now implemented | | Emergency commands | now reachable | The last four were configurable and reported by /lagxpert emergency but had no code acting on them. A new response coordinator consumes them: - Natural spawn blocking filters by SpawnReason. Only environmental pressure is suppressed — spawn eggs, breeding, plugin spawns and commands keep working, because blocking those makes a server look broken rather than busy. The list is operator-configurable, and players with lagxpert.bypass.mobs are exempt. - Forced cleanup runs off-cycle with a CAS latch and a cooldown, so a flapping state cannot stack cleanup sweeps. - AI freeze dispatches per chunk (correct under Folia), skips players and NPC-metadata entities, and on recovery respects your mobs.yml AI rules instead of blanket re-enabling everything. A shutdown mid-emergency lifts the freeze so mobs are never left permanently frozen.

Adaptive Thresholds

Previously this computed multipliers correctly and nothing consumed them — turning the feature on or off only changed numbers in /lagxpert status. It is now the single authority for "what is the limit right now", and is wired into: - StorageListener — hopper, chest, furnace, barrel, dropper, dispenser, shulker, TNT, piston and observer placement limits - EntityListener and SmartMobManager — mob caps - RedstoneCircuitTracker — pulse tolerance and maximum circuit duration - AutoChunkScanTask — so a chunk flagged as overloaded matches what players are actually restricted to - EntityCleanupTask — the per-chunk entity ceiling Two independent inputs are combined: continuous server health (from TPS and memory) and the discrete emergency state. The more restrictive of the two wins; they are never multiplied, which would compound into unplayable values. Limits are only ever scaled down — your configured value is always the ceiling — and a configurable minimum-multiplier floor prevents collapse. Per-player limits granted through lagxpert.limits.* permissions are honored verbatim and never scaled. The health factor now takes the worse of the TPS and memory components rather than averaging them, so healthy memory can no longer mask a genuinely bad TPS reading.

Optimization Profiles

profiles.yml previously shipped with zero Java code behind it: no reader, no profile subcommand. /lagxpert profile balanced returned "unknown subcommand". Now implemented:

/lagxpert profile list       show profiles and which is active
/lagxpert profile <name>     apply a profile
/lagxpert profile revert     restore the pre-profile configuration

Applying a profile writes into the real config files, saves them, and reloads every subsystem. Before the first write the replaced values are snapshotted, so revert restores your configuration rather than an assumed default — and applying a second profile over a first still reverts all the way back. safety.auto-revert-minutes guards against applying an aggressive profile during an incident and forgetting. Profile keys that do not map to a real sett