ChunkGenerator

First beta release after a round of production tuning on real hosts. The plugin now ships with auto-scaled throttle defaults calibrated against an actual container-host crash, six chunk-visit patterns, a resume command, and a simplified config aimed at public use.

Added

• Traversal patterns as an optional last argument to /cg start. Six patterns: center (default, spiral outward from the zone center), edge (concentric rings inward from the outer perimeter to the center), and north / south / east / west (linear row or column sweep starting at the named edge). All patterns visit the same set of chunks; only the order differs, and the saved iterator index is pattern-aware so a paused job resumes in its own sequence.
• /cg resume <world> to restart a paused job from where it left off, completing the start / stop / resume / cancel quartet. Paused jobs were previously only reactivated by a server restart with auto-resume: true. New permission node chunkgenerator.command.resume.
• inflight/target pipeline pressure reading in /cg status and the periodic console log. When chunks are being submitted but not completing (slow Paper chunk worker, stuck futures, far-out generation) the diagnosis is obvious from one command.
• Startup log lines that report the resolved throttle values and recommend a paper-global.yml > chunk-system > worker-threads setting matching the host's core count, so operators do not have to derive the Paper-side tuning themselves.
• README sections on container hosts and MaxRAMPercentage (Pterodactyl, Docker — heap-based monitoring cannot observe the cgroup RSS ceiling) and on maximising sustained throughput (the real ceiling is Paper's chunk worker count, not the plugin's queue depth).

Changed

• config.yml flattened and trimmed for public use. A short Basics block at the top (target-tps, bossbar / actionbar / console toggles, auto-resume) and an Advanced throttle section that most servers never need to touch. A banner at the very top points operators at paper-global.yml > chunk-system for the actual speed lever. The display, monitoring and persistence wrapper sections are gone — every knob lives at the root now.
• Auto-scaled defaults retuned against production data and tied to the core count rather than the heap alone, with a conservative 30 MB-per-inflight-chunk transient estimate.
  • max-inflight = max(64, min(cores × 32, heapMB / 60))
  • start-inflight = max(16, cores × 6)
• Default memory thresholds lowered: memory-backoff-pct 85 → 70, memory-pause-pct 92 → 80, to keep a margin from the container RSS limit that heap-based monitoring cannot see directly.

Fixed

• Kernel-OOM crash on container hosts (Pterodactyl, exit code 137) when the throttle saturated to its previous auto-scaled ceiling and Paper's chunk save thread fell behind: held chunks plus native memory exceeded the cgroup limit even while JVM heap stayed under the old 85 % backoff threshold.
• ETA no longer overflows to Long.MAX_VALUE (rendered as 2562047788015215h30m) when the smoothed speed decays toward zero. Below 0.01 chunk/s or for predicted ETAs over a year, the status returns ? instead of a meaningless number.
• Speed readout shows two decimals below 10 chunk/s and one decimal below 100, so sub-1 chunk/s progress no longer rounds to a flat 0.

Migration

• Delete plugins/ChunkGenerator/config.yml after upgrading so the new flat layout (with the paper-global.yml tuning banner) regenerates. Old YAMLs still load but stay on the previous keys that the plugin no longer reads.