#[Self-hosted dedicated server] Sailing rubber-banding worsens progressively over weeks of play

1 messages · Page 1 of 1 (latest)

civic sphinx
#

Summary

Position desync / rubber-banding that is slight on land but severe on water, with severity scaling with the number of players actively sailing.

The issue has developed gradually over weeks of play — there was no rubber-banding at launch and the server ran smoothly for days. It has been getting progressively worse and now makes ocean play effectively unplayable.

png-clipart-rubber-bands-natural-rubber-yermis-flexibility-elastic-band-material-band-thumbnail.png
#

Server setup

  • Self-hosted dedicated server running indifferentbroccoli/windrose-server-docker:latest on Ubuntu Server 24.04 + Docker + Wine
  • Hardware: Intel i5-6500 (4c/4t, 3.2 GHz), 32 GB RAM, Samsung 860 EVO SSD
  • Game build: 0.10.0.3.104-256f9653
  • 3 players, guild-only, same region, running since Early Access launch (April 14)
#

Symptoms

  • Mild/Slight rubber-banding on land when players are actively moving
  • Heavy rubber-banding / position desync the moment anyone boards a ship and sails
  • Severity scales with the number of players sailing simultaneously — 2+ ships on the water makes it dramatically worse
  • Started clean at launch, has been worsening progressively over the weeks of play — constant now, vs subtle a week or two ago
  • Player ping measured from the server: 21–47 ms (fine, all same region)
  • Same behaviour in P2P invite-code mode and direct connection mode
#

Server-side evidence

Full log extracts and 48h CPU/memory history in the attached archive.

Stalls in the async data-layer commit queue during sailing:
​log R5LogBLDalAQ: Warning: [s:1546] Slow task. Task was finished in 1682 ms. R5LogBLDalAQ: Warning: [s:1583] Slow task. Task was finished in 1764 ms. R5LogBLDalAQ: Warning: [s:1651] Slow task. Task was finished in 1605 ms. R5LogBLDalAQ: Warning: [s:1665] Slow task. Task was finished in 2115 ms. ​

  • 0 Slow task warnings in a recent 65 min of land-based play
  • 18 Slow task warnings in 15 min of sailing immediately after

CPU:

  • GameThread pegs at ~212 % CPU during sailing (more than 2 full cores)
  • CPU remains at ~212 % even after every player has logged out — idle CPU jumped from the historical baseline ~30 % to ~53 % after a single session that included a Blackbeard AI fleet encounter
  • CPU does not drop on container restart — points to state being loaded from the world save
#

Why this looks like cumulative save-state

The progression (clean → subtle → constant) without any config, mod, or hardware change on my side suggests something in the world save accumulates with play — persistent entities, loaded/streamed world chunks from exploration, AI ship wrecks, quest state, or similar. Fresh container + same world save = same bad CPU. This also lines up with the observation that severity scales with the number of players actively sailing (each moving player loads and ticks more world regions).

#

What I've already tested and ruled out

Mods are not the cause.
I run UE4SS + caffeinemodz's WindroseAPI (Nexus #44) + one in-house Lua position-tracker. Mods were installed ~April 19 and the server ran smoothly with them for days. The rubber-banding appeared later and has been worsening ever since, entirely decoupled from any mod change. No mods were added, removed, updated, or reconfigured during the progression.

#

Network / Kraken backend — direct connection mode (UseDirectConnection: true) has the same sailing stutter, so not the TURN relay or the Apr 14–16 ISP issue.

Container / Wine version — tested the wine-staging variant image, no change.

Host resource starvation — RAM fine, disk fine, other containers idle. The CPU is saturated by the game itself, not competing processes.

#

Attached

  • Server bundle — full logs from recent sessions, crash dumps, R5LogBLDalAQ Slow task extract, 48 h CPU/memory history from my custom server manager, docker-compose.yml, ServerDescription.json, WorldDescription.json, container-inspect.json
  • Client bundle — client logs, client crashes, sentry-native, DxDiag, system summary

Full technical writeup in EVIDENCE-SUMMARY.md inside the server archive.

Repro assistance

Happy to share the world save (several GB, separate archive) — given the CPU load persists across restarts and the progression correlates with play, this is probably the most useful thing I can give you.

waxen osprey
#

Thanks for the report! Our dedicated server does not natively support linux, so we can't do much here. But we're aiming to release a linux dedicated server in the future.

civic sphinx