Palworld's dedicated server has leaked memory since launch in 2024, and — let's clear this up first — the 1.0 update did not fix it. Pocketpair's performance work has softened it over the years, but the fundamental behavior remains: the longer a server process runs, the more RAM it holds, until players feel it.
Here's how to recognize it and what actually works, from running a fleet of Palworld servers.
The symptoms
The leak follows a predictable arc:
- Fresh restart: the server sits at its baseline — snappy, responsive.
- Day 2–3: RAM has crept up noticeably. Players start reporting rubber-banding — moving forward and snapping back — and delayed Pal actions.
- Day 4–6: lag spikes during busy moments; base automation stutters.
- Eventually: on a machine without spare memory, the process gets killed or crashes outright. Unmitigated servers commonly die within about a week.
If your server "gets worse the longer it's up" and a restart makes everything instantly fine — that's the leak, not your hardware.
Fix 1: A scheduled daily restart (the one that matters)
The leak resets completely when the process restarts. A clean daily restart keeps the server permanently in that fresh-baseline zone, and no amount of RAM headroom achieves the same thing — bigger hardware just makes the arc take longer.
Two things make restarts safe rather than disruptive:
- Save before stopping. Trigger a world save, then stop. A hard kill mid-save is how you corrupt a world.
- Pick a quiet hour. Restart when your region sleeps, not mid-raid. The downtime is seconds.
If you self-host, cron this. On Connect Hosting every Palworld server gets this automatically — one restart a day, scheduled in your region's quiet hours, world saved first.
Fix 2: Turn off invader raids
The setting bEnableInvaderEnemy (the "hostile invaders" raid toggle) is strongly associated with the leak's growth rate — server operators consistently report RAM climbing at roughly half the speed with invaders disabled. If your group doesn't care about base raid events, this is the single best settings-level mitigation. It's a normal option in PalWorldSettings.ini; on our panel it's the "Hostile Pals & Enemies" toggle in world settings.
Fix 3: Sensible headroom
Mitigations reduce the slope; they don't make memory static. Size the server so between-restart drift never touches your ceiling — our RAM guide has concrete numbers per group size. The short version: 16GB for most groups, 24GB for packed 32-player worlds.
What doesn't work
- "Just add more RAM." Delays the symptoms, changes nothing structurally.
- Community "leak fix" mods. Nothing mod-side can fix a server-process leak; at best they mask symptoms, at worst they break on every patch (1.0 broke a wave of stale early-access mods exactly this way).
- Waiting for the patch that fixes it. People have been doing that since 2024. Run restarts.
The zero-effort version
This is one of those problems that's genuinely boring to solve — a saved world and a nightly restart — but only if something reliably does it every single day. That's exactly what our Palworld hosting automates: daily leak-protection restarts on every plan, with the world saved cleanly first, plus the full 32 player slots included whichever plan size you pick.
