What Actually Happens When Your Modded Minecraft Server Crashes

It's Friday night. Four of you finally logged on at the same time, which took two weeks to arrange. Someone places a machine, the chunk loads, and the server dies.

Now what?

For most people hosting a modded server today, the answer is: *you* figure it out. You read a crash log you don't understand, you Google a stack trace, you find a two-year-old Reddit thread that says "remove Optifine," and at some point everyone goes to bed.

We think that's the wrong answer, and we changed how our own service works because of it. But to explain why, it's worth being precise about what actually breaks and what hosting companies actually promise - because those two things overlap far less than most people assume.

Why modded servers crash

Vanilla Minecraft servers are stable. Modded servers are a different category of software entirely: 300 mods written by 300 different people, loaded into one JVM, expected to cooperate.

The common causes, roughly in order of how often we see them:

• Out of memory. The single most common cause. A pack that runs fine with four players starts thrashing the garbage collector with six, or a world grows past what the allocated heap can hold. The server doesn't crash cleanly - it slows, then stops responding, then dies.
• Mod conflicts after an update. One mod updates, another expects the old API, and a crash appears three days later when someone finally triggers the code path.
• Corrupted chunks. A world-generation mod writes something malformed. The server crashes every time anyone walks into that chunk, forever, until someone removes it.
• Bad configs. A single mistyped value in one of several hundred config files.
• Mod loader and Java mismatches. Forge, NeoForge, or Fabric expecting a Java version the server isn't running.

Notice what these have in common: almost none of them are the hosting company's fault. The machine is fine. The network is fine. The disk is fine. The software you chose to run on it fell over.

That distinction is exactly where the industry's promises quietly stop.

What "99.9% uptime" actually means

Nearly every Minecraft host advertises 99.9% uptime. We do too. Two things are worth knowing about that number.

First, the arithmetic. 99.9% of a 30-day month allows 43 minutes and 12 seconds of downtime. Not much - but not nothing either. A host can go dark for 43 minutes a month, every month, and still be fully within its promise.

Second, and more important: it usually measures the platform, not your server. An uptime SLA covers the provider's infrastructure - network, hardware, control panel. It's a statement that the machine your server sits on was powered, connected, and reachable.

Your Minecraft server can be completely dead while the platform is at 100% uptime. The host's monitoring sees a healthy machine. Their SLA is intact. Your friends are staring at "Connection refused."

This isn't a scam, and it isn't hidden - it's how infrastructure SLAs have always worked, in Minecraft hosting and in cloud computing generally. But it does mean the guarantee most people are shopping on doesn't cover the failure they're actually worried about.

So who fixes your server?

In practice, at most hosts: you do.

You'll get a control panel with a restart button, a file manager, and a backup you can restore yourself if you know which one to pick and what you'll lose. Support will help - usually well, sometimes quickly - but the ticket is a request for assistance with *your* problem, not a commitment to resolve it in a defined time.

That model makes complete sense at scale. If you have tens of thousands of servers, you cannot promise to personally debug a modpack. The economics don't work. So the industry converged on renting you infrastructure and being helpful about the rest.

It's a reasonable business. It's just not what a group of five friends who want to play ATM10 actually needs.

What we changed

We decided that if the server is down, getting it back up is our job - whatever caused it.

Our SLA now guarantees: you report that your server is down, and it's playable again within 60 minutes, or your next billing period is free.

Two details make that promise deliverable rather than marketing:

We promise to restore, not to repair. Restoring means getting you a working, joinable server. Depending on what broke, that might be a restart, a rebuild, or a rollback to your most recent backup. Some modpack problems genuinely take longer than an hour to diagnose properly - and we'd rather get you playing tonight and find the root cause tomorrow than leave you offline while we read logs. The fix and the diagnosis are different jobs with different clocks.

The restore applies even when you broke it. If you installed a mod that killed the server, that's still us getting you back online. Helping you is the service. We do exclude those cases from the *credit* - a mod author's bug isn't our failure to pay for - but never from the help.

Underneath it, three things do the actual work:

• Monitoring that checks servers are joinable, not just that the machine is powered on. That distinction is the whole point of the previous section.
• Automatic restart on crash detection, running 24 hours a day. Most modded crashes are resolved in minutes this way, with nobody woken up.
• Daily offsite backups, kept for 7 days. Offsite means separate infrastructure - if a machine fails completely, the backups are unaffected. A rollback can cost you up to 24 hours of progress, and we'll always tell you exactly what you'd lose before we do it.

The honest limits

A guarantee with no limits is a lie, so here are ours.

The 60-minute clock runs 08:00-22:00 Amsterdam time, seven days a week. Outside those hours, monitoring and automatic recovery keep running, and in practice we often reply anyway - but we don't promise it, because we're a small team and we sleep. We'd rather write down a promise we can keep every time than a bigger one we break at 4am.

The clock starts when you report it, not when the crash happened, and it pauses while we're waiting on you - most often when we need permission before rolling your world back.

Credits are capped: one per month, three per year. If we're missing our target more often than that, credits aren't the right conversation. We're failing you, and we'd rather refund you properly.

The full text is at piggo.host/sla, including everything that's excluded and why.

Why we think this is the right trade

Running a modded server for a friend group is a small, thankless job that lands on one person. That person becomes the group's sysadmin: they get the messages when it's down, they're the reason nobody could play on Saturday, and eventually the server quietly dies and the group stops playing together.

We'd rather be that person than let it be you. Charging more than a budget host and taking responsibility for the outcome is a coherent trade - and being specific about it in writing, with hours and exclusions and caps, is how you tell the difference between a company that means it and a company with a nice sentence on its homepage.

---

Frequently asked questions

Does 99.9% uptime mean my Minecraft server will always be online? No. Uptime SLAs typically measure the hosting provider's infrastructure - network, hardware, control panel - not the game server process running your modpack. Your server can crash from a mod conflict or run out of memory while the host's platform is at 100% uptime.

How much downtime does a 99.9% uptime SLA allow? 43 minutes and 12 seconds per 30-day month. 99.5% allows about 3.6 hours, and 99% allows about 7.2 hours.

Whose fault is it when a modded Minecraft server crashes? Usually nobody's, in the sense people mean. The most common causes - running out of memory, mod conflicts after an update, corrupted chunks, misconfigured files - come from the modpack rather than the hosting hardware. The more useful question is not whose fault it is, but whose job it is to fix.

What's the difference between restoring a server and fixing it? Restoring means getting you a working, joinable server again - by restart, rebuild, or rollback to a backup. Fixing means finding and correcting the root cause. Restoring can almost always be done in minutes; a proper diagnosis sometimes takes hours. Good hosts do the first one immediately and the second one afterwards.

Do Minecraft hosts guarantee they'll fix a crashed server? Most guarantee infrastructure uptime and offer support, but don't commit to a time limit for restoring a crashed game server. Piggo guarantees restoration within 60 minutes of a report, between 08:00 and 22:00 Amsterdam time, or your next billing period is free.

How often should a Minecraft server be backed up? Daily is the practical minimum for a modded server, ideally stored on separate infrastructure from the server itself so a hardware failure can't take both. Piggo takes daily offsite backups and keeps them for 7 days. You should still keep your own copy of any world you'd be devastated to lose.

How much RAM does a modded Minecraft server need? As a rough guide: 6 GB for lighter packs with a couple of players, 8 GB for most modern packs with a few friends, 12 GB for heavy packs like All the Mods 10 with a full group, and 16 GB or more for large packs with bigger communities. Running under-provisioned is one of the most common causes of crashes.