To handle player spikes at a multiplayer launch, pre-warm enough capacity for the first wave, start one server per match on demand, and spread that capacity across more than one infrastructure provider. Then load test the whole path from matchmaker to running server, so launch day is not the first time it runs at scale.
This post covers what breaks on launch day, how to plan and test for it, and what two large launches on Gameye looked like.
Launch-day failures at a glance
| Failure point | What players see | How to prevent it |
|---|---|---|
| Server allocation can’t keep up with match creation | Long waits after a match is found | Keep warm containers ready and start one server per match on demand |
| A cloud account hits its instance or vCPU quota | New matches fail in one region while others work | Raise quotas weeks ahead and spread capacity across providers |
| One provider has an outage | A whole region stops starting matches | Run more than one provider per region with automatic failover |
| Matchmaker retries pile up | Queues keep growing after capacity comes back | Retry after a short delay and fall back to a second region |
| Players arrive in a region you didn’t plan for | High ping or no matches for those players | Use a platform that can add regions without long reservations |
What actually fails on launch day
When you plan for launch, look past the game server process. Load builds up in the steps before a match starts, and those steps are where players get stuck.
Allocation. A match forms in the matchmaker. Something then has to find a machine, start the server and return an address. If that step slows down under load, players sit in a queue that looks broken to them. If the fleet was sized for a forecast, it runs out as soon as demand passes the forecast.
Provider quotas. Cloud accounts limit how many instances or vCPUs you can run per region. The default limits on an account can sit well below a launch peak, and raising them takes a support request and time. You can have budget to spend and still have the provider refuse new machines.
Matchmaker backlog. When allocation fails, the matchmaker retries. If every pending match retries at once with no delay, the retries add load at the worst moment. The queue keeps growing after capacity returns, because the backlog has to drain first.
Capacity planning and load testing
A forecast is a starting point. Chivalry 2’s demand in its first hour was roughly double its best-case projection. Nubs! Arena saw roughly three times its forecast. Plan for the forecast to be wrong in both size and location.
- Size the baseline from the forecast and the ceiling from a multiple of it. Put the predictable part on reserved capacity. Make sure overflow capacity can cover several times that without a manual step.
- Pre-warm the first wave. On Gameye, a warm pool keeps a set number of containers running for each application and region. When a session is requested, one is assigned at once and the pool refills. Gameye bills warm pool containers while the pool holds them, so size the pool to expected concurrent demand, not to peak. Containers outside the pool start in 0.5 seconds on average.
- Load test the full path. Test the matchmaker, the allocator and the game server together. Run the test across several regions and use sudden concurrency spikes, not a smooth ramp.
- Test with real players. Chivalry 2 ran a 15-month closed alpha with regular stress tests against live infrastructure. Community reports on Discord during that alpha found a performance problem that internal testing missed. The team fixed it before launch.
- Check your incident tools under load. Make sure logs and debugging access work at peak before launch day.
Gameye’s orchestration layer has been load tested to 3 million concurrent users. That is an internal test figure. The highest production peak so far is 250,000 concurrent players, at the Chivalry 2 launch.
Multi-provider failover
If your whole fleet sits in one cloud account, an outage at that provider takes the region down. So does a quota you can’t raise in time.
The fix is more than one provider per region, with placement that moves new sessions on its own. Gameye runs on 21 infrastructure providers across more than 200 data centers. If a provider has issues, new sessions go to healthy providers in the same region. Existing sessions on healthy nodes keep running. Your matchmaker makes the same API call either way.
Placement is metal first. Every session tries bare metal first and only bursts to cloud when bare metal in that region is full. When demand drops, the cloud capacity is released. You get a fast, cheaper baseline on normal days and cloud headroom for the launch peak.
Your matchmaker still needs to handle a full region. Gameye returns a 420 when it can’t place a session in the requested region. The docs recommend a retry after 2 to 5 seconds, then a fallback to a secondary region if retries fail. Define those fallback regions before launch.
Chivalry 2 at launch
Torn Banner Studios and Tripwire Interactive launched Chivalry 2 in June 2021 on PC, PlayStation and Xbox. It reached 250,000 concurrent players in the first 30 minutes and nearly 1 million players in the first month, on dedicated servers in 30 locations.
Gameye allocated sessions on demand and added capacity across providers as concurrency climbed. Nobody had to intervene by hand, and there were no major downtime events.
Demand also came from Japan and Korea, which early capacity planning had not prioritized. Gameye added capacity in Tokyo and Seoul within hours. Per-second billing meant the studio paid for sessions that ran, not for idle capacity. Read the full Chivalry 2 case study.
No More Room in Hell 2 at launch
No More Room in Hell 2 ran more than 500,000 game server containers in one launch weekend on Gameye. Bare metal, flex metal and cloud capacity scaled together automatically. The game reached the Top 5 trending on PS5 at release.
On Gameye, each session runs in its own container. When a match ends, Gameye stops the container and reclaims the capacity for the next session, so capacity follows the number of live matches.
The Nubs! Arena launch is a smaller example of the same pattern. Demand came in at roughly three times forecast, and Gameye handled it without manual capacity changes. The figures behind these launches are on the platform data page.
Launch-day checklist
Before launch:
- Load test the matchmaker, allocator and servers together, with sudden spikes in several regions.
- Raise cloud quotas in every region you plan to burst into.
- Set up warm pools in your main regions, sized for expected concurrent demand.
- Define fallback regions and a retry delay for capacity errors.
- Confirm logs and debugging access work under load.
- Upload the final server image early, so the build you tested is the build that runs.
On launch day:
- Watch allocation time and failed session starts, not only concurrent players.
- Watch demand by region, and add regions where players show up.
- Keep a hotfix path ready. On Gameye, a new image rolls out to every region and provider within a few minutes, and a rollback points at the previous image.
After the peak:
- Shrink warm pools and let overflow capacity go.
- Compare real demand by region against the forecast, and use it for the next launch.
How Gameye handles a launch spike
Your matchmaker calls POST /session with a region and a Docker image. Gameye places the session and returns an IP address and port. There is no Gameye SDK in your server.
- Placement. Bare metal first, cloud only when bare metal in the region is exhausted.
- Hardware. Dedicated bare metal averaging 4.7 GHz across the fleet, using AMD Zen 4 and Zen 5 where available.
- Pricing. $0.07 per vCPU-hour on demand and $0.035 reserved, billed per second, with no egress fees.
- Reliability. A 99.7% contractual uptime SLA, and 99.9% measured uptime over the trailing 12 months.
See how it works for the session flow and pricing for the full rate card.
Summary
- Pre-warm the first wave with warm pools, then start one server per match on demand.
- Plan for demand above forecast and in regions you did not expect.
- Load test the full path from matchmaker to running server, with sudden spikes.
- Spread capacity across providers with automatic failover, and give your matchmaker a retry delay and fallback regions.
To test your launch setup, start a free trial and run a match on Gameye, or talk to an engineer about your launch plan.