On the evening the Forever beta opened on September 17, 2026, players kept getting kicked off, quests took a minute to open, and the server had to be restarted. Senior producer Tom Ellis broke down all three failures — and none of them were about the game itself.

On the evening the Forever beta opened on September 17, 2026, things got slippery: players were getting kicked off, quests and loot took a full minute to open, and the server had to be restarted with progress lost. Blizzard published a breakdown of what went wrong — and not a single complaint was about the game itself.

Where it all runs

The beta lives on the PTR stand, not on live. There's only one of them, and it's regional. Historically, WoW betas have put almost no load on the infrastructure, and nobody sized the hardware for them seriously. That was the assumption going into the Forever launch.

That's where all three of that evening's problems came from.

First: mass disconnects

They searched for half an hour — and there was nothing to find. CPU and memory across all services were fine. The Battle.net Game Service team saw the same on their side. The culprit turned out to be a load counter that protects BGS from going down: a small regional stand was pushing a huge stream of logins through two connections, and the arithmetic the counter uses to decide when to panic was breaking.

Connections were bumped to eight — disconnects stopped. According to senior producer Tom Ellis, the counter triggered for the first time without any accompanying CPU load. His thread on X caught on with the line '20 years, always something new'.

Second: lag on looting and turning in quests

Loot opened slowly, quests took a long time to pick up. The culprit was the PTR database: new tables and a heavy stream of inserts left the query planner with stale assumptions about the data. Oracle engineers ran statistics manually and turned on automatic recalculation. Speed came back immediately.

Third: the restart

A few hours in, everyone had to be kicked off. The WORLD pools, which run the actual game, were burning CPU and hitting memory limits, while the OOM killer was already taking out the first virtual machines — risking taking the hypervisor down with them.

Classic engineers found the cause: empty maps weren't shutting down properly, and the servers were slowly eating their own resources. The fix passed testing but required a restart. They also added backup WORLD pools and a second instance of the regional service. After that, the night went quietly.

Why the load wasn't anticipated

The stand was sized based on past WoW betas. But what showed up was launch-day audience — numbers like that haven't been seen in a long time. Lead designer Josh Greenfield (Aggrend) wrote on X: 'in 17 years at Blizzard and almost 8 on the Classic team, I've never seen numbers like this.'

What this means for November 4

Directly — nothing. The live stand has nothing to do with the beta, and none of the three failures says anything about the game's readiness. Indirectly — it says something: demand turned out to be higher than Blizzard's internal forecasts, which means launch queues are worth planning for.

In practice

• The beta failures were about the PTR test stand, not about the game itself; by November 4 that's no longer relevant.
• If there are queues at launch — it's the same story as September 17 at the beta: demand higher than Blizzard expected.
• If you get disconnected from your character in the first hours of November 4 — that's no longer a 'broken launch', it's the normal story of server oversaturation, like in past WoW launches.