VFX · Tools
Why we replaced our render farm orchestrator this year
From Deadline to a custom in-house scheduler — what broke, what works, what's next.
01 · Why we left Deadline
Eight years on Deadline. We loved it. We rewrote our scheduler anyway. The reason was not Deadline — it was that we couldn't talk to it the way we wanted to. Our shot database had grown opinions Deadline couldn't represent.
02 · What we built
A custom orchestrator in Go that reads our shot database, prioritises by deadline, owner, and dependency, and dispatches to a pool of ephemeral render workers on AWS Spot. Four engineers, three months, one production incident along the way.
03 · What broke
Cold start times on Spot were brutal. A 90-second cold start on a 20-second render is comedy. We solved it by keeping a small warm pool of ten always-on instances plus a prebaked AMI with all plugins installed.
"Less magic, more dashboards."
— Daniel Cho, Pipeline TD
04 · What's next
Per-shot cost reporting back to the producer. Real-time priority changes from the crit room. Less magic, more dashboards.
