Two transcodes at a time, and the bill is the box

The web desk's export endpoint takes a master and returns three renditions. Every export is a full ffmpeg pass, and on a busy news day thirty arrive in ten minutes.

The handler is a hoody-exec script with one line at the top: // @concurrent 2. The runtime holds the cap: at most two exports run in parallel, so the newsroom's other containers keep their cores. // @timeout 20m keeps a long feature from being cut off by the default hour. No worker pool, no semaphore in the code: the concurrency limit is a comment the runtime enforces.

On a per-invocation platform, thirty transcodes are thirty bills and a fight with the execution ceiling. Here they are thirty requests to a box you already rent, and the thirty-first costs the same as the first.

The render cap is a comment, and the bill is the box.