Discord music bot: what uses the RAM

A music bot does more than answer commands: it starts external processes, keeps a queue in memory and decodes audio continuously. On a free plan with 315 MB of RAM, every process counts. Here is what uses memory, how to measure it and what to reduce first.

• A music bot starts several processes: the bot, ffmpeg and sometimes yt-dlp.
• The queue weighs little as long as it only holds titles and identifiers.
• Memory that never comes back down points to a leak, to fix before buying more RAM.

The processes that use memory

Each playback can start several processes: the bot itself, ffmpeg to decode and convert the sound, and yt-dlp to fetch the stream when the source is a web address. Each one has its own memory.

• The bot process: your Node.js or Python code and the libraries loaded at startup.
• ffmpeg: decodes and converts the sound. Each server playing music adds its share.
• yt-dlp: called to extract a stream from an address, it only lives while the fetch runs.

The queue and the data

The queue lives in memory. A list of titles and addresses weighs little. What really weighs is the data fetched and kept: large metadata, images, playback buffers that are never released.

• Keep only titles and identifiers in the queue, not whole audio files.
• Release the buffers and playback objects when a track ends.
• Limit the queue size, for example to a few dozen tracks per server.

Measure and reduce

Measure during real playback, then an hour later. Memory that keeps rising and never comes back down points to a leak. Also check for ffmpeg processes still running after a track ends.

• Note the usage during playback, then after an hour of listening, and compare.
• An ffmpeg left running after a track uses memory for nothing: stop it explicitly.
• Usage that never comes back down is a leak: fix it before changing plan.

Why ACLClouds for a music bot

ffmpeg, yt-dlp and deno are already installed in the Node.js and Python images: nothing to compile. A music bot that goes beyond 315 MB can move to a plan with more memory.

• Free plan: 315 MB of RAM and 0.25 vCPU, to test a light music bot.
• ffmpeg, yt-dlp and deno included in the Node.js and Python images.
• For more memory, the Linux Essentiel VPS gives 16 GB of RAM, 4 vCPU and 80 GB of storage, at 9.99 € per month.