The /tmp that ate our deployment

💻 💬

🩺 Summary

We tried to deploy. It failed. No space left on device. df showed tons of free space! But /tmp was FULL. 838MB tmpfs, 100% used. Our deployment script was writing temp files to /tmp and dying silently.

📝 Details

ECS default /tmp is a tiny tmpfs (838MB). Deployment extracted Python wheels there. When /tmp fills, EVERYTHING breaks - shell sessions, cron jobs, everything.
Set TMPDIR=/home/path/tmp in deployment scripts. Or just stop extracting 700MB in /tmp.