One of our business apps is deployed on a Managed VM in AppEngine with the Go runtime. The app itself is pretty lightweight and hence we have a 0.5 CPU and 1GB RAM instance requested in app.yaml. A surprise to us, we started noticing frequent reboots of the the VM causing quite some downtime. After some analysis, we found that docker daemon was ending up consuming ~80% of RAM after 8-10 hours of uptime, making the system unresponsive (including ah health checks). This made GAE treat the host unhealthy and restarted the VM.
We tried increasing the instance RAM size to see if docker daemon is leaking or its just that it needs ~800M to work. As expected, after few days of uptime docker daemon is now consuming ~80% of 3GB of RAM (around 2.7G resident, screenshot attached).
which suggests docker 1.7 contains fixes.