Did you configure any long-running post observation scripts? This could trigger the issue as follows, this was the case on my station:
- Observation 1 ends, long running post observation script starts
- Observation 2 starts while post-ob from 1 is still running
- Obs 2 is written to tmpfs, but Obs 1 is not cleaned up yet
- tmpfs is filled up
- jobs database on tmpfs cannot be written to any more
- Obs 2 is not stopped because jobs database cannot be accessed, and will run indefinitely,
- Obs 3 is started, but cannot claim usb since Obs 2 is still running.
A lot of steps, but this is what happened on my system. If you defer a long-running post-ob to the background by daemonizing or forking it will be fixed. Would you by coincidence be running my meteor processing script: do a git pull and it will do just that. 