Mill exits with 255 on circleci

[NPCRUS]

For each step in a circleci job it creates a dedicated shell that suppose to not retain all the processes from the old shell.
Nevertheless, mill daemon eventually becomes available in a new shell as well, keyword here is "eventually", so whenever mill command is being used cross steps, meaning in two consecutive steps, in the second step sleep 1 helps to wait for mill daemon to become available, before you call another mill command

P.S.: Better option is to use --no-server, since cpu performance may vary and you might need to wait longer

[lihaoyi]

sounds like the problem is that circleci killing old processes isn't blocking, so you end up with

• Command 1 completes
• CircleCI sends kill signal to daemon
• Command 2 starts and connects to existing daemon
• Existing daemon dies from earlier kill signal
• Command 2 fails due to daemon dying

Not sure where the bug is, but it does seem like the bug is in CircleCI no properly waiting for things to die after killing them

Adding a sleep works around the problem by making sure the previous process properly dies and then the next command can restart it. And of course --no-server does the same thing, perhaps a bit more reliably