fix: Write os-var replaced .src file to temp file, and rename

[chsukivra]

The replace_os_vars function is called every time for any command, not only those that boot the release. If you call <release> status in a readiness check, that will re-write the file.

This might lead to race conditions, where the release starts up, the script creates the replace_os_vars function, is slowed down, but the health check is invoked and it will re-render the sys.config and write it halfway while the starting release gets around to read that very file.

We do see intermittent examples of this, where one container out of a bunch on a slow machine start up, and a container that has no changes from previous successful runs will fail with and error like this, with same sys.config.src files, same env.

could not start kernel pid (application_controller) (error in config file \"/tmp/sys.config\" (none): no ending <dot> found)

However this container has a healthcheck implemented with the status command running in it.

    healthcheck:
      test: [ "CMD", "/opt/forms/bin/forms", "status" ]

There are many ways to attack this issue. Just opening with a suggestion.

[tsloughter]

Neat edge case. My only concern is whats the availability of mktemp?

[chsukivra]

I've also been thinking about introducing an environment variable like RELX_EXPAND_SRC=false that completely disable this generator unless explicitly asked for. If not set or set to true it will keep current behavior.

#!/bin/sh
/opt/myapp/bin/myapp expand_src
exec /opt/myapp/bin/myapp foreground

Its just that it doesnt do the right thing out of the box and that could be annoying.

Also, maybe i can make a bigger change that chose to only generate these files for the commands that will actually start the service? I dont see why status should lead to it being expanded.