Signed-off-by: Jan Dubois <[email protected]>

…ss fails

To avoid terminating early and killing a still running drain script it is
necessary to keep the container-run process running even if the main
process has already exited.

Once container-run receives the SIGTERM signal it will pass it on to the
child processes and then wait until all direct children have exited before
exiting itself. This should allow the processes to terminate cleanly as
long as the grace period has not yet expired.

container-run will also terminate immediately if the main process terminates
with an error. This allows the container to fail and be restarted by k8s.

Signed-off-by: Jan Dubois <[email protected]>

This is used to trigger graceful shutdown of nginx.

Signed-off-by: Jan Dubois <[email protected]>

…ning

Can be used to wait for a process to stop after sending the STOP command.

Signed-off-by: Jan Dubois <[email protected]>

It only supports the "start" and "stop" commands from the real bpm,
but has additional "quit" and "running" commands to help reimplementing
drain scripts that make use of the pid files.

Signed-off-by: Jan Dubois <[email protected]>

Signed-off-by: Jan Dubois <[email protected]>

It is important to process errors synchronously because their handling
depends on the value of the 'active' flag. To avoid deadlocks any code
sending errors to the channel must run in a blockable goroutine.

Signed-off-by: Jan Dubois <[email protected]>

(and give them a chance to notify their child processes). If
the processes haven't terminated in 20s, send SIGKILL.

Signed-off-by: Jan Dubois <[email protected]>

Signed-off-by: Jan Dubois <[email protected]>

Signed-off-by: Jan Dubois <[email protected]>

That way it also starts the timeout trigger if the process
doesn't stop within 20s.

Signed-off-by: Jan Dubois <[email protected]>

The rest are just confusing the debug output, and passing
on e.g. SIGCHLD to the child processes is neither correct
nor useful.

Signed-off-by: Jan Dubois <[email protected]>

container-run should send a SIGKILL if the process is still alive
after 20s, so if it isn't gone after 30, then it probably will not
stop.

The `bpm running` subcommand will also now print "yes" or "no" in
addition to setting the exit code.

Signed-off-by: Jan Dubois <[email protected]>

(but they no longer pass because they have not been updated
with the new application logic).

Signed-off-by: Jan Dubois <[email protected]>

Also modify `bpm running` to only print yes/no when stdout is a tty.

Signed-off-by: Jan Dubois <[email protected]>

Signed-off-by: Jan Dubois <[email protected]>

The errors channel is unbuffered, so stopProcesses cannot deliver
errors if it is running from the same goroutine as the receiver.

Signed-off-by: Jan Dubois <[email protected]>

Signed-off-by: Jan Dubois <[email protected]>