marathonctl logs

[ashwanthkumar]

This issue contains my thoughts on how it must be implemented.

Usage

$ marathonctl logs -h
Tail the logs of an application deployed via Marathon.

Usage:
  marathonctl logs <app-id> [flags]

Flags:
  -o, --stdout       Tail the stdout file from the current executor directory
  -e, --stderr        Tail the stderr file from the current executor directory
  -l, --log string    Log file with relative to the current executor directory. (Default: stdout)

Implementation

1. Query the Marthon's /apps/<app-name> endpoint to get the list of active tasks for the app.
2. Build a unique list of host from the list of tasks
3. Query each of the mesos-slave host's /slave(1)/state.json to get the corresponding current working directory for each of the task.
4. Now use that along with the log file name on the /files/read.json?path=$CWD/$LOG_FILE.log on mesos slaves to tail the logs.

We might want to fan out the 4th operation into goroutines on a 1 sec ticket.

Questions

• How do we detect a failed / killed / finished task and remove them from the list of goroutines.
• Can mesos slaves handle so much load while tailing logs for a huge number of instances?