Process management

[at15]

The benchmark need to have the ability to start and stop process

• attach perf to worker process when job is submitted
• detach/stop (ctrl+c) perf when job is finished
• start and stop hadoop process programmatically (so I don't need to ssh into every server/vm to start /stop hadoop master and/or slave

TODO MUST

• [ ] find pid file for data node, name node etc.
• [ ] use jps to find
  • [ ] spark executor
  • [ ] hadoop YarnChild
• [ ] start perf and attach perf stat to it
• [ ] assign unique id for each job
• [ ] start hibench programmantically
• [ ] collect useful system data
• [ ] how to down grade to single node
  • [ ] single node == master + slave on the same node?

It is also possible to use ssh instead of having multiple reset client

• https://www.zaiste.net/2014/08/executing_commands_via_ssh_using_go/
• https://gist.github.com/josephspurrier/e83bcdbf9e6865500004
• https://github.com/hypersleep/easyssh/blob/master/easyssh.go

TODO Fancy

• [ ] web shell
• [ ] grafana GUI for node health etc.
• [ ] flame graph
• [ ] attach etc.

Ref

• https://github.com/topfreegames/apm (simply use os.exec is not enough, you can't obtain pid and need to redirect output)
  • https://github.com/sevlyar/go-daemon (seems to be a fork and exec in go?)
• https://github.com/shirou/gopsutil
• https://github.com/yudai/gotty share terminal as web application
• https://github.com/mitchellh/go-ps