ssh 'cd X && cmd && nohup setsid job &' backgrounds the whole && list: $! is a subshell, not the job

Pattern: ssh host 'cd /dir && chmod +x run.sh && nohup setsid ./run.sh > out 2>&1 < /dev/null & echo $! > run.pid'.

Gotcha: in sh, & binds looser than &&, so the WHOLE cd && chmod && nohup setsid ./run.sh list is backgrounded as one subshell. $! is that subshell's PID (it shows as bash -c ...), not the job's. setsid then puts the real script in a new session/pgid. Killing the recorded PID won't stop the job. Also the ssh call can hang until your client times out, because the backgrounded subshell still holds the session.

Fix: run the setup steps first, then background only the job, with braces or a separate line: cd /dir && chmod +x run.sh; nohup setsid ./run.sh >out 2>&1 </dev/null & echo $! > run.pid. Even then, with setsid the job's PID can differ from $! (setsid forks when the caller is a process-group leader). It is more reliable to have the script write its own PID: echo $$ > run.pid as its first line.

If you already have a wrong PID, find the real one by parent instead of by name pattern: ps -o pid,pgid,sid,cmd --ppid <recorded pid>. Then kill the job with kill -TERM -- -<pgid>. Stopping the local ssh client afterwards didn't kill the setsid'd job (it's in its own session, under nohup).