SSH command within a script terminates prematurely

Question

From myhost.mydomain.com, I start a nc listener. Then login to another host to start a netcat push to my host:

nc -l 9999 > data.gz &
ssh repo.mydomain.com "cat /path/to/file.gz | nc myhost.mydomain.com 9999"

These two commands are part of a script. Only 32K bytes are sent to the host and the ssh command terminates, the nc listener gets an EOF and it terminates as well.

When I run the ssh command on the command line (i.e. not as part of the script) on myhost.mydomain.com the complete file is downloaded. What's going on?

Note: You can achieve the same with ssh repo.mydomain.com "cat /path/to/file.gz" > data.gz Maybe your script is more complex but maybe someone overlooked the simple solution. — Aaron Digulla
– Aaron Digulla, Commented Aug 11, 2015 at 11:41

Aaron Digulla · Accepted Answer · 2015-08-11 11:45:52Z

1

I think there is something else that happens in your script which causes this effect. For example, if you run the second command in the background as well and terminate the script, your OS might kill the background commands during script cleanup.

Also look for set -o pipebreak which terminates all the commands in a pipeline when one of them returns with != 0.

On a second note, the approach looks overly complex to me. Try to reduce it to

ssh repo.mydomain.com "cat /path/to/file.gz" > data.gz

(ssh connects stdout of the remote with the local). It's more clear when you write it like this:

ssh > data.gz repo.mydomain.com "cat /path/to/file.gz"

That way, you can get rid of nc. As far as I know, nc is synchronous, so the second invocation (which sends the data) should only return after all the data has been sent and flushed.

answered Aug 11, 2015 at 11:45

Aaron Digulla

330k111 gold badges626 silver badges840 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

dr.house Over a year ago

I can't stream all the data via ssh (or scp it) because of the low compute capacity available on the sender. It is faster (at least 4-5x in my case) to simply netcat it. There's no pipebreak as well. I tweaked the second command to be: ssh repo.mydomain.com "nc myhost.mydomain.com 9999 < /path/to/file.gz". Same observation. Only that this time 4K bytes were sent. This behavior is independent of the file

Aaron Digulla Over a year ago

The 4KB is an indication that nc could only fill the send buffer (which is 4KB) once before it was stopped. Try to wrap nc in a script which runs date 1>&2 (connects stdout with stderr) before and after the command just to see how long it runs. Also try nc -v; that might give you more information how many bytes are written/received. As for speed: The speedup probably happens because ssh encrypts the connection. See serverfault.com/questions/116875/… for ways to optimize

timoiste · Accepted Answer · 2024-07-08 08:05:01Z

0

I had a similar problem and in the end found out the the login-daemon's configuration was set to termninate all processes on logout.

You may system daemon login configuration for KillUserProcesses (on exit) This overrides nohup, disown a.s.o

The file is called /etc/systemd/logind.conf

# See logind.conf(5) for details.
[Login]
#NAutoVTs=6
#ReserveVT=6
KillUserProcesses=no

The KillUserProcesses - Option was set to yes in my case by default.

answered Jul 8, 2024 at 8:05

timoiste

1

Collectives™ on Stack Overflow

SSH command within a script terminates prematurely

2 Answers 2

2 Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

2 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related