question/request: reuse same SSH connection

[ccoenen]

I have a situation where I routinely pull or push from dozens of separate repositories on the same server. Oftentimes, only a few of the repositories have any changes at all. MOST of my time is spent doing SSH setup.

Is there any way to keep the underlying ssh connection open while we're going from repo to repo? I think I could speed up my script by a whole lot. As an added benefit, I would also look less like a DDoS-attack to my repository server.

[steveukx]

Hi, looks like you can use the GIT_SSH_COMMAND environment variable to do something like:

export GIT_SSH_COMMAND='ssh -o ControlMaster=auto -o ControlPersist=10m -o ControlPath=~/.ssh/cm-%r@%h:%p'

(disclaimer - I've not tested this).

If that's working for you, you can include environment variables in the use of simple-git by following the environment variables guide

[ccoenen]

I tried this:

const GIT_SSH_COMMAND = 'ssh -o ControlMaster=auto -o ControlPersist=10m -o ControlPath=~/.ssh/controlmaster-%r@%h:%p';

await simpleGit({ baseDir: localPath })
	.env({...process.env, GIT_SSH_COMMAND})
	.pull();

Pulling 30 repos from the same server, all 30 of them are already up to date before the first run (efficiency-worst-case, which shows the biggest relative difference):

No ControlMaster

Run 1
real	0m22,178s
user	0m0,952s
sys	0m0,693s

Run 2
real	0m19,294s
user	0m0,819s
sys	0m0,546s

Wireshark counts 1471 packets, for 31 TCP connections

Including ControlMaster

Run 3
real	0m8,132s
user	0m0,574s
sys	0m0,471s

Run 4
real	0m8,698s
user	0m0,611s
sys	0m0,521s

Wireshark counts 460 packets, for 2 TCP connections

Not bad :-)

Thanks!

[ccoenen]

This fixes another interesting weirdness in my personal script: if this is loooooong-running, my ssh-agent would time out at some point, and subsequent pulls would then fail the authentication. With the persistent connection, if the first pull works, I can just walk away. :-D