I run Byobu (backed by Tmux) on a remote machine. I have a basic window layout which is what I generally want, and I've configured Byobu to launch at login so when I ssh to the remote machine I resume my session and when I want to leave I F6 to detach and disconnect. This works well.
My problem is when I reboot the machine all my Byobu windows are erased. I'm not trying (yet?) to resume exactly where I left off, but just getting all the windows with their names would be a big help.
I can use .byobu/window.tmux to have a window layout that's executed when I connect. But I've tried this
new-session 'remote';
new-window -n irc;
new-window -n code;
new-window -n logs;
and that causes Byobu to create a new session each time I connect, so I can't resume anymore.
Then I tried adding the -A flag to attach:
new-session -A 'remote';
new-window -n irc;
new-window -n code;
new-window -n logs;
This goes back to resuming the old session, but now every time I connect I get an additional three windows added to the windows I had before.
Here's pseudo-code for what I want:
if session_exists('remote') {
attach('remote');
} else {
new-session 'remote';
new-window -n irc;
new-window -n code;
new-window -n logs;
}
Can I do this?
I found this article where the guy explains how to create one main, unattached session and then create a shared session from that one.
This results in the behavior you're aiming for, I believe.
Check out this article:
byobu-tmux tabbed windows
I´m a happy user of byobu, but recently I noted that I cannot attach anymore to open sessions.
With ps aux | grep tmux I can clearly see many tmux processes, but unfortunatly, when I try to attach with:
tmux attach
byobu attach
I get no results but a no session error. Moreover, with byobu-select-session I got a failed to connect to server instead.
There is a commant to connect tmux to a given socket, which I found using
lsof -U | grep '^tmux'. But still no session attached. My session files are in /tmp/user/tmux-1000/default, but I can see some sockets being used.
From ps aux I can see that byobu launches tmux with: tmux -2 -f /usr/share/byobu/profiles/tmuxrc new-session -n - /usr/bin/byobu-shell
Unfortunately, either with byobu -S path or byobu -L socketname I am not able to attach to previously open session, and byobu simply start a new session.
I run into a similar situation caused by accidentally removing the tmux socket in /tmp. The method described here solved the problem for me (either killall -SIGUSR1 tmux or kill -USR1 $PID_FOR_RUNNING_TMUX).
Main question
I do a lot of interactive work in R using SSH and tmux. I connect using the -Y flag, e.g.
ssh -Y user#server
Then, I typically start a new tmux session:
tmux new -s name
When I first connect, my plots work fine. However, my connection often gets interrupted or I might disconnect for lunch. For this case, I have a line in my .bashrc that saves my current display:
if [ ! -f ~/.my.display ]; then
touch ~/.my.display
fi
display=$(echo $DISPLAY)
if [[ $display != "" ]]; then
echo "export DISPLAY="$DISPLAY > ~/.my.display
fi
source ~/.my.display
After disconnecting, when I tmux into my interactive R session, I typically need to reset my X11 window:
# reconnect to tmux session
ssh -Y user#server
tmux a -t name
# reset x11 window
graphics.off()
display = gsub(".*DISPLAY.", "", tail(readLines("~/.my.display"), n = 1))
x11(display, type = "Xlib")
This works most of the time. However, around 25% of the time, my graphics.off() call hangs for 30 minutes, which is incredibly frustrating. I often need to kill my session, losing all of my work.
Is there a way to call graphics.off() without it hanging for so long? Ideally, it would try this command for like 10 seconds, then throw an error if there is a problem.
Bonus
Within tmux, I often connect to other user accounts using SSH -Y. Therefore, the sequence of commands is:
# connect laptop -> user1
ssh -Y user1#server1
tmux a -t name
# connect user1 -> user2
ssh -Y user2#server2
I am having trouble making interactive plots with user2's account. Right now, my solution is to save user2's DISPLAY variable to their .my.display. Before plotting with user2 through tmux, I first SSH directly into user2's account, which properly properly configures their .my.display file. Then, I follow the same steps as before: graphics.off() and starting a new x11 window.
This seems unnecessarily complicated, so my second question is if there is a cleaner way to set this up?
I am using XQuartz 2.7.11 on Mac OS 10.12.6
I am trying to find a nice way to restore the SSH agent when I reconnect a disconnected tmux session.
The cause seems to be that the SSH agent session changes but the environment variable from the tmux session is not updated.
How can I automate this, before attaching the session itself? Because the session I am attaching to does not always have a bash prompt, so I cannot afford to type something inside it. It has to be something to run before creating or attaching the tmux session.
An example of the code I'm running is at https://gist.github.com/ssbarnea/8646491 -- a small ssh wrapper that is using tmux to create persistem ssh connections. This works quite well, but sometimes the ssh agent stops working so I am no longer able to use it to connect to other hosts.
There's an excellent gist by Martijn Vermaat, which addresses your problem in great depth, although it is intended for screen users, so I'm adjusting it for tmux here.
To summarize:
create ~/.ssh/rc if it doesn't exist yet, and add the following content:
#!/bin/bash
# Fix SSH auth socket location so agent forwarding works with tmux.
if test "$SSH_AUTH_SOCK" ; then
ln -sf $SSH_AUTH_SOCK ~/.ssh/ssh_auth_sock
fi
Make it work in tmux, add this to your ~/.tmux.conf:
# fix ssh agent when tmux is detached
setenv -g SSH_AUTH_SOCK $HOME/.ssh/ssh_auth_sock
Extra work is required if you want to enable X11 forwarding, see the gist.
While tmux updates SSH variables by default, there is no need to
change/add socket path
change the SSH_AUTH_SOCKET variable
I like the solution by Chris Down which I changed to add function
fixssh() {
eval $(tmux show-env \
|sed -n 's/^\(SSH_[^=]*\)=\(.*\)/export \1="\2"/p')
}
into ~/.bashrc. Call fixssh after attaching session or before ssh/scp/rsync.
Newer versions of tmux support -s option for show-env, so only
eval $(tmux show-env -s |grep '^SSH_')
is possible.
Here's what I use for updating SSH_AUTH_SOCK inside a tmux window (based on Hans Ginzel's script):
alias fixssh='eval $(tmux showenv -s SSH_AUTH_SOCK)'
Or for tmux that does not have showenv -s:
alias fixssh='export $(tmux showenv SSH_AUTH_SOCK)'
Here is my solution which includes both approaches, and does not require extra typing when I reconnect to tmux session
alias ssh='[ -n "$TMUX" ] && eval $(tmux showenv -s SSH_AUTH_SOCK); /usr/bin/ssh'
There are lots of good answers here. But there are cases where tmux show-environment doesn't see SSH_AUTH_SOCK. In that case you can use find to locate it explicitly.
export SSH_AUTH_SOCK=$(find /tmp -path '*/ssh-*' -name 'agent*' -uid $(id -u) 2>/dev/null | tail -n1)
That's long and complicated, so I'll break it down...
01 export SSH_AUTH_SOCK=$(
02 find /tmp \
03 -path '*/ssh-*'
04 -name 'agent*'
05 -uid $(id -u)
06 2>/dev/null
07 | tail -n1
08 )
export the SSH_AUTH_SOCK environment variable set to the output of the $() command substitution
find files starting in /tmp
limit results to only those with /ssh- in the path
limit results to only those whose name begins with agent
limit results to only those with a user id matching the current user
silence all (permissions, etc.) errors
take only the last result if there are multiple
You may be able to leave off 6 & 7 if you know that there will only be 1 result and you don't care about stderr garbage.
I use a variation of the previous answers:
eval "export $(tmux show-environment -g SSH_AUTH_SOCK)"
assuming that you did the ssh agent started from the outer environment. Same goes for other environment variables such as DISPLAY.
I prefer to avoid configuring TMUX (etc) and keep everything purely in ~/.ssh/. On the remote system:
Create ~/.ssh/rc:
#!/bin/bash
# Fix SSH auth socket location so agent forwarding works within tmux
if test "$SSH_AUTH_SOCK" ; then
ln -sf $SSH_AUTH_SOCK ~/.ssh/ssh_auth_sock
fi
Add following to ~/.ssh/config so it no longer relies on $SSH_AUTH_SOCK, which goes stale in detached terminals:
Host *
IdentityAgent ~/.ssh/ssh_auth_sock
Known limitations
ssh-add doesn't use ~/.ssh/config and so cannot communicate with ssh-agent. Commands like ssh-add -l produce errors, even though ssh user#host works fine, as does updating git remotes which are accessed via SSH.
I may have worked out a solution that is fully encapsulated in the ~/.tmux.conf configuration file. It is a different approach than modifying the ~/.bash_profile and ~/.ssh/rc.
Solution only using ~/.tmux.conf
Just cut and paste the following code into your ~/.tmux.conf
# ~/.tmux.conf
# SSH agent forwarding
#
# Ensure that SSH-Agent forwarding will work when re-attaching to the tmux
# session from a different SSH connection (after a dropped connection).
# This code will run upon tmux create, tmux attach, or config reload.
#
# If there is an SSH_AUTH_SOCK originally defined:
# 1) Remove all SSH related env var names from update-environment.
# Without this, setenv cannot override variables such as SSH_AUTH_SOCK.
# Verify update-environment with: tmux show-option -g update-environment
# 2) Force-set SSH_AUTH_SOCK to be a known location
# /tmp/ssh_auth_sock_tmux
# 3) Force-create a link of the first found ssh-agent socket at the known location
if-shell '[ -n $SSH_AUTH_SOCK ]' " \
set-option -sg update-environment \"DISPLAY WINDOWID XAUTHORITY\"; \
setenv -g SSH_AUTH_SOCK /tmp/ssh_auth_sock_tmux; \
run-shell \"ln -sf $(find /tmp/ssh-* -type s -readable | head -n 1) /tmp/ssh_auth_sock_tmux\" \
"
Caveat
The above solution along with the other solutions are susceptible to a race condition when initiating multiple connections to the same machine. Consider this:
Client 1 Connect: SSH to machineX, start/attach tmux (writes ssh_auth_sock link)
Client 2 Connect: SSH to machineX, start/attach tmux (overwrites ssh_auth_sock link)
Client 2 Disconnect: Client 1 is left with a stale ssh_auth_sock link, thus breaking ssh-agent
However, this solution is slightly more resilient because it only overwrites the ssh_auth_sock link upon tmux start/attach, instead of upon initialization of a bash shell ~/.bash_profile or ssh connection ~/.ssh/rc
To cover this last race condition, one may add a key binding to reload the tmux configuration with a (Ctrl-b r) key sequence.
# ~/.tmux.conf
# reload config file
bind r source-file ~/.tmux.conf
From within an active tmux session, executing this sequence when the ssh_auth_sock link goes stale will refresh the ssh-agent connection.
In case other fish shell users are wondering how to deal with this when using fish (as well as for my future self!). In my fish_prompt I added a call to the following function:
function _update_tmux_ssh
if set -q TMUX
eval (tmux show-environment SSH_AUTH_SOCK | sed 's/\=/ /' | sed 's/^/set /')
end
end
I suppose that more advanced *nix users would know how to replace sed with something better, but this works (tmux 3.0, fish 3.1).
Following up on #pymkin's answer above, add the following, which worked with tmux 3.2a on macOS 11.5.3:
To ~/.tmux.conf:
# first, unset update-environment[SSH_AUTH_SOCK] (idx 3), to prevent
# the client overriding the global value
set-option -g -u update-environment[3]
# And set the global value to our static symlink'd path:
set-environment -g SSH_AUTH_SOCK $HOME/.ssh/ssh_auth_sock
To ~/.ssh/rc:
#!/bin/sh
# On SSH connection, create stable auth socket path for Tmux usage
if test "$SSH_AUTH_SOCK"; then
ln -sf "$SSH_AUTH_SOCK" ~/.ssh/ssh_auth_sock
fi
What's going on? Tmux has the semi-helpful update-environment variable/feature to pick up certain environment variables when a client connects. I.e. when you do tmux new or tmux attach, it'll update the tmux environment from when you ran those commands. That's nice for new shells or commands you run inside tmux afterwards, but it doesn't help those shells you've started prior to the latest attach. To solve this, you could use some of the other answers here to have existing shells pick up this updated environment, but that's not the route I chose.
Instead, we're setting a static value for SSH_AUTH_SOCK inside tmux, which will be ~/.ssh/ssh_auth_sock. All shells inside tmux would pick that up, and never have to be updated later. Then, we configure ssh so that, upon connection, it updates that static path with a symlink to the latest real socket that ssh knows.
The missing piece from #pymkin's answer is that Tmux will have the session value override the global value, so doing set-environment -g isn't sufficient; it gets squashed whenever you re-attach. You also have to also tell tmux not to update SSH_AUTH_SOCK in the session environment, so that the global value can make it through. That's what the set-option -g -u is about.
After coming across so many suggestions, I finally figured out a solution that enables TMUX update the stale ssh agent after being attached. Basically, both the zshrc files on the local and remote machines need to be modified.
Insert the following codes into the local zshrc, which is based on this reference.
export SSH_AUTH_SOCK=~/.ssh/ssh-agent.$(hostname).sock
ssh-add -l 2>/dev/null >/dev/null
# The error of executing ssh-add command denotes a valid agent does not
# exist.
if [ $? -ge 1 ]; then
# remove the socket if it exists
if [ -S "${SSH_AUTH_SOCK}" ]; then
rm "${SSH_AUTH_SOCK}"
fi
ssh-agent -a "${SSH_AUTH_SOCK}" >/dev/null
# one week life time
ssh-add -t 1W path-to-private-rsa-file
fi
Insert the following code into the remote zshrc, where the tmux session will be attached.
alias fixssh='eval $(tmux showenv -s SSH_AUTH_SOCK)'
Then ssh into the remote machine. The -A option is necessary.
ssh -A username#hostname
Attach the TMUX session. Check the TMUX evironment variables
# run this command in the shell
tmux showenv -s
# or run this command after prefix CTRL+A or CTRL+B
:show-environment
Run fixssh in the previously existed panes to update the ssh agent. If a new pane is created, it will automatically get the new ssh-agent.
Here's another simple Bash solution, using PROMPT_COMMAND to update the SSH_* vars inside tmux before each prompt is generated. The downside to this solution is that it doesn't take effect in existing shells until a new prompt is generated, because PROMPT_COMMAND is only run before creating new prompts.
Just add this to your ~/.bashrc:
update_tmux_env () {
# Only run for shells inside a tmux session.
if [[ -n "$TMUX" ]]; then
eval $(tmux show-env -s | grep '^SSH_')
fi
}
export PROMPT_COMMAND=update_tmux_env
Here's a new fix to an old problem: I think it's simpler than the other fixes and there's no need to make a static socket or mess with the shell prompt or make a separate command you have to remember to run.
I added this code added to my .bashrc file:
if [[ -n $TMUX ]]; then
_fix_ssh_agent_in_tmux () { if [[ ! -S $SSH_AUTH_SOCK ]]; then eval export $(tmux show-env | grep SSH_AUTH_SOCK); fi }
ssh () { _fix_ssh_agent_in_tmux; command ssh $#; }
scp () { _fix_ssh_agent_in_tmux; command scp $#; }
git () { _fix_ssh_agent_in_tmux; command git $#; }
rsync () { _fix_ssh_agent_in_tmux; command rsync $#; }
fi
If the shell is running within tmux, it redefines 'ssh' and its ilk to bash functions which test and fix SSH_AUTH_SOCK before actually running the real commands.
Note that tmux show-env -g also returns a value for SSH_AUTH_SOCK but that one is stale, I assume it's from whenever the tmux server started. The command above queries the current tmux session's environment which seems to be correct.
I'm using tmux 2.6 (ships with with Ubuntu 18.04) and it seems to work well.
I have two unix machines, both running AIX 5.3
My $HOME is mounted on machine1.
Using NFS, login machine2 will go to the same $HOME
I login machine2 first, then machine1.
Both using telnet.
The 2 sessions will share the same .sh_history file.
I found out that the fc -l behavior very strange.
In machine2, I issue the commands in telnet:
fc -l
ksh fc -l
Both give the same output.
In machine1,
fc -l
ksh fc -l
give DIFFERENT results
The result for ksh fc -l
is the same as /usr/bin/fc -l
Also, when I run a script like this:
#!/usr/bin/ksh
fc -l
The result is same as /usr/bin/fc -l
Could anyone tell me what happened?
Alvin SIU
Ah, wisdom of the ancients... (Since this post is over a year old.)
Anyway, I just encountered this problem in Solaris 10. Issue seems to be this: When you define a function in /etc/profile, or in any file called by /etc/profile, your HISTFILE variable gets ignored by the Korn shell, and the shell instead uses ".sh_history" when accessing its history. Not sure why this is.
Result is that you see other root shell's commands. You can test it with :
lsof -p $$
or
cat /proc/$$/fd/63
It's possible that the login shell is not ksh or that $HISTFILE is being reset. One thing you can do is echo $HISTFILE in the various situations and see if it's different. Another thing to check is to see what shell you're in using ps.
Bash (default $HOME/.bash_history), for example, will have a different $HISTFILE than ksh (default $HOME/.sh_history).
Another possible reason for the difference is that the builtin fc may be able to see in-memory history that hasn't been written to disk yet (which the external /usr/bin/fc wouldn't be able to see). If this is true, it may be version dependent. Bash, for example, doesn't write history to the file until the shell exits. Ksh (at least the version I'm using) writes it immediately.