bug#8929: 24.0.50; tramp hangs when process died

bug#8929: 24.0.50; tramp hangs when process died

[Sébastien Gross]

Hi,

Using tramp to edit remote files creates some ssh processes.

When this process dies for some reasons (such ad IP change or network
outrage), the associated buffer is still alive and makes tramp hanging
when trying to reconnect the remote host.

This behavior happens every time I carry my laptop (in suspend to disk
or suspend to ram mode) from my home to my office (or the way back).


Here is a patch (for tramp-sh.el) to add a sentinel that kills the
process buffer when the process exit:

--- tramp-sh.el.o	2011-06-24 14:44:09.023169775 +0200
+++ tramp-sh.el	2011-06-24 14:52:57.383172776 +0200
@@ -4277,6 +4277,13 @@
 		       (tramp-get-connection-buffer vec)
 		       tramp-encoding-shell))))
 
+	    ;; Kill buffer when process died.
+	    (set-process-sentinel
+	     p
+	     (lambda (proc change)
+	       (when (eq (process-status proc) 'exit)
+		 (kill-buffer (process-buffer proc)))
+
 	    (tramp-message
 	     vec 6 "%s" (mapconcat 'identity (process-command p) " "))


Hope this doesn't break anything.

Cheers.



In GNU Emacs 24.0.50.1 (x86_64-pc-linux-gnu, GTK+ Version 2.20.1)
 of 2011-05-23 on builder1-tc2ams, modified by Debian
 (emacs-snapshot package, version 1:20110520-1+squeeze)
Windowing system distributor `The X.Org Foundation', version 11.0.10707000
configured using `configure  '--build' 'x86_64-linux-gnu' '--host' 'x86_64-linux-gnu' '--prefix=/usr' '--sharedstatedir=/var/lib' '--libexecdir=/usr/lib' '--localstatedir=/var' '--infodir=/usr/share/info' '--mandir=/usr/share/man' '--with-pop=yes' '--enable-locallisppath=/etc/emacs-snapshot:/etc/emacs:/usr/local/share/emacs/24.0.50/site-lisp:/usr/local/share/emacs/site-lisp:/usr/share/emacs/24.0.50/site-lisp:/usr/share/emacs/site-lisp' '--without-compress-info' '--with-x=yes' '--with-x-toolkit=gtk' '--with-imagemagick=yes' 'build_alias=x86_64-linux-gnu' 'host_alias=x86_64-linux-gnu' 'CFLAGS=-DDEBIAN -DSITELOAD_PURESIZE_EXTRA=5000 -g -O2' 'LDFLAGS=-g -Wl,--as-needed' 'CPPFLAGS=''

-- 
Sébastien Gross

bug#8929: 24.0.50; tramp hangs when process died

[Michael Albinus]

Sébastien Gross <seb@chezwam.org> writes:

> Hi,

Hi,

> When this process dies for some reasons (such ad IP change or network
> outrage), the associated buffer is still alive and makes tramp hanging
> when trying to reconnect the remote host.
>
> This behavior happens every time I carry my laptop (in suspend to disk
> or suspend to ram mode) from my home to my office (or the way back).
>
> Here is a patch (for tramp-sh.el) to add a sentinel that kills the
> process buffer when the process exit:
>
> Hope this doesn't break anything.

I wouldn't like to add such a radical buffer kill. Asynchronous
processes on remote hosts would loose their output buffer, when they
have finished. Debugging of Tramp problems would be harder.

On GNU Linux systems, it might be possible to catch D-Bus signals for
resuming the system after hibernate or suspend. Could you, please, eval
the following lines, and see whether Tramp behaves better?

--8<---------------cut here---------------start------------->8---
(require 'dbus)
(dbus-register-signal
 :system "org.freedesktop.UPower" "/org/freedesktop/UPower"
 "org.freedesktop.UPower" "Resuming" 'tramp-cleanup-all-connections)
--8<---------------cut here---------------end--------------->8---

If this works for you (it does for me), I would add a custom option for
enabling this behaviour.

Maybe we shall even introduce a variable `system-resume-hook', which
keeps functions to be called when Emacs continues to run after a
hibernate or suspend of the machine.

> Cheers.

Best regards, Michael.

bug#8929: 24.0.50; tramp hangs when process died

[Sébastien Gross]

Michael Albinus <michael.albinus@gmx.de> writes:

Hi,


> I wouldn't like to add such a radical buffer kill. Asynchronous
> processes on remote hosts would loose their output buffer, when they
> have finished. Debugging of Tramp problems would be harder.

I do totally agree with you about the debugging purposes. So why not
setting up a variable such as:

(defcustom tramp-kill-process-buffer-on-exit nil 
  "Kill remote shell buffer process if connection with remote host is
  lost."
  :group 'tramp
  :type 'boolean)

Setting by default this variable to `nil' doesn't affect current tramp
behavior. Then one can chose how tramp should handle process ending by
changing `tramp-kill-process-buffer-on-exit'.

And add in `tramp-maybe-open-connection':

  ;; Kill buffer when process died.
  (when tramp-kill-process-buffer-on-exit
    (set-process-sentinel
      p
      (lambda (proc change)
        (when (eq (process-status proc) 'exit)
 	 (kill-buffer (process-buffer proc)))


>
> On GNU Linux systems, it might be possible to catch D-Bus signals for
> resuming the system after hibernate or suspend. Could you, please, eval
> the following lines, and see whether Tramp behaves better?
>
> (require 'dbus)
> (dbus-register-signal
>  :system "org.freedesktop.UPower" "/org/freedesktop/UPower"
>  "org.freedesktop.UPower" "Resuming" 'tramp-cleanup-all-connections)
>
> If this works for you (it does for me), I would add a custom option for
> enabling this behaviour.

This works fine but on in one use case on power resuming. If for some
reason the connection with a remote host is lost, the dbus solution acts
more like a workaround, not like a global issue solving.


Cheers,

-- 
Sébastien Gross

bug#8929: 24.0.50; tramp hangs when process died

[Michael Albinus]

Sébastien Gross <seb@chezwam.org> writes:

> Hi,

Hi Sébastien,

>> I wouldn't like to add such a radical buffer kill. Asynchronous
>> processes on remote hosts would loose their output buffer, when they
>> have finished. Debugging of Tramp problems would be harder.
>
> I do totally agree with you about the debugging purposes.

I've checked the code: we have already a sentinel, which adds some
cleanup in case a process died. But this is applied only for
asynchronous processes.

I've extended the sentinel a little bit, and I've moved the sentinel
setup to `tramp-maybe-open-connection'. It is activated for all
processes now.

Could you, please, check the patch, whether it is sufficient to your needs?

--8<---------------cut here---------------start------------->8---
*** /home/albinus/src/tramp/lisp/tramp-sh.el.~1~	2011-07-01 11:25:11.000000000 +0200
--- /home/albinus/src/tramp/lisp/tramp-sh.el	2011-07-01 14:35:41.304448000 +0200
***************
*** 2669,2674 ****
--- 2669,2675 ----
      (let ((vec (tramp-get-connection-property proc "vector" nil)))
        (when vec
  	(tramp-message vec 5 "Sentinel called: `%s' `%s'" proc event)
+         (tramp-flush-connection-property proc)
          (tramp-flush-directory-property vec "")))))
  
  ;; We use BUFFER also as connection buffer during setup. Because of
***************
*** 2721,2729 ****
  		       v 'file-error
  		       "pty association is not supported for `%s'" name)))))
  	      (let ((p (tramp-get-connection-process v)))
! 		;; Set sentinel and query flag for this process.
! 		(tramp-set-connection-property p "vector" v)
! 		(set-process-sentinel p 'tramp-process-sentinel)
  		(tramp-compat-set-process-query-on-exit-flag p t)
  		;; Return process.
  		p)))
--- 2722,2728 ----
  		       v 'file-error
  		       "pty association is not supported for `%s'" name)))))
  	      (let ((p (tramp-get-connection-process v)))
! 		;; Set query flag for this process.
  		(tramp-compat-set-process-query-on-exit-flag p t)
  		;; Return process.
  		p)))
***************
*** 4300,4310 ****
  				 tramp-encoding-command-interactive)
  			 (list tramp-encoding-shell))))))
  
  	    (tramp-message
  	     vec 6 "%s" (mapconcat 'identity (process-command p) " "))
  
  	    ;; Check whether process is alive.
- 	    (tramp-compat-set-process-query-on-exit-flag p nil)
  	    (tramp-barf-if-no-shell-prompt
  	     p 60 "Couldn't find local shell prompt %s" tramp-encoding-shell)
  
--- 4299,4313 ----
  				 tramp-encoding-command-interactive)
  			 (list tramp-encoding-shell))))))
  
+ 	    ;; Set sentinel and query flag.
+ 	    (tramp-set-connection-property p "vector" vec)
+ 	    (set-process-sentinel p 'tramp-process-sentinel)
+ 	    (tramp-compat-set-process-query-on-exit-flag p nil)
+ 
  	    (tramp-message
  	     vec 6 "%s" (mapconcat 'identity (process-command p) " "))
  
  	    ;; Check whether process is alive.
  	    (tramp-barf-if-no-shell-prompt
  	     p 60 "Couldn't find local shell prompt %s" tramp-encoding-shell)
  
--8<---------------cut here---------------end--------------->8---

> Cheers,

Best regards, Michael.

bug#8929: 24.0.50; tramp hangs when process died

[Michael Albinus]

Michael Albinus <michael.albinus@gmx.de> writes:

Hi Sébastien,

> I've extended the sentinel a little bit, and I've moved the sentinel
> setup to `tramp-maybe-open-connection'. It is activated for all
> processes now.
>
> Could you, please, check the patch, whether it is sufficient to your needs?

Does the patch work for you? I have submitted it to the trunk, meanwhile.

If yes, we could close the bug.

>> Cheers,

Best regards, Michael.

bug#8929: 24.0.50; tramp hangs when process died

[Michael Albinus]

Sébastien Gross <seb@chezwam.org> writes:

> Hi Michael,

Hi Sébastien,

> Seems to be OK but I haven't tested it in all cases.
> I'll come back to you if I find something wrong in the future.

Thanks for testing. I'll close the bug; feel free to come back if there
are still problems.

> Thanks for your support.
>
> Cheers

Best regards, Michael.