Re: Slow Tramp issue

[Michael Albinus]

Sandro Romanzetti <roman.sandro@gmail.com> writes:

> Dear Tramp Developers,

Hi Sandro,

> I am trying to execute the following R code from within a org-mode
> file but it take more than 60 seconds to get the results back on my
> screen. For this reason I usually login to my remote server and start
> emacs from the terminal. In that case I have no delays whatsoever.
>
> Here is the source code block:
>
> #+begin_src R :session myplot :results value :colnames yes :exports results
> library(tidyverse)
> T<-tibble(x=(1:20),y=(1:20))
> T
> #+end_src
>
> I attach the debug message (using tramp-verbose 10). I would be very grateful 
> if you could give me any hint on how to solve this issue.

Thanks. I've tried to analyze the traces. Note that I don't use R, so I
must guess something.

The traces don't show any remote process call. So I assume that there is
already a remote R process, running as REPL loop.

The majority of the Tramp actions is about the remote file
"/tmp/R-Gpow4E". It doesn't exist at the beginning, and Tramp marks it
as such in the cache:

--8<---------------cut here---------------start------------->8---
00:52:36.548413 tramp-send-command (6) # test -e /tmp/R-Gpow4E 2>/dev/null; 
echo tramp_exit_status $?
00:52:36.591744 tramp-wait-for-regexp (6) #
tramp_exit_status 1
///1b9b06976240a21aa1381a07a40053bd#$
00:52:36.591976 tramp-get-connection-property (7) # process-buffer nil; cache 
used: nil
00:52:36.592165 tramp-set-file-property (8) # /tmp/R-Gpow4E file-exists-p nil
--8<---------------cut here---------------end--------------->8---

But then, your session waits that this file exists:

--8<---------------cut here---------------start------------->8---
00:52:36.876143 tramp-get-file-property (8) # /tmp/R-Gpow4E file-exists-p nil; 
inhibit: 60; cache used: t; cached at: 00:52:36
00:52:37.137672 tramp-get-file-property (8) # /tmp/R-Gpow4E file-exists-p nil; 
inhibit: 60; cache used: t; cached at: 00:52:36

[...]

00:53:36.565685 tramp-get-file-property (8) # /tmp/R-Gpow4E file-exists-p nil; 
inhibit: 60; cache used: t; cached at: 00:52:36
00:53:36.831883 tramp-get-file-property (8) # /tmp/R-Gpow4E file-exists-p 
undef; inhibit: 60; cache used: nil; cached at: 00:52:36
--8<---------------cut here---------------end--------------->8---

For 60 seconds, Tramp reports the non-existence according to its
cache. Then the cache expires, and Tramp does a real check:

--8<---------------cut here---------------start------------->8---
00:53:36.881497 tramp-send-command (6) # test -e /tmp/R-Gpow4E 2>/dev/null; 
echo tramp_exit_status $?
00:53:36.930700 tramp-wait-for-regexp (6) #
tramp_exit_status 0
///1b9b06976240a21aa1381a07a40053bd#$
00:53:36.931118 tramp-get-connection-property (7) # process-buffer nil; cache 
used: nil
00:53:36.931540 tramp-set-file-property (8) # /tmp/R-Gpow4E file-exists-p t
--8<---------------cut here---------------end--------------->8---

Tramp knows about a stale cache when it starts a remote
process. However, when a remote process is running, and a client program
(your R session) sends statements which change the existence of a remote
file, Tramp cannot know this. Your client program must inhibit Tramp's
cache while doing so.

I recommend you to contact the org maintainers about. At a proper place,
where the code waits for the existence of the tmpfile, they shall
let-bind remote-file-name-inhibit-cache to t.

> Kind regards,
> Sandro Romanzetti

Best regards, Michael.