bugGNU Octave - Bugs: bug #54444, Octave CLI doesn't terminate at...

 
 

bug #54444: Octave CLI doesn't terminate at sigterm until <ENTER> pressed at prompt

Submitter:  Olaf Till <i7tiol>
Submitted:  Sat 04 Aug 2018 04:16:28 PM UTC
   
 
Category:  Interpreter Severity:  3 - Normal
Priority:  5 - Normal Item Group:  Regression
Status:  In Progress Assigned to:  None
Originator Name:  Olaf Till Open/Closed:  * Open
Release:  * dev Operating System:  * GNU/Linux
Fixed Release:  None Planned Release:  None
* Mandatory Fields

Add a New Comment Rich Markup
   

Jump to the original submission

Fri 11 Sep 2020 03:55:39 PM UTC, comment #15: 

We could just have Octave call exit in the signal handler immediately when the TERM signal arrives, but that would skip the usual shutdown procedure and that doesn't seem like the best behavior.  So, we just have the signal handler set a global variable and respond to that when it is safe to do so (at a place where octave_quit is called).  So typically Octave does not respond to signals while it is waiting for input or for a library function to return (see, for example, bug #56828 or bug #37795).  It only works with readline because the readline function that waits for user input also allows us to periodically check for events.

John W. Eaton <jwe>
Group administrator
Fri 11 Sep 2020 08:59:53 AM UTC, comment #14: 


> But, presumably Web users of the Octave Online Server do not have a way of sending an IPC signal to the running process.


Right; it's more a problem for sysadmins running their own instance of Octave Online Server.  Several Octave processes get spawned, but extra care is required to clean them up, since a simple SIGTERM is insufficient.  Web users don't see this.

Shane <sffc>
Thu 10 Sep 2020 09:51:36 PM UTC, comment #13: 

But, presumably Web users of the Octave Online Server do not have a way of sending an IPC signal to the running process.

Rik <rik5>
Group administrator
Thu 10 Sep 2020 09:22:22 PM UTC, comment #12: 
Shane <sffc>
Thu 10 Sep 2020 04:24:45 PM UTC, comment #11: 

Verified fix.  Changing Status to "In Progress" to reflect the fact that this may need to be re-visited if we want to get the last 10% of cases where readline is not being used.

Rik <rik5>
Group administrator
Thu 10 Sep 2020 03:48:04 PM UTC, comment #10: 

I pushed the change on stable and merged with default:

http://hg.savannah.gnu.org/hgweb/octave/rev/f0a3f4527ba6

John W. Eaton <jwe>
Group administrator
Thu 10 Sep 2020 12:09:02 AM UTC, comment #9: 

I verified that the patch does what it says.  Most of the time I think readline is compiled in so this will certainly help on a lot of installations.  It is probably worth applying as it would get rid of 90% of complaints.

I would still leave this bug open though in case we can find a universal fix.

Rik <rik5>
Group administrator
Wed 09 Sep 2020 09:03:17 PM UTC, comment #8: 

The attached patch will allow signals to be processed while readline is waiting for input.  But it won't fix the problem if Octave is not using readline, either because it is disabled or Octave is started with the --no-line-editing option.  But in that case, it won't handle graphics events either.  So, should we apply this patch to at least handle the case when Octave is using readline?


(file #49773)

John W. Eaton <jwe>
Group administrator
Sat 04 Jan 2020 04:30:33 AM UTC, comment #7: 

I got hit by this, too, when upgrading Octave Online to version 5.1.  Previously, I could send SIGTERM to the Octave process, and it would gracefully exit, no matter what state it was in.  Now, sending SIGTERM has no effect until the process gets some other kind of interaction.

I think this change is a bug in Octave, because the convention of SIGTERM is that the process promptly begins a graceful shutdown procedure.  If the process does not exit on SIGTERM after some amount of time, then the only other option is to send SIGKILL to the process.  *SIGTERM-wait-SIGKILL is expected to work regardless of the identity of the process*.  Since Octave does not do anything when receiving a SIGTERM, it forces agents to send SIGKILL, which doesn't allow Octave to perform any cleanup logic.

As a workaround, for the purposes of terminating an Octave process, I've changed the logic to first execute "exit" in the process, and then send SIGKILL after 10 seconds if the process hasn't gracefully exited.  This only works because I know I have an Octave process.

Shane <sffc>
Sat 09 Mar 2019 12:22:12 PM UTC, comment #6: 

FTR: Correction: The reported behaviour was not what affected the parallel package. I didn't see that Octaves signal handling is now done in an extra thread. The server of the parallel package forks and closes the parent, so this extra thread is lost. So the server has to install its own signal handling anyway (and unblock the signals).

Olaf Till <i7tiol>
Group Member
Sat 02 Mar 2019 08:02:59 PM UTC, comment #5: 

According to current libinterp/corefcn/sighandlers.cc, this seems to be due to a deliberate change -- signals for which the default action is to terminate the program are now handled by generic_sig_handler(), meaning that terminating the program is delayed until respond_to_pending_signals() is called, which again is only called by octave_quit().

I think this change is problematic. In a resting Octave instance, octave_quit() is not called. So now it is not possible any more to kill a resting Octave instance 'from outside', except with a fatal signal for which the Octave instance can't perform any clean-up actions.

The necessity for killing a resting Octave instance 'from outside', while still enabling clean-up, is given for Octave servers of the 'parallel' package. But I think it is also a generally desirable feature.

@jwe: What should be done here?

Olaf Till <i7tiol>
Group Member
Mon 04 Feb 2019 05:39:09 PM UTC, comment #4: 

Re-titling since this is not just an issue for 4.4, but also for 5.1 and the development branch.

Rik <rik5>
Group administrator
Mon 31 Dec 2018 05:30:40 PM UTC, comment #3: 

I just re-confirmed that this issue is still present (12/31/18).

Rik <rik5>
Group administrator
Thu 27 Sep 2018 08:36:58 PM UTC, comment #2: 

Amending title to add the part about pressing the <ENTER> key.  This also affects the development branch.

Rik <rik5>
Group administrator
Sat 04 Aug 2018 07:03:15 PM UTC, comment #1: 

I tested Octave 3.4 and 3.6, and octave-cli 3.8 through 4.2, and all exit immediately when they receive a SIGHUP, SIGTERM, or SIGQUIT.

In Octave 4.4 (either octave-cli or octave-gui), the interpreter does exit after receiving SIGHUP, SIGTERM, or SIGQUIT, but only after pressing Enter at the interpreter prompt.

Mike Miller <mtmiller>
Group Member
Sat 04 Aug 2018 04:16:28 PM UTC, original submission:  

Octave-4.4(.0), started with 'octave-cli', doesn't terminate at receiving sigterm, it seems to need sigkill, which leaves files marked in Octave as temporary open. Affects the parallel package.

Olaf Till <i7tiol>
Group Member

 

(Note: upload size limit is set to 16384 kB, after insertion of the required escape characters.)

Attach Files:
   
   
Comment:
   

Attached Files
file #49773:  event-hook-quit-patch.txt added by jwe (3KiB - text/plain)

 

Depends on the following items: None found

Items that depend on this one: None found

 

Carbon-Copy List
  • -email is unavailable- added by jwe (Updated the item)
  • -email is unavailable- added by sffc (Posted a comment)
  • -email is unavailable- added by i7tiol (I'm adding jwe to the CC list since his decision on how to proceed is needed.)
  • -email is unavailable- added by rik5 (Posted a comment)
  • -email is unavailable- added by i7tiol (Submitted the item)
  •  

    There are 0 votes so far. Votes easily highlight which items people would like to see resolved in priority, independently of the priority of the item set by tracker managers.

    Only group members can vote.

     

    Follow 11 latest changes.

    Date Changed by Updated Field Previous Value => Replaced by
    2020-09-10 rik5 StatusReady For Test In Progress
        Release4.4.0 dev
    2020-09-10 jwe StatusConfirmed Ready For Test
    2020-09-09 jwe Attached File- Added event-hook-quit-patch.txt, #49773
    2019-03-02 i7tiol Carbon-Copy- Added "john w. eaton" <jwe@gnu.org>
    2019-02-04 rik5 SummaryOctave-4.4.X CLI doesn't terminate at sigterm until &lt;ENTER&gt; pressed at prompt Octave CLI doesn't terminate at sigterm until <ENTER> pressed at prompt
    2018-09-28 mtmiller Carbon-CopyRemoved 80942 -
    2018-09-27 rik5 Summaryoctave-cli of octave-4.4.0 doesn't terminate at sigterm. Octave-4.4.X CLI doesn't terminate at sigterm until <ENTER> pressed at prompt
    2018-08-04 mtmiller CategoryNone Interpreter
        Item GroupOther Regression
        StatusNone Confirmed

    Back to the top

    Powered by Savane 3.13-cf05.
    Corresponding source code