diff mbox series

[bug#68357] shepherd: service: fix `unload all`

Message ID 20240110013058.1822277-2-shinyzero0@tilde.club
State New
Headers
Series [bug#68357] shepherd: service: fix `unload all` |

Commit Message

ShinyZero0 Jan. 10, 2024, 1:30 a.m. UTC 
* modules/shepherd/service.scm: fix `unload all`

---
 modules/shepherd/service.scm | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)

Comments

Ludovic Courtès Jan. 14, 2024, 9:33 p.m. UTC | #1 
Hi,

"zero@fedora" <shinyzero0@tilde.club> skribis:

> * modules/shepherd/service.scm: fix `unload all`

Thanks for the patch!

> -      (('unregister-all)                          ;no reply
> -       (let ((root (cdr (vhash-assq 'root registered))))
> -         (loop (fold (cut vhash-consq <> root <>)
> -                     vlist-null
> -                     (service-provision root)))))
>        (('lookup name reply)
>         ;; Look up NAME and return it, or #f, to REPLY.
>         (put-message reply
> @@ -2638,8 +2633,11 @@ requested to be removed."
>    (let ((name (string->symbol service-name)))
>      (cond ((eq? name 'all)
>             ;; Special 'remove all' case.
> -           (put-message (current-registry-channel) `(unregister-all))
> -           #t)
> +           (unregister-services
> +             (filter
> +               (lambda (sv)
> +                 (not (eq? (service-canonical-name sv) 'root)))
> +               (service-list))))

Do I get it right that the problem with the current implementation is
that services are removed from the registry but not actually stopped?

Ludo’.
ShinyZero0 Jan. 14, 2024, 10:53 p.m. UTC | #2 
Hi Ludo! Thanks for replying.

Yes, that was the problem. `herd reload root all` was leaving the
processes running and many of them crashed or had strange behaviour
after loading, so one had to kill them manually and then restart 
the services.

Of course `herd unload root all` was the same.

Best wishes, Paul.
Ludovic Courtès Jan. 21, 2024, 10:23 p.m. UTC | #3 
Hi!

"ShinyZero0" <shinyzero0@tilde.club> skribis:

> Yes, that was the problem. `herd reload root all` was leaving the
> processes running and many of them crashed or had strange behaviour
> after loading, so one had to kill them manually and then restart 
> the services.
>
> Of course `herd unload root all` was the same.
>
> Best wishes, Paul.

Thanks for clarifying.  I pushed something similar along with a test as
commit b7fcdad75dff73fb4a8f531d0834eb9a965bcfd6.

Ludo’.
diff mbox series

Patch

diff --git a/modules/shepherd/service.scm b/modules/shepherd/service.scm
index 5be78bd..54d3400 100644
--- a/modules/shepherd/service.scm
+++ b/modules/shepherd/service.scm
@@ -1136,11 +1136,6 @@  requests arriving on @var{channel}."
                  (length lst))
            (map service-canonical-name lst))
           (loop registered))))
-      (('unregister-all)                          ;no reply
-       (let ((root (cdr (vhash-assq 'root registered))))
-         (loop (fold (cut vhash-consq <> root <>)
-                     vlist-null
-                     (service-provision root)))))
       (('lookup name reply)
        ;; Look up NAME and return it, or #f, to REPLY.
        (put-message reply
@@ -2638,8 +2633,11 @@  requested to be removed."
   (let ((name (string->symbol service-name)))
     (cond ((eq? name 'all)
            ;; Special 'remove all' case.
-           (put-message (current-registry-channel) `(unregister-all))
-           #t)
+           (unregister-services
+             (filter
+               (lambda (sv)
+                 (not (eq? (service-canonical-name sv) 'root)))
+               (service-list))))
           (else
            ;; Removing only one service.
            (match (lookup-service name)