[Drbd-dev] DRBD8: Resync of device PauseSync state does not resume after an I/O error.

[Drbd-dev] DRBD8: Resync of device PauseSync state does not resume after an I/O error.

[Montrose, Ernest]

[-- Attachment #1: Type: text/plain, Size: 1017 bytes --]

Hi all,
Looks like what's going on is that on an I/O error during resync (on the
sync source), the io error handler set the new state to 
"Connected" and after_state_ch() is not called but scheduled it seems.
Next time after_state_ch()
is called the conn state is "Connected" for both old and new and the
disk state is Diskless.
We do not cal resume_next_sg() for this combination of state.  Could it
be as simple as this patch?

Index: drbd/drbd_main.c
===================================================================
--- drbd/drbd_main.c    (revision 10812)
+++ drbd/drbd_main.c    (working copy)
@@ -1009,7 +1009,8 @@
        // A resync finished or aborted, wake paused devices...
        if ( (os.conn > Connected && ns.conn <= Connected) ||
            (os.peer_isp && !ns.peer_isp) ||
-            (os.user_isp && !ns.user_isp) ) {
+            (os.user_isp && !ns.user_isp) ||
+        (ns.conn <= Connected && ns.disk == Diskless)) {
                resume_next_sg(mdev);
        }

[-- Attachment #2: Type: text/html, Size: 6007 bytes --]

Re: [Drbd-dev] DRBD8: Resync of device PauseSync state does not resume after an I/O error.

[Philipp Reisner]

Am Montag, 26. Februar 2007 22:30 schrieb Montrose, Ernest:
> Hi all,
> Looks like what's going on is that on an I/O error during resync (on the
> sync source), the io error handler set the new state to
> "Connected" and after_state_ch() is not called but scheduled it seems.
> Next time after_state_ch()
> is called the conn state is "Connected" for both old and new and the
> disk state is Diskless.
> We do not cal resume_next_sg() for this combination of state.  Could it
> be as simple as this patch?

Hi Ernest,

Ok, I now understand what was going on there. The __drbd_chk_io_error()
functions sets the disk to failed.
As a consequence in _drbd_set_state:

	if( ns.conn > Connected && (ns.disk <= Failed || ns.pdsk <= Failed )) {
		warn_sync_abort=1;
		ns.conn = Connected;
	}

The ns gets set to Connected. But the after_state_ch() is not called nor
scheduled...


I propose to fix it with this patch. Could you please test this patch?
If this patches fixes the issue as well ?

--- drbd_int.h  (revision 2773)
+++ drbd_int.h  (working copy)
@@ -1473,7 +1473,8 @@
        case Detach:
        case CallIOEHelper:
                if (mdev->state.disk > Failed) {
-                       _drbd_set_state(_NS(mdev,disk,Failed),ChgStateHard);
+                       _drbd_set_state(_NS(mdev,disk,Failed),
+                                       ChgStateHard|ScheduleAfter);
                        ERR("Local IO failed. Detaching...\n");
                }
                break;

In case it does, I will commit this, since it is the more generic fix 
to the issue. 

-Phil
-- 
: Dipl-Ing Philipp Reisner                      Tel +43-1-8178292-50 :
: LINBIT Information Technologies GmbH          Fax +43-1-8178292-82 :
: Vivenotgasse 48, 1120 Vienna, Austria        http://www.linbit.com :

RE: [Drbd-dev] DRBD8: Resync of device PauseSync state does not resume after an I/O error.

[Montrose, Ernest]

Phil,
No problem...I will test it and let you know.

EM--

-----Original Message-----
From: Philipp Reisner [mailto:philipp.reisner@linbit.com] 
Sent: Friday, March 02, 2007 8:33 AM
To: drbd-dev@linbit.com
Cc: Montrose, Ernest
Subject: Re: [Drbd-dev] DRBD8: Resync of device PauseSync state does not
resume after an I/O error.

Am Montag, 26. Februar 2007 22:30 schrieb Montrose, Ernest:
> Hi all,
> Looks like what's going on is that on an I/O error during resync (on
the
> sync source), the io error handler set the new state to
> "Connected" and after_state_ch() is not called but scheduled it seems.
> Next time after_state_ch()
> is called the conn state is "Connected" for both old and new and the
> disk state is Diskless.
> We do not cal resume_next_sg() for this combination of state.  Could
it
> be as simple as this patch?

Hi Ernest,

Ok, I now understand what was going on there. The __drbd_chk_io_error()
functions sets the disk to failed.
As a consequence in _drbd_set_state:

	if( ns.conn > Connected && (ns.disk <= Failed || ns.pdsk <=
Failed )) {
		warn_sync_abort=1;
		ns.conn = Connected;
	}

The ns gets set to Connected. But the after_state_ch() is not called nor
scheduled...


I propose to fix it with this patch. Could you please test this patch?
If this patches fixes the issue as well ?

--- drbd_int.h  (revision 2773)
+++ drbd_int.h  (working copy)
@@ -1473,7 +1473,8 @@
        case Detach:
        case CallIOEHelper:
                if (mdev->state.disk > Failed) {
-
_drbd_set_state(_NS(mdev,disk,Failed),ChgStateHard);
+                       _drbd_set_state(_NS(mdev,disk,Failed),
+                                       ChgStateHard|ScheduleAfter);
                        ERR("Local IO failed. Detaching...\n");
                }
                break;

In case it does, I will commit this, since it is the more generic fix 
to the issue. 

-Phil
-- 
: Dipl-Ing Philipp Reisner                      Tel +43-1-8178292-50 :
: LINBIT Information Technologies GmbH          Fax +43-1-8178292-82 :
: Vivenotgasse 48, 1120 Vienna, Austria        http://www.linbit.com :

RE: [Drbd-dev] DRBD8: Resync of device PauseSync state does not resume after an I/O error.

[Montrose, Ernest]

Phil,
I just tested it! The patch works just fine. Please check it in.

EM--

-----Original Message-----
From: Philipp Reisner [mailto:philipp.reisner@linbit.com] 
Sent: Friday, March 02, 2007 8:33 AM
To: drbd-dev@linbit.com
Cc: Montrose, Ernest
Subject: Re: [Drbd-dev] DRBD8: Resync of device PauseSync state does not
resume after an I/O error.

Am Montag, 26. Februar 2007 22:30 schrieb Montrose, Ernest:
> Hi all,
> Looks like what's going on is that on an I/O error during resync (on
the
> sync source), the io error handler set the new state to
> "Connected" and after_state_ch() is not called but scheduled it seems.
> Next time after_state_ch()
> is called the conn state is "Connected" for both old and new and the
> disk state is Diskless.
> We do not cal resume_next_sg() for this combination of state.  Could
it
> be as simple as this patch?

Hi Ernest,

Ok, I now understand what was going on there. The __drbd_chk_io_error()
functions sets the disk to failed.
As a consequence in _drbd_set_state:

	if( ns.conn > Connected && (ns.disk <= Failed || ns.pdsk <=
Failed )) {
		warn_sync_abort=1;
		ns.conn = Connected;
	}

The ns gets set to Connected. But the after_state_ch() is not called nor
scheduled...


I propose to fix it with this patch. Could you please test this patch?
If this patches fixes the issue as well ?

--- drbd_int.h  (revision 2773)
+++ drbd_int.h  (working copy)
@@ -1473,7 +1473,8 @@
        case Detach:
        case CallIOEHelper:
                if (mdev->state.disk > Failed) {
-
_drbd_set_state(_NS(mdev,disk,Failed),ChgStateHard);
+                       _drbd_set_state(_NS(mdev,disk,Failed),
+                                       ChgStateHard|ScheduleAfter);
                        ERR("Local IO failed. Detaching...\n");
                }
                break;

In case it does, I will commit this, since it is the more generic fix 
to the issue. 

-Phil
-- 
: Dipl-Ing Philipp Reisner                      Tel +43-1-8178292-50 :
: LINBIT Information Technologies GmbH          Fax +43-1-8178292-82 :
: Vivenotgasse 48, 1120 Vienna, Austria        http://www.linbit.com :