[PATCH 1/2] crm-fence-peer.9.sh: fix parsing in_ccm crmd fields of node_state with Pacemaker 2.1.7

[PATCH 1/2] crm-fence-peer.9.sh: fix parsing in_ccm crmd fields of node_state with Pacemaker 2.1.7

[Su Yue]

From: Su Yue <glass.su@suse.com>

If pacemaker version < 2.1.7, in_ccm of node_state is "true" or "false"
and crmd is "online" or "offline".

pacemaker 2.1.7 changed the two fields into timestamps.
For in_ccm, the value is timestamp since when node has been a cluster
member("true"). A value 0 of means the node is not a cluster member("false").
For crmd, the value is timestamp since when peer has been online in
CPG("online"). A value 0 means the peer is offline in CPG("offline").

The original code doesn't handle these fields in timestamp format.
Since there are many comprare of strings in context, converting in_ccm and crmd
from timestamps/0 to old strings is simpler and clearer.

Link: https://github.com/ClusterLabs/pacemaker/blob/Pacemaker-2.1.7/lib/pengine/unpack.c#L1581
Signed-off-by: Su Yue <glass.su@suse.com>
---
 scripts/crm-fence-peer.9.sh | 25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)

diff --git a/scripts/crm-fence-peer.9.sh b/scripts/crm-fence-peer.9.sh
index 44da6516bf3f..a3353a7354a6 100755
--- a/scripts/crm-fence-peer.9.sh
+++ b/scripts/crm-fence-peer.9.sh
@@ -888,6 +888,31 @@ guess_if_pacemaker_will_fence()
 		esac
 	done
 
+	# Copied from pacemaker-2.1.7:lib/pengine/unpack.c:
+	# Since crm_feature_set 3.18.0 (pacemaker-2.1.7):
+	#
+	# - in_ccm		::= <timestamp>|0
+	# Since when node has been a cluster member. A value 0 of means the
+	# node is not a cluster member.
+	# - crmd		::= <timestamp>|0
+	# Since when peer has been online in CPG. A value 0 means the peer
+	# is offline in CPG.
+	if [[ $in_ccm =~ ^[0-9]+$ ]]; then
+		if [[ $in_ccm = "0" ]]; then
+			in_ccm="false"
+		else
+			in_ccm="true"
+		fi
+	fi
+
+	if [[ $crmd =~ ^[0-9]+$ ]]; then
+		if [[ $crmd = "0" ]]; then
+			crmd="offline"
+		else
+			crmd="online"
+		fi
+	fi
+
 	# if it is not enabled, no point in waiting for it.
 	if ! $stonith_enabled ; then
 		# "normalize" the rest of the logic
-- 
2.44.0

[PATCH 2/2] crm-fence-peer.9.sh: use join of node_state to judge whether node is banned

[Su Yue]

From: Su Yue <glass.su@suse.com>

crmd in node_state can't be "banned". join should be used instead
of crmd.

Signed-off-by: Su Yue <glass.su@suse.com>
---
 scripts/crm-fence-peer.9.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/scripts/crm-fence-peer.9.sh b/scripts/crm-fence-peer.9.sh
index a3353a7354a6..b326a1656c15 100755
--- a/scripts/crm-fence-peer.9.sh
+++ b/scripts/crm-fence-peer.9.sh
@@ -934,7 +934,7 @@ guess_if_pacemaker_will_fence()
 
 	# for further inspiration, see pacemaker:lib/pengine/unpack.c, determine_online_status_fencing()
 	[[ -z $in_ccm ]] && will_fence=true
-	[[ $crmd = "banned" ]] && will_fence=true
+	[[ $join = "banned" ]] && will_fence=true
 	if [[ ${expected-down} = "down" && $in_ccm = "false"  && $crmd != "online" ]]; then
 		: "pacemaker considers this as clean down"
 	elif [[ $in_ccm = false ]] || [[ $crmd != "online" ]]; then
-- 
2.44.0

Re: [PATCH 2/2] crm-fence-peer.9.sh: use join of node_state to judge whether node is banned

[Su Yue]

Hi, dear drbd folks
  Would you help review the patches for drbd-utils? Many thanks!

— 
Su

> On Mar 3, 2024, at 17:12, Su Yue <glass.su@suse.com> wrote:
> 
> From: Su Yue <glass.su@suse.com>
> 
> crmd in node_state can't be "banned". join should be used instead
> of crmd.
> 
> Signed-off-by: Su Yue <glass.su@suse.com>
> ---
> scripts/crm-fence-peer.9.sh | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/scripts/crm-fence-peer.9.sh b/scripts/crm-fence-peer.9.sh
> index a3353a7354a6..b326a1656c15 100755
> --- a/scripts/crm-fence-peer.9.sh
> +++ b/scripts/crm-fence-peer.9.sh
> @@ -934,7 +934,7 @@ guess_if_pacemaker_will_fence()
> 
> # for further inspiration, see pacemaker:lib/pengine/unpack.c, determine_online_status_fencing()
> [[ -z $in_ccm ]] && will_fence=true
> - [[ $crmd = "banned" ]] && will_fence=true
> + [[ $join = "banned" ]] && will_fence=true
> if [[ ${expected-down} = "down" && $in_ccm = "false"  && $crmd != "online" ]]; then
> : "pacemaker considers this as clean down"
> elif [[ $in_ccm = false ]] || [[ $crmd != "online" ]]; then
> -- 
> 2.44.0
> 

Re: [PATCH 2/2] crm-fence-peer.9.sh: use join of node_state to judge whether node is banned

[Su Yue]

> On Mar 7, 2024, at 10:54, Su Yue <glass.su@suse.com> wrote:
> 
> Hi, dear drbd folks
>  Would you help review the patches for drbd-utils? Many thanks!
> 
Gentle ping.

— 
Su
> — 
> Su
> 
>> On Mar 3, 2024, at 17:12, Su Yue <glass.su@suse.com> wrote:
>> 
>> From: Su Yue <glass.su@suse.com>
>> 
>> crmd in node_state can't be "banned". join should be used instead
>> of crmd.
>> 
>> Signed-off-by: Su Yue <glass.su@suse.com>
>> ---
>> scripts/crm-fence-peer.9.sh | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>> 
>> diff --git a/scripts/crm-fence-peer.9.sh b/scripts/crm-fence-peer.9.sh
>> index a3353a7354a6..b326a1656c15 100755
>> --- a/scripts/crm-fence-peer.9.sh
>> +++ b/scripts/crm-fence-peer.9.sh
>> @@ -934,7 +934,7 @@ guess_if_pacemaker_will_fence()
>> 
>> # for further inspiration, see pacemaker:lib/pengine/unpack.c, determine_online_status_fencing()
>> [[ -z $in_ccm ]] && will_fence=true
>> - [[ $crmd = "banned" ]] && will_fence=true
>> + [[ $join = "banned" ]] && will_fence=true
>> if [[ ${expected-down} = "down" && $in_ccm = "false"  && $crmd != "online" ]]; then
>> : "pacemaker considers this as clean down"
>> elif [[ $in_ccm = false ]] || [[ $crmd != "online" ]]; then
>> -- 
>> 2.44.0
>> 
> 

Re: [PATCH 2/2] crm-fence-peer.9.sh: use join of node_state to judge whether node is banned

[Lars Ellenberg]

On Thu, Mar 14, 2024 at 11:18:22AM +0800, Su Yue wrote:
> 
> 
> > On Mar 7, 2024, at 10:54, Su Yue <glass.su@suse.com> wrote:
> > 
> > Hi, dear drbd folks
> >  Would you help review the patches for drbd-utils? Many thanks!
> > 
> Gentle ping.

So you apparently have some kind of test suite
excercising this fencing handler?

Thank you for making us aware, and for your patience.
Looks plausible.

	Lars

Re: [PATCH 2/2] crm-fence-peer.9.sh: use join of node_state to judge whether node is banned

[Glass Su]

> On Apr 17, 2024, at 17:57, Lars Ellenberg <lars.ellenberg@linbit.com> wrote:
> 
> On Thu, Mar 14, 2024 at 11:18:22AM +0800, Su Yue wrote:
>> 
>> 
>>> On Mar 7, 2024, at 10:54, Su Yue <glass.su@suse.com> wrote:
>>> 
>>> Hi, dear drbd folks
>>> Would you help review the patches for drbd-utils? Many thanks!
>>> 
>> Gentle ping.
> 
> So you apparently have some kind of test suite
> excercising this fencing handler?
> 

Yes. SUSE does have internal tests for drbd in HA environment.
The issue of patch1 was exposed with pacemaker-2.1.7 in SLES-15SP6.  
Patch2 was filed during debug process of the above issue. It looks like a 
small mistake when the judgement was introduced.

And the patches survived in our test scenarios.

Thanks,
Su
> Thank you for making us aware, and for your patience.
> Looks plausible.
> 
> Lars
>