resource-agents/bz2174911-LVM-activate-failover-with-missing-pvs.patch

From 51dd5d5d051aa3b3f0c104f8e80f212cd5780fc3 Mon Sep 17 00:00:00 2001
From: Oyvind Albrigtsen <oalbrigt@redhat.com>
Date: Tue, 14 Mar 2023 09:14:28 +0100
Subject: [PATCH] LVM-activate: failover with missing PVs

There area two changes included:

- Allow the system ID to be changed on a VG when the VG is
  missing PVs, as long as a majority of PVs are still present.
  This requires a recent version of lvm that supports the
  --majoritypvs option for vgchange.

- Use --activationmode degraded when activating LVs so that
  raid LVs can be activated when legs are missing, as long as
  sufficient devices are available for raid to provide all the
  data in the LV.

By David Teigland.
---
 heartbeat/LVM-activate | 82 ++++++++++++++++++++++++++++++++----------
 1 file changed, 64 insertions(+), 18 deletions(-)

diff --git a/heartbeat/LVM-activate b/heartbeat/LVM-activate
index e951a08e9c..f6f24a3b52 100755
--- a/heartbeat/LVM-activate
+++ b/heartbeat/LVM-activate
@@ -50,6 +50,8 @@ OCF_RESKEY_vg_access_mode_default=""
 OCF_RESKEY_activation_mode_default="exclusive"
 OCF_RESKEY_tag_default="pacemaker"
 OCF_RESKEY_partial_activation_default="false"
+OCF_RESKEY_degraded_activation_default="false"
+OCF_RESKEY_majority_pvs_default="false"
 
 : ${OCF_RESKEY_vgname=${OCF_RESKEY_vgname_default}}
 : ${OCF_RESKEY_lvname=${OCF_RESKEY_lvname_default}}
@@ -57,6 +59,8 @@ OCF_RESKEY_partial_activation_default="false"
 : ${OCF_RESKEY_activation_mode=${OCF_RESKEY_activation_mode_default}}
 : ${OCF_RESKEY_tag=${OCF_RESKEY_tag_default}}
 : ${OCF_RESKEY_partial_activation=${OCF_RESKEY_partial_activation_default}}
+: ${OCF_RESKEY_degraded_activation=${OCF_RESKEY_degraded_activation_default}}
+: ${OCF_RESKEY_majority_pvs=${OCF_RESKEY_majority_pvs_default}}
 
 # If LV is given, only activate this named LV; otherwise, activate all
 # LVs in the named VG.
@@ -191,6 +195,29 @@ logical volumes.
 <content type="string" default="${OCF_RESKEY_partial_activation_default}" />
 </parameter>
 
+<parameter name="degraded_activation" unique="0" required="0">
+<longdesc lang="en">
+Activate RAID LVs using the "degraded" activation mode. This allows RAID
+LVs to be activated with missing PVs if all data can be provided with
+RAID redundancy. The RAID level determines the number of PVs that are
+required for degraded activation to succeed. If fewer PVs are available,
+then degraded activation will fail. Also enable majority_pvs.
+</longdesc>
+<shortdesc lang="en">Activate RAID LVs in degraded mode when missing PVs</shortdesc>
+<content type="string" default="${OCF_RESKEY_degraded_activation_default}" />
+</parameter>
+
+<parameter name="majority_pvs" unique="0" required="0">
+<longdesc lang="en">
+If set, the VG system ID can be reassigned to a new host if a majority
+of PVs in the VG are present. Otherwise, VG failover with system ID
+will fail when the VG is missing PVs. Also enable degraded_activation
+when RAID LVs are used.
+</longdesc>
+<shortdesc lang="en">Allow changing the system ID of a VG with a majority of PVs</shortdesc>
+<content type="string" default="${OCF_RESKEY_majority_pvs_default}" />
+</parameter>
+
 </parameters>
 
 <actions>
@@ -524,24 +551,27 @@ lvm_validate() {
 		exit $OCF_ERR_GENERIC
 	fi
 
-	# Inconsistency might be due to missing physical volumes, which doesn't
-	# automatically mean we should fail.  If partial_activation=true then
-	# we should let start try to handle it, or if no PVs are listed as
-	# "unknown device" then another node may have marked a device missing
-	# where we have access to all of them and can start without issue.
-	case $(vgs -o attr --noheadings $VG | tr -d ' ') in
-	???p??*)
-		if ! ocf_is_true "$OCF_RESKEY_partial_activation" ; then
-			# We are missing devices and cannot activate partially
-			ocf_exit_reason "Volume group [$VG] has devices missing.  Consider partial_activation=true to attempt to activate partially"
-			exit $OCF_ERR_GENERIC
+	vg_missing_pv_count=$(vgs -o missing_pv_count --noheadings ${VG} 2>/dev/null)
+
+	if [ $vg_missing_pv_count -gt 0 ]; then
+		ocf_log warn "Volume Group ${VG} is missing $vg_missing_pv_count PVs."
+
+		# Setting new system ID will succeed if over half of PVs remain.
+		# Don't try to calculate here if a majority is present,
+		# but leave this up to the vgchange command to determine.
+		if ocf_is_true "$OCF_RESKEY_majority_pvs" ; then
+			ocf_log warn "Attempting fail over with missing PVs (majority.)"
+
+		# Setting new system ID will fail, and behavior is undefined for
+		# other access modes.
+		elif ocf_is_true "$OCF_RESKEY_partial_activation" ; then
+			ocf_log warn "Attempting fail over with missing PVs (partial.)"
+
 		else
-			# We are missing devices but are allowed to activate partially.
-			# Assume that caused the vgck failure and carry on
-			ocf_log warn "Volume group inconsistency detected with missing device(s) and partial_activation enabled.  Proceeding with requested action."
+			ocf_exit_reason "Volume group [$VG] has devices missing.  Consider majority_pvs=true"
+			exit $OCF_ERR_GENERIC
 		fi
-		;;
-	esac
+	fi
 
 	# Get the access mode from VG metadata and check if it matches the input
 	# value. Skip to check "tagging" mode because there's no reliable way to
@@ -601,7 +631,18 @@ lvm_validate() {
 do_activate() {
 	do_activate_opt=$1
 
-	if ocf_is_true "$OCF_RESKEY_partial_activation" ; then
+	if ocf_is_true "$OCF_RESKEY_degraded_activation" ; then
+		# This will allow a RAID LV to be activated if sufficient
+		# devices are available to allow the LV to be usable
+		do_activate_opt="${do_activate_opt} --activationmode degraded"
+
+	elif ocf_is_true "$OCF_RESKEY_partial_activation" ; then
+		# This will allow a mirror LV to be activated if any
+		# devices are missing, but the activated LV may not be
+		# usable, so it is not recommended.  Also, other LV
+		# types without data redundancy will be activated
+		# when partial is set.
+		# RAID LVs and degraded_activation should be used instead.
 		do_activate_opt="${do_activate_opt} --partial"
 	fi
 
@@ -661,11 +702,16 @@ clvmd_activate() {
 }
 
 systemid_activate() {
+	majority_opt=""
 	set_autoactivation=0
 	cur_systemid=$(vgs --foreign --noheadings -o systemid ${VG} | tr -d '[:blank:]')
 
+	if ocf_is_true "$OCF_RESKEY_majority_pvs" ; then
+		vgchange --help | grep '\--majoritypvs' >/dev/null 2>&1 && majority_opt="--majoritypvs"
+	fi
+
 	# Put our system ID on the VG
-	vgchange -y --config "local/extra_system_ids=[\"${cur_systemid}\"]" \
+	vgchange -y $majority_opt --config "local/extra_system_ids=[\"${cur_systemid}\"]" \
 		--systemid ${SYSTEM_ID} ${VG}
 	vgchange --help | grep '\--setautoactivation' >/dev/null 2>&1 && set_autoactivation=1