fix(LVMSR): deactivate unused LVM snap base before deletion (#130)

During normal operations, unused LV bases appear to stay enabled on slaves after they get deleted, causing an accumulation of activated deleted LVs on slaves.
This is mostly benign and this automatically gets resolved after a reboot of the slave. But if it accumulates a lot (on a big infra for example) it causes LVM commands to slow down so much that the whole SR gets unusable.

Signed-off-by: Antoine Bartuccio <antoine.bartuccio@vates.tech>

Author: klmp200

drivers/LVHDSR.py

@@ -1248,13 +1248,6 @@ def call_on_slave(self, args, host_refs, message: str):
             if not rv:
                 raise Exception('plugin %s failed' % self.PLUGIN_ON_SLAVE)
 
-    def _updateSlavesPreClone(self, hostRefs, origOldLV):
-        args = {"vgName": self.vgname,
-                "action1": "deactivateNoRefcount",
-                "lvName1": origOldLV}
-        message = "Deactivate VDI"
-        self.call_on_slave(args, hostRefs, message)
-
     def _updateSlavesOnClone(self, hostRefs, origOldLV, origLV,
             baseUuid, baseLV):
         """We need to reactivate the original LV on each slave (note that the
@@ -1292,6 +1285,16 @@ def _updateSlavesOnRemove(self, hostRefs, baseUuid, baseLV):
         message = f"Cleaning locks for {baseLV}"
         self.call_on_slave(args, hostRefs, message)
 
+    def _deactivateOnSlave(self, hostRefs, lvname):
+        """Tell the slave we need to deactivate the base image"""
+        args = {
+            "vgName": self.vgname,
+            "action1": "deactivateNoRefcount",
+            "lvName1": lvname}
+
+        message = f"Deactivating {lvname}"
+        self.call_on_slave(args, hostRefs, message)
+
     def _cleanup(self, skipLockCleanup=False):
         """delete stale refcounter, flag, and lock files"""
         RefCounter.resetAll(lvhdutil.NS_PREFIX_LVM + self.uuid)
@@ -1760,7 +1763,7 @@ def _snapshot(self, snapType, cloneOp=False, cbtlog=None, cbt_consistency=None):
         self.sr._ensureSpaceAvailable(size_req)
 
         if hostRefs:
-            self.sr._updateSlavesPreClone(hostRefs, self.lvname)
+            self.sr._deactivateOnSlave(hostRefs, self.lvname)
 
         baseUuid = util.gen_uuid()
         origUuid = self.uuid
@@ -1890,6 +1893,13 @@ def _finishSnapshot(self, snapVDI, snapVDI2, hostRefs, cloneOp=False, snapType=N
             util.SMlog("%s != %s != %s => deleting unused base %s" % \
                     (snapParent, self.uuid, snap2Parent, self.lvname))
             RefCounter.put(self.uuid, False, lvhdutil.NS_PREFIX_LVM + self.sr.uuid)
+
+            # The removed LV could still be activated on a slave host if it's
+            # part of a VM currently running there, we need to deactivate it
+            # before it gets removed to avoid a LV leak.
+            if hostRefs:
+                self.sr._deactivateOnSlave(hostRefs, self.lvname)
+
             self.sr.lvmCache.remove(self.lvname)
             self.sr.lvActivator.remove(self.uuid, False)
             if hostRefs: