fix(coalesce): Big vhd-blocks cause exception

When handling bitmap, the GC is trying to write it in sm-config of the
VDI.
In the case of QCOW2, the bitmap can be bigger with 64KiB blocksize
and they can't be written in the VDI sm-config.
In this case, the GC would silently fail before running again and
failing again.
The written information is apparently never used again but the existence
of the key in sm-config is checked by the GC to know if it has work to
do.
By not raising, the GC would run normally but never stop.
Writing a dummy value then allow the GC to run and stop if it doesn't
have work to do.
Since the value used is always read from the VDI, the one from sm-config
is not really relevant.

Signed-off-by: Damien Thenot <damien.thenot@vates.tech>

Author: Nambrok

drivers/cleanup.py

@@ -769,7 +769,15 @@ def getAllLeaves(self) -> List["VDI"]:
 
     def updateBlockInfo(self) -> Optional[str]:
         val = base64.b64encode(self._queryCowBlocks()).decode()
-        self.setConfig(VDI.DB_VDI_BLOCKS, val)
+        try:
+            self.setConfig(VDI.DB_VDI_BLOCKS, val)
+        except Exception:
+            if self.vdi_type != VdiType.QCOW2:
+                raise
+            # Sometime with QCOW2, our allocation table is too big to be stored in XAPI, in this case we do not store it
+            # and we write `skipped` instead so that hasWork is happy (and the GC doesn't run in loop indefinitely).
+            self.setConfig(VDI.DB_VDI_BLOCKS, "skipped")
+
         return val
 
     def rename(self, uuid) -> None: