vdsm: get gluster volume info from any gluster peer

The function _get_gluster_volinfo query the glusterfs volume
info the the storage server, this is translated to the gluster
client adding the parameter --remote-host which limits the
query to one server, so we are converting the storage server as a
single point of failure, if it is not available, it can led to cluster
outtage. The proposed changed let the cluster cli to use any
available gluster peer.

Signed-off-by: José Enrique Gutiérrez Mazón <[email protected]>
Signed-off-by: Albert Esteve <[email protected]>

lib/vdsm/storage/storageServer.py

@@ -387,9 +387,21 @@ def _get_gluster_volinfo(self):
             volinfo = superVdsmProxy.glusterVolumeInfo(self._volname,
                                                        self._volfileserver)
             return volinfo[self._volname]
-        except ge.GlusterCmdExecFailedException as e:
-            log.warning("Failed to get volume info: %s", e)
-            return {}
+        except ge.GlusterException as e:
+            # The remote host may be down.
+            # If we are running on a hyperconverged system the gluster client
+            # can use one of the other gluster servers.
+            log.info("Failed to get volume info from remote server %s: %s",
+                     self._volfileserver, e)
+            log.debug("Trying to get volume info from backup servers: %s",
+                      self._options)
+            try:
+                volinfo = superVdsmProxy.glusterVolumeInfo(self._volname)
+                return volinfo[self._volname]
+            except ge.GlusterException as e:
+                log.warning(
+                    "Failed to get volume info from backup servers: %s", e)
+        return {}
 
 
 class NFSConnection(Connection):

tests/storage/storageserver_test.py

@@ -250,6 +250,40 @@ def test_glusterfs_cli_missing(self, monkeypatch):
         gluster = GlusterFSConnection(id="id", spec="192.168.122.1:/music")
         assert gluster.options == ""
 
+    @pytest.mark.parametrize("userMountOptions", [
+        '',
+        'backup-volfile-servers=192.168.122.1:192.168.122.2',
+    ])
+    def test_glusterfs_retry_withotut_volfile_server(self, monkeypatch,
+                                                     userMountOptions):
+        """
+        The test will fail if called with volfileserver, simulating the case
+        when a remote server is down. If the caller try again without
+        volfileserver the call will succeed, simulating the case when running
+        on a hyperconverged system when the local glsuter glsuter client can
+        use one of the connected servers.
+        """
+        monkeypatch.setattr(storageServer, 'supervdsm', FakeSupervdsm())
+        monkeypatch.setattr(gluster_cli, 'exists', lambda: True)
+
+        def glusterVolumeInfo(volname=None, volfileServer=None):
+            assert volname == "music"
+            if volfileServer is not None:
+                raise ge.GlusterException()
+            return {'music': {'brickCount': '2',
+                              'bricks': ['192.168.122.1:/tmp/music',
+                                         '192.168.122.2:/tmp/music']
+                              }
+                    }
+        storageServer.supervdsm.glusterVolumeInfo = glusterVolumeInfo
+        gluster = GlusterFSConnection(id="id", spec="192.168.122.3:/music",
+                                      options=userMountOptions)
+        expected_volinfo = {'brickCount': '2',
+                            'bricks': ['192.168.122.1:/tmp/music',
+                                       '192.168.122.2:/tmp/music']
+                            }
+        assert gluster.volinfo == expected_volinfo
+
 
 class TestGlusterFSNotAccessibleConnection: