Ayal Baron has posted comments on this change.
Change subject: Invalidate filters on HSMs before rescanning extended VG ......................................................................
Patch Set 17: Code-Review+2
(2 comments)
.................................................... Commit Message Line 15: devices are filtered . The only way to invalidate filter now is flush LvmCache. Line 16: To avoid perfomance issues the patch introduces public method for filter invalidation Line 17: and calls it from the getDevicesVisibility. As far as getDevicesVisibility Line 18: is called to all added devices this allows to reload filters every time Line 19: after new devices has been mapped but before vgscan. no we don't. there is no problem on the spm, that works fine. This patch makes sure that the other hosts in the cluster work properly as well Line 20: Line 21: Change-Id: If1eeed1c203f2c8c73370987048565d665932299 Line 22: Bugzilla-Url: https://bugzilla.redhat.com/1022976
.................................................... File vdsm/storage/hsm.py Line 2004: boolean Line 2005: :rtype: dict Line 2006: """ Line 2007: visibility = self.scanDevicesVisibility(guids) Line 2008: lvm.invalidateFilter() absolutely not, that is raceful. you first scan for new devices, then update the filter to include anything new found in the scan. Line 2009: for guid in guids: Line 2010: if visibility[guid]: Line 2011: visibility[guid] = (os.stat('/dev/mapper/' + guid).st_mode & Line 2012: stat.S_IRUSR != 0)