I think mxzid should be fine, it should be a ZK-wide unique value. On Sun, Mar 12, 2017 at 6:03 PM, Jordan Zimmerman < [email protected]> wrote:
> So, is mxzid enough? Do we check version too? Is compared the data even > necessary? > > -JZ > > On Mar 12, 2017, at 2:14 PM, Scott Blum <[email protected]> wrote: > > Oh, this is related to CURATOR-391 https://github. > com/apache/curator/pull/202 > > You wrote: > >> Using only Stat version to note node change (for UPDATE event) is not >> enough. In a split brain if a node delete plus a recreate is missed, the >> Stat version can appear to be the same. > > > I never considered this corner case! So yeah, I agree that mxzid would be > more reliable. > > > On Sun, Mar 12, 2017 at 5:09 PM, Scott Blum <[email protected]> wrote: > >> I think "version" and "mzxid" should be equally reliable for determining >> that the contents changed. Whenever a change happens the version gets >> incremented while the mzxid gets set to the associated zk transaction id. >> So I don't see a problem with ifVersion == data.getStat().getVersion() but >> I haven't looked through all the PathChildrenCache code in a while, was >> there something else you were referring to going on in there? >> >> On Sun, Mar 12, 2017 at 1:23 PM, Jordan Zimmerman < >> [email protected]> wrote: >> >>> I'm working on CURATOR-391 and notice that TreeCache uses Mzxid to >>> determine if a ZNode has changed. PathChildrenCache uses a less reliable >>> method. I _think_ that Mzxid is enough. Does everyone else agree? >>> >>> -Jordan >> >> >> > >
