> -----Original Message----- > From: Wei Liu [mailto:[email protected]] > Sent: 31 October 2018 10:17 > To: Paul Durrant <[email protected]> > Cc: 'Marek Marczykowski-Górecki' <[email protected]>; xen- > [email protected]; Wei Liu <[email protected]>; Ian Jackson > <[email protected]> > Subject: Re: [Xen-devel] [PATCH] libvchan: create xenstore entries in one > transaction > > On Wed, Oct 31, 2018 at 10:08:26AM +0000, Paul Durrant wrote: > > > -----Original Message----- > > > From: Xen-devel [mailto:[email protected]] On > Behalf > > > Of Marek Marczykowski-Górecki > > > Sent: 30 October 2018 23:49 > > > To: [email protected] > > > Cc: Wei Liu <[email protected]>; Ian Jackson > <[email protected]>; > > > Marek Marczykowski-Górecki <[email protected]> > > > Subject: [Xen-devel] [PATCH] libvchan: create xenstore entries in one > > > transaction > > > > > > This will prevent race when client waits for server with xs_watch - > all > > > entries should appear at once. > > > > Watches should be put on the "state" key to avoid this kind of thing. > That key should always be the last thing modified and hence there should > never really be the need for xenstore transactions at all. > > AIUI libxenvchan doesn't have a "state" key.
It appears you're right. OMG. How on earth is one and supposed to know the state of the other end? This is an awful 'protocol'. > We can introduce one now > but we still need to handle old code which doesn't know about the new > key. > Indeed, but this really should be replaced by a protocol which actually has defined state transitions. Paul > I could be wrong about the protocol though. > > Wei. _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
