Re: Comments on Partial Locking -01

["Tomasz Mikołajczyk" <miksiu@gmail.com>]

On Dec 11, 2007 9:53 PM, Mehmet Ersue <m_ersue@yahoo.de> wrote:
> IMHO:
> a) A partial lock after a global lock MUST fail.
> b) A global lock after a partial lock SHOULD be allowed.

I don't agree with the point a). In the case when a session A wants to
lock (partial lock) /foo/bar and /foo/baz, it can execute the global
lock first to protect against the situation, when during partial
locking /foo/bar another session (B) would lock partially /foo/baz. It
means that after the following steps:
1) global lock
2) partial lock of /foo/bar
3) partial lock of /foo/baz
4) global unlock
session A keeps lock on /foo/bar and /foo/baz.

Regards
Tomasz Mikolajczyk

> I) If above happens it is the easiest way of implementation (for the
> managing session and the managed node) if the global lock and partial lock
> are handled independently.
> I.e. the proposal would be that the partial lock survives the global lock
> and can be freed when the session decides to do so in its sw module
> hierarchy (whereever in the module the decision is taken). Since it is one
> session doing it I would assume the locks are done and released based on a
> FIFO principle.
> II) The same behaviour I would also assume in case of an overlapping partial
> lock.
>
> Martin Bjorklund wrote:
> > Sent: Tuesday, December 11, 2007 9:10 PM
> > To: ietf@andybierman.com
> ...
>
> > Suppose you write a routine that grabs a lock of /foo/bar/baz, does
> something and then
> > releases the lock.  Later, you might call this routine from code that
> > already have /foo/bar locked (or /foo/bar/baz/xxx/yyy).  Thus one can
> > argue that the code on the manager will get more complicated if we
> > don't allow overlapping locks.
> >
> > I will also argue that the code in the agent does not necessarily
> > become more complicated with overlapping locks (we have implemented
> > overlapping locks actually).  But maybe that wasn't your point.
>
> This is case II) and it seems to be the modular way of implementing a
> managing session.
>
>
> Balazs Lengyel wrote:
> > Sent: Tuesday, December 11, 2007 6:49 PM
> > To: Andy Bierman
> ...
>
> > However I don't see this as important, so it is up to Mehmet to fight for
> it. For the time being I will not add it to the draft.
> > Balazs
>
> I wasn't in the discussion when the clausel in RFC 4741 has been finalized.
> I believe it would make sense to allow case b).
> Nevertheless if there is nobody then me supporting this kind of change then
> I am willing to accept the decision which has been taken in the WG at that
> time.
>
> Cheers,
> Mehmet
>
>
>
> > -----Original Message-----
> > From: ext Balazs Lengyel [mailto:balazs.lengyel@ericsson.com]
> > Sent: Tuesday, December 11, 2007 6:15 PM
> > To: Mehmet Ersue
> > Cc: netconf@ops.ietf.org; Martin Bjorklund
> > Subject: Re: Comments on Partial Locking -01
> >
> > Hello Mehmet,
> > We could propose to allow a global lock after a partial lock
> > by the same session, it is even
> > logical in a way.
> >
> > However the main NETCONF standard says:
> > "      An attempt to lock the configuration MUST fail if an existing
> >        session or other entity holds a lock on any portion of the lock
> >        target."
> > This means we can not allow the global lock. I think what you
> > propose does not violate the
> > spirit of the RFC4741, but it does violate the exact rule in the RFC.
> >
> > Also we would have to deal with some tricky cases:
> > 1) session A issues partial-lock-A
> > 2) session A issues global-lock
> > - Should the partial-lock still be kept, or does this automatically remove
> all partial locks
> > for session-A for the specified datastore? (propose YES)
> > - after the successful global lock should it be possible to use a partial
> lock for the same
> > datastore by the same session ? (propose NO)
> > 3) session-A issues global-unlock
> > - should the partial locks stay alive ? (propose NO)
> >
> > regards Balazs
> >
> > Mehmet Ersue wrote:
> > >
> > > I wouldn't mind if the full lock does not fail when both of
> > the locks
> > > have been initiated by the same session.
> > > Then the question is whether the partial lock should
> > survive which seems
> > > to be useful.
> > >
> > > Cheers,
> > > Mehmet
> >

--
to unsubscribe send a message to netconf-request@ops.ietf.org with
the word 'unsubscribe' in a single line as the message text body.
archive: <http://ops.ietf.org/lists/netconf/>