RE: [RFC2518 Issue] PROPFIND 'allprop' usage

> -----Original Message-----
> From: Hartmut Warncke [mailto:hwarncke@Adobe.COM]
> Sent: 22 November 2000 13:12
> To: Lisa Dusseault
> Cc: WebDAV WG
> Subject: Re: [RFC2518 Issue] PROPFIND 'allprop' usage
> 
> 
> 
> Lisa,
> 
> 
> > One more question, because maybe, as Richard Humphreys 
> suggests, "depth
> > infinity" is the problem.
> >
> > Hartmut, does GoLive5 use depth: infinity PROPFIND requests 
> at all?  If yes,
> > does it use them for custom property discovery?
> 
> Yes, GoLive uses those infinity requests because it's a very 
> effiicient way to
> synchronize the WebDAV server content with the content of a 
> site on the local
> machine with  *o n e*  network request. We are requesting build-in DAV
> properties (like "getlastmodified" or "lockdiscovery") as 
> well as custom
> properties (which were PROPPATCHED by GoLive before) with 
> depth infinity.
> 
> To my mind depth infinity requests are -from the client 
> perspective - a very
> efficient and interesting WebDAV feature.
> Dropping  infinite requests would break GoLive 5!

The implications of PROPFIND with an infinite depth affect both the server
and/or the client. Supposing you have a large WebDAV enabled repository. The
PROPFIND request (a few hundred bytes) could potentially mean that (tens of)
GBs of data is to be returned. Even if the server could generate such a
large response (which is a feat in its own right), it will load the network,
and what will happen when the client receives it? Pity the poor user sitting
there for a long time (hours?), only to be told "timed out" or "out of
memory" or something similar.

> 
> By the way if a server would be allowed to refuse depth 
> infinity requests (like
> mod_dav with its "DavDepthInfinity" flag in its configuration 
> file), the client
> needs a mechanism to figure out if the server is willing to 
> respond to depth
> infinity requests or not. This mechanism has to be precisely 
> defined by RFC2518.
> 

Changing the specification could break existing products, which is a bad
thing. Backwards compatibilty should be maintained.

We think the way forward for the WebDAV specification is to allow servers
the ability to refuse such requests and inform the client. A mechanism
should be defined for the client to understand this. If the client received
a response which basically stated that the server was refusing to service an
infinite depth request, it could issue multiple requests with a Depth:1.

Regards

Shaun Hall
Xerox Europe

> 
> 
> > If not, then we could compromise our way through this by 
> stating that a
> > WebDAV server SHOULD (MUST?) respond to a PROPFIND depth-0 
> 'allprop' request
> > with all custom properties, but that it MAY respond to a 
> PROPFIND depth>0
> > 'allprop' request with a more limited set of properties 
> (suggested to be the
> > non-locking-related properties defined in RFC2518 presently).
> >
> > Would that work?
> 
> I think "allprop" should always return all properties (custom 
> and DAV build-in).
> Perhaps it makes sense to drop PROPFIND-ALLPROP requests with 
> depth infinity but
> please keep the PROPFIND-PROP requests with depth infinity 
> because they are a
> very powerful and efficient instrument. Dropping those 
> requests would lead us
> -in some situations- into a time consuming disaster with a 
> huge amount of
> network traffic.
> 
> Best, Hartmut
> 
> 
> > > -----Original Message-----
> > > From: Hartmut Warncke [mailto:hwarncke@Adobe.COM]
> > > Sent: Monday, November 13, 2000 8:07 AM
> > > To: Lisa Dusseault
> > > Cc: WebDAV WG
> > > Subject: Re: [RFC2518 Issue] PROPFIND 'allprop' usage
> > >
> > >
> > >
> > >
> > > > To clarify, does the GoLive 5 WebDAV client rely on using
> > > PROPFIND 'allprop'
> > > > requests to get all custom properties on **any** webDAV server?
> > >  or against
> > > > a specific webDAV server?
> > >
> > > On  **any** WebDAV server.
> > >
> > > > Have you tested this against various servers?
> > > > Does it work with all of them?
> > >
> > > Yes, it's working against IIS and mod_dav for example.
> > >
> > > Best, Hartmut
> > >
> > >
> > > > > -----Original Message-----
> > > > > From: w3c-dist-auth-request@w3.org
> > > > > [mailto:w3c-dist-auth-request@w3.org]On Behalf Of 
> Hartmut Warncke
> > > > > Sent: Monday, November 13, 2000 4:07 AM
> > > > > To: Lisa Dusseault; WebDAV WG
> > > > > Subject: Re: [RFC2518 Issue] PROPFIND 'allprop' usage
> > > > >
> > > > >
> > > > >
> > > > > Hi all,
> > > > >
> > > > > I have major concerns regarding the change of  the response
> > > on a PROPFIND
> > > > > 'allprop' request in the way you described it.
> > > > >
> > > > > Such a change would be  *v e r y  harmful*   for the GoLive 5
> > > > > WebDAV client.
> > > > > When we send a PROPFIND 'allprop' request we expect all
> > > > > properties which are
> > > > > defined on the resource, especially  the 
> Lockproperties and all custom
> > > > > properties defined by GoLive 5 (which we have 
> PROPPATCHED before).
> > > > >
> > > > > We are probably able to replace PROPFIND 'allprop' requests by
> > > > > PROPFIND 'prop'
> > > > > requests in future GoLive releases (which would be indeed much
> > > > > more efficient)
> > > > > but the suggested change in the protocol would be a  
> *disaster*
> > > > > for GoLive 5
> > > > > which is already in the box.
> > > > >
> > > > > Best, Hartmut
> > > > >
> > > > >
> > > > > Lisa Dusseault wrote:
> > > > >
> > > > > > Past discussions have shown that servers frequently 
> have trouble
> > > > > > implementing PROPFIND 'allprop'.  Jim asked me to 
> summarize the past
> > > > > > discussion & list the open issues so that we can get this
> > > > > fixed, if it can
> > > > > > be fixed, in revisions to 2518.
> > > > > >
> > > > > > There are already cases where not all properties 
> will be returned:
> > > > > > RFC2518: "In the case of allprop and propname, if a 
> principal
> > > > > does not have
> > > > > > the
> > > > > >    right to know whether a particular property exists then
> > > the property
> > > > > >    should be silently excluded from the response."
> > > > > >
> > > > > > John Stracke's proposal for reducing/specifying the scope of
> > > > > 'allprop', and
> > > > > > discussion of the motivation:
> > > > > >  -
> > > > > 
> http://lists.w3.org/Archives/Public/w3c-dist-auth/1999JulSep/0092.html
> > > > > >  -
> > > > > 
> http://lists.w3.org/Archives/Public/w3c-dist-auth/1999JulSep/0310.html
> > > > > >
> > > > > > It has been a point of discussion for Advanced Collections:
> > > > > >  -
> > > > > 
> http://lists.w3.org/Archives/Public/w3c-dist-auth/1999JanMar/0008.html
> > > > > > "Clients need to know whether the property is computed on
> > > the fly before
> > > > > > requesting it.  There is no way to find out.  The impact of
> > > > > computing on the
> > > > > > fly is especially significant when a client 
> requests allprop.
> > > > > There may be
> > > > > > other properties that are computed on the fly as well.
> > > DAV:getetag is
> > > > > > computed, and some versioning history properties may also
> > > be computed."
> > > > > >
> > > > > > Also in Versioning:
> > > > > >  -
> > > > 
> http://lists.w3.org/Archives/Public/w3c-dist-auth/1999JanMar/0075.html
> > > > > "There has also been a massive growth in the number 
> of available DAV
> > > > > properties.  PROPFIND allprop operations may lead to 
> very large
> > > > > responses even with Depth: 1, which would slow down 
> performance
> > > > > for users due to network speeds.  It might be 
> worthwhile to add this
> > > > > facet to the open issue ALLPROP_AND_COMPUTED."
> > > > >
> > > > > Also in ACLs, Babich argues that clients who request 
> 'allprop' don't
> > > > really
> > > > > want to see the ACL property, thus they ought to specifically
> > > ask for it.
> > > > >  -
> > 
> http://lists.w3.org/Archives/Public/w3c-dist-auth/1998JulSep/0101.html
> > > >
> > > > Several server implementors have voiced the opinion 
> that 'allprop'
> > should
> > > be
> > > > "put out of its misery" (GMC) or at least weakened.  
> Often this is
> > because
> > > > of standard or custom properties that must be 
> calculated by the server
> > > (e.g.
> > > > 'lockdiscovery'), and the calculation can become extraordinarily
> > expensive
> > > > with an 'allprop' of depth: infinity.
> > > >
> > > > The only server-side argument for keeping 'allprop' is that
> > > server-to-server
> > > > COPY requires it; but if anybody has implemented this 
> yet and can't use
> > > > 'propname' instead, please speak up.
> > > >
> > > > Summary:
> > > > There thus seems to be a consensus among server 
> implementors and those
> > > > designing new features for DAV.  What's missing in 
> order to resolve this
> > > > issue for fixing RFC2518 is input from clients.
> > > >
> > > > 1.  Is anybody aware of clients that rely on particular 
> properties being
> > > > returned in 'allprop'?  If the properties relied upon 
> include any more
> > > than
> > > > the set DAV:{creationdate, displayname, getcontentlanguage,
> > > > getcontentlength, getcontenttype, getetag, getlastmodified} (the
> > > properties
> > > > required for DAV level 1 support) I would be very 
> surprised.  Thus,
> > > servers
> > > > may be able to restrict the required property set to this set.
> > > >
> > > > 2.  Is anybody aware of clients that rely on 'allprop', 
> rather than
> > > > 'propname', for property discovery?  This would be a 
> more serious issue
> > if
> > > > major client implementations actually rely on doing 
> property discovery
> > > using
> > > > 'allprop', and attempt this against various 
> implementations of WebDAV
> > > > servers.
> > > >
> > > > These seem to be our options for modifying RFC2518 
> (remember, it has to
> > be
> > > a
> > > > simple mod):
> > > >  - deprecate 'allprop' and tell clients not to use it, 
> but to use
> > > 'propname'
> > > > instead
> > > >  - define 'allprop' to be the set of properties 
> required for DAV level 1
> > > > support (although servers could freely return more properties if
> > desired)
> > > >  - explicitly allow servers to return an error code (507?) for
> > properties
> > > > that were too expensive to calculate for a 'allprop' 
> request, but still
> > > > allowing the client to do property discovery through 'allprop'
> > > >
> > > > Please voice your preferences among these options, 
> objections, or other
> > > > ideas.
> > > >
> > > > thanks,
> > > > Lisa
> > > > Xythos
>

Received on Wednesday, 22 November 2000 11:17:39 UTC