Does Fuseki support conditional GSP requests?

[namedgraph]

Conditional requests meaning support for If-Match, If-None-Match, If-Modified-Since, If-Unmodified-Since headers in the Graph Store Protocol implementation.

[rvesse]

No

Mainly because the underlying storage doesn't typically have any way of easily determining whether or not a graph has changed other than loading the whole graph at which point you've already done 90% of the work (bar serialising the graph back to the client)

Such a feature could conceivably be implemented with a change aware wrapper DatasetGraph implementation, and by extending request handling

[namedgraph]

This might require changes at the TDB level (not sure), but couldn't this be implemented as a graphURI -> hash or graphURI -> eTag map? In which the value would only change during PUT and POST requests?

• PUT replaces the value
• POST adds hash of the request graph to the existing hash value
• DELETE removes the map entry

That way the current state of all graphs would be known at all times - using a rather small map?

[afs]

What about SPARQL Updates? Patch changes?

If-Match and If-Modified-Since are quite different.

Won't the epoch timestamp and the eTag both have to be persistent?

[namedgraph]

I'm speculating without knowing how SPARQL updates are implemented, but each affected graph is effectively applied a diff that the update operation produces? In that case the hash can be re-calculated without having to rehash full graph contents.

I haven't used If-Modified-Since much, but it also could be an entry in some sort of graphURI -> lastModified map, or attached to the graph as metadata somehow?

This would have performance implications obviously, and I don't know how large, but on the other hand the could be savings on the HTTP level like sending 304 Not Modified response instead of 200 OK with a response body.

[afs]

each affected graph is effectively applied a diff that the update operation produces?

No, there is no graph grouping in applying the changes.

[namedgraph]

OK. But there could be? :)

UUID per dataset that you suggest #2305 does not seem that useful to me, because updating one graph would invalidate the state of all graphs?

[afs]

See #2305.

[namedgraph]

Right, I forgot about that one 😅