[
https://issues.apache.org/jira/browse/JENA-602?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13843399#comment-13843399
]
Andy Seaborne commented on JENA-602:
------------------------------------
GSP POST does what is required - the graph destination is in the query string.
When used with files with content type (not HTML multi-part file uploads by
HTML form), insertion is streamed and transactional already.
The HTML form can be assumed to be just files (there are be many) with no
{{graph}} field. (It is better to POST to the content type, and raw stream than
use multipart form upload.)
Posting quads using GSP POST (and no {{?graph=}} or {{?default}}) will work in
the new UI codebase or when running --uber. Dynamic dispatch
({{SPARQL_UberServlet}} and {{SPARQL_UberFilter}}) will be the only way to
execution all operations in the new codebase.
What is left to do is a version of {{SPARQL_Upload.uploadWorker}} that streams
inside a transaction and does not need the {{graph}} form parameter.
{{REST_Quads}} adds functionality for whole dataset operations (get quads, POST
quads).
> Fuseki upload should avoid having to cache the data into memory where possible
> ------------------------------------------------------------------------------
>
> Key: JENA-602
> URL: https://issues.apache.org/jira/browse/JENA-602
> Project: Apache Jena
> Issue Type: Improvement
> Components: Fuseki
> Affects Versions: Fuseki 1.0.0
> Reporter: Rob Vesse
>
> Currently when uploading data files to Fuseki it always caches the data
> in-memory even if the underlying dataset is transactional. The reason stated
> in code is that since a form multipart request is used there is no way of
> knowing which graph the data is destined for because you may see the graph
> name after the data.
> With the ongoing Fuseki UI refactor it may be worth rethinking this, what we
> could do is require that any graph name is a query string parameter thus
> guaranteeing up front access to it and that only the data is POSTed in the
> form multipart request.
> If this change was made we could then support transactional uploads since
> you'd always know the graph to write to in advance and this would make file
> uploads significantly faster for systems like TDB.
--
This message was sent by Atlassian JIRA
(v6.1.4#6159)
