[jira] [Comment Edited] (JENA-2270) Improve Fuseki data upload functionality.

["Andy Seaborne (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/JENA-2270?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17486350#comment-17486350 ] 

Andy Seaborne edited comment on JENA-2270 at 2/3/22, 10:43 AM:
---------------------------------------------------------------

No work in the UI, all server side refinement.
I've added a JIRA component tag "Fuseki UI" so tickets can mark this.

The UI isn't triggering a path that went to the old "upload" endpoint.

The {{file-upload}} JS component does use MIME type multipart uploads for files but it seems the GSP endpoint already supports this. 

The GSP endpoint supports regular POST/PUT to the dataset when no {{?default}} or {{?graph=}} is present.

The ticket is about clearing up the code in the server. That code is using [Apache Common FileUpload|https://commons.apache.org/proper/commons-fileupload/] which is {{javax}}. If the code is tidied up and usage of the java FileUpload is in only two functions. 

It is already only supporting {{multipart/form-data}}, looking the filename extension, and sending all data to the same destination. There are other, more complicated cases, like {{multipart/mixed}} where there is {{Content-Type}} per part.

For everyone's sanity, users and implementers, sending a TriG file for these cases is cleaner, especially if {{multipart/mixed}}-like transfer is required.

_Update:_

{{file-upload}} does not send multiple files in one HTTP request, nor did the old UI. They both do multiple parallel requests, which is fine. The server log has one request per file, and the request overlap.

The only difference is that pre-4.4 uses the dataset URL, and now the UI uses {{/datasets/data}}. Fuseki has "dispatch by content-type" on the dataset URL although the UI uses {{/data}}.

All good - the UI does not have to know the file extension to content-type mapping. JSON-LD 1.1 upload work if the extension if {{.jsonld11}} which is a new feature in Jena and no UI changes were necessary.



was (Author: andy.seaborne):
No work in the UI, all server side refinement.
I've added a JIRA component tag "Fuseki UI" so tickets can mark this.

The UI isn't triggering a path that went to the old "upload" endpoint.

The {{file-upload}} JS component does use multipart uploads for files but it seems the GSP endpoint already supports this. The GSP endpoint supports regular POST/PUT to the dataset when no {{?default}} or {{?graph=}} is present.

The ticket is about clearing up the code in the server.

Fuseki has "dispatch by content-type" on the dataset URL although the UI uses {{/data}}.

The UI does not allow the complicated case of "many files, different graph names" in one upload which the full {{application/multipart}} does. The name field can be after the file contents (odd but legal) so the receiver does not know the destination at the start of each part and it is buffered in memory.

For everyone's sanity, users and implementers, sending a TriG file for this case is cleaner.

_Update:_

{{file-upload}} does not send multiple files in one go, nor did the old UI. They both do multiple parallel requests.
The server log has one request per file, and the request overlap.

The only difference is that pre-4.4 uses the dataset URL, and now the UI uses {{/datasets/data}}.


> Improve Fuseki data upload functionality.
> -----------------------------------------
>
>                 Key: JENA-2270
>                 URL: https://issues.apache.org/jira/browse/JENA-2270
>             Project: Apache Jena
>          Issue Type: Improvement
>          Components: Fuseki
>    Affects Versions: Jena 4.4.0
>            Reporter: Andy Seaborne
>            Assignee: Andy Seaborne
>            Priority: Minor
>
> Continues JENA-2248.
> The upload code has become messy with duplication. 
> Functionality for HTML file upload {{<input type="file">}} has a dependency on code using {{javax}}, and not {{jakarta}}.
> {{HTML_FileUpload}} is dated and it buffers its input.
> This is a step towards having the option to remove HTML file upload capability.
> A new general data uploader for sending data to a dataset will cover the functionality better.
> The only use of HTML file upload not covered is when the graph name is in the multipart content. This predates SPARQL Graph Store Protocol which is a better approach because the graph name is available at the start and so the data can transactionally stream to the right destination. HTML file upload can, in theory, supply the graph name after the data.
> The UI "add data" tab uses GSP style to load data.



--
This message was sent by Atlassian Jira
(v8.20.1#820001)