Improve attribute table performance display

[haubourg]

[haubourg]

Topic discussed on the 18 01 2017 in a technical meeting with Matthias Kuhn. Displaying large attribute table really need improvements.

related issues :

• http://hub.qgis.org/issues/12926 : "improve attribute table performance"
• http://hub.qgis.org/issues/10619 : "Attribute table 'Show features visible on map' option does not speed up loading of large postgis table " > closed two weeks ago, seems fixed in master
• filtering attribute with map extent as a default option prevent user from displaying the whole attribute table
• Pull request prototype to speed up attribute table loading https://github.com/qgis/QGIS/pull/2518 . Was not merge, uses mutlithreaded loading

[nyalldawson]

FYI - I'm also working on a related fix to improve performance when "show selected features" option is used.

[haubourg]

Thanks for keeping us informed Nyall :) Is that a part of http://hub.qgis.org/issues/12926 ? Are you trying the "Moved feature loader into its own thread" path @elpaso tested in https://github.com/qgis/QGIS/pull/2518 ? I 'd like to understand if we just do polishing or if a major refactor is required here.

[haubourg]

@mhugo @pblottiere @vmora : FIY guys, the attribute table performance is identified as a bottleneck for QWAT and QGEP projects (and INO most other projects too). If you have ideas or memory of things done before, please join the discussion (or the referenced tickets).

[nyalldawson]

Is that a part of http://hub.qgis.org/issues/12926 ? Are you trying the "Moved feature loader into its own thread" path @elpaso tested in qgis/QGIS#2518 ? I 'd like to understand if we just do polishing or if a major refactor is required here.

No - it's a separate issue. The caching mechanism isn't working correctly in this case and is forcing the entire table to load to show just the selected features. More of a specific bug fix than any large reworking.

[haubourg]

@nyalldawson thanks !

[haubourg]

Note: QGIS user group decided to fund Matthias Kuhn to work on that issue

[m-kuhn]

Note: The main target will be the opening speed for feature forms. If there is time left after looking at the loading speed for the relation reference widget, that will be spent on the attribute table. So no promise yet.

[haubourg]

@m-kuhn same here, priority 1 is on feature form loading, attribute table is P2.

[tudorbarascu]

Attribute table improvements already in master and 2.18 (see https://github.com/qgis/QGIS/pull/4230)

[haubourg]

@nyalldawson I just tested and it seems quite fast. Is there any more attribute table work in progress?

[nyalldawson]

@haubourg

I just tested and it seems quite fast. Is there any more attribute table work in progress?

No more plans here for now, sorry. There's still a lot we can/should do though.

[ponceta]

I get 2 seconds to load 20k point features and more than 50 sec to load the 10 k linestring features (pipes) on QGIS 2.18.7 Any hints on how to investigate this?

I have a 10 sec loading time on pgadmin for the same table (with all alternative geometries).

Loading it without joins takes approximatively 2 seconds. On QWAT pipe table we have 5 joins for values list (function, protection, pressure zones, status, materials)

[haubourg]

Ho, I didn't test using joins in QWAT context. Good catch Arnaud

[ponceta]

I optimized a bit the joins by omitting all unusefull joins and its much much faster.

For me it's OK.

Thanks for the speedup @nyalldawson !

[3nids]

What did you optimized? Column choice?

I would have thought that performance issues would have come from edit widgets rather than joins, that's interesting.

Le ven. 5 mai 2017 à 15:17, Arnaud Poncet-Montanges < [email protected]> a écrit :

I optimized a bit the joins by omitting all unusefull joins and its much much faster.

For me it's OK.

Thanks for the speedup @nyalldawson https://github.com/nyalldawson !

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/qwat/QWAT/issues/164#issuecomment-299461939, or mute the thread https://github.com/notifications/unsubscribe-auth/AAHxG3y9xoOxl3orMHH74TWIZ0TDuAODks5r2yFggaJpZM4LoKWE .

[haubourg]

@ponceta can you test again with 2.8.9, it seems a lot of performance regressions have been tackled last week.

[ponceta]

It's generally better as I felt. Since not many of my users are on 2.18 yet they can't feel it right now.