fabianmenges commented on issue #3518: WIP Annotations URL: https://github.com/apache/incubator-superset/pull/3518#issuecomment-332604969 @mistercrunch > using table viz slices as input for annotation is fragile I'm using 'table viz' for horizontal lines and intervals, I'm using 'line viz' for point markers and line overlays, You will be able to pick your column for title and description in the 'Annotation overlay' configuration. Things like `onclick` should be handled in a separate PR, you would probably need to support some templating for this anyways. If you still think we need a special slice for it, feel free to create one, I'll hook it up. > can we break down the data driven annotations to a different PR? ... All the annotations types but _Formula_ are data driven. Are you asking me to support only Formulas for now? This would result in deleting a whole bunch of code and make some of the architecture look pretty useless. >I'm all for formula, vertical, and point markers, though I think we need a bit of a plan as to how all of these annotations come together... I'm making the popover dynamic for each Annotation type right now. It already is, if you consider that when its a _Formula_ you can enter the formula, but when it is any other type it is a "Slice Selector." For the slice selector you will be able to pass down 'Since' and 'Until', there is even more things we could pass down, e.g. (Filters, Mapping) This is also where you could specify a 'Tilte' and 'Description' and 'Link' coloumn for a dropdown of a list of all columns, but I probably won't be able to get that stuff into this PR. > for the data-driven annotations, there are considerations around whether they are fetched... Fetching them independently, asynchronously and on the client has many advantages. - We can fetch all annotation layers in parallel (async), I did not see anything in your current python project configuration that would allow me to do this in a reasonable way in python, (e.g. no gevent, etc). I feel very strongly about not doing this synchronously. - Separation of concerns: Your charts will still load fast even if there is something wrong with your annotations that might be stored on a different system. Even if we would implement this in python asynchronously we would still have to wait for the timeout of each annotation layer before we could return data. - If AnnotationLayers are hidden and you 'unhide' them you only have to load the missing data and not all the data > to make all of this comprehensive, we'll need a s-ton of tooltips I can take care of this. Would like to get that into a 2nd PR. This will get deployed to our production environment by Friday. It would be great if we could nail out the architecture questions beforehand. I hate writing JSON migration scripts. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
With regards, Apache Git Services
