Inflight _msearch requests are retried if visualizations are resized on dashboard

[adityasraghav]

Kibana version:
OSS 6.3.1

Elasticsearch version:
OSS 6.3.1

Server OS version:
Amazon Linux AMI 2018.03.0.20180811 x86_64 HVM GP2

Browser version:
Chrome 71.0.3578.98

Browser OS version:
OSX 10.13.6

Original install method (e.g. download page, yum, from source, etc.):
Download page linux tar

Describe the bug:
On the dashboard, if there are any _msearch requests in flight and the visualization hasn't loaded for any reason; if the window/visualizations are resized, the _msearch requests are retried.
Similarly, if an _msearch request times out and the error message is dismissed, all inflight _msearch requests will be retried.

Steps to reproduce:

1. Create a dashboard with a visualization.
2. Setup elasticsearch behind a proxy and send the /_msearch path to unreachable address to make sure requests remain in-flight till they timeout
3. Open up the dashboard and resize the browser.

Expected behavior:
Any inflight requests shouldn't be retried as it can overload elasticsearch, especially if there are timeouts happening.

Screenshots (if relevant):

[elasticmachine]

Pinging @elastic/kibana-app

[nreese]

#27440 sounds related and may be caused by the same underlying problem

[timroes]

There are basically two issues here. One is, that requests are not cancelled when they have become unnecessary. This is tracked via #17845 (and I updated the description with this very example). The other "issue" is - and I would like to track this via this issue - that visualizations emit a request when resizing. That is actually desired behavior, because visualizations might need different data depending on their size, e.g. the map visualizations actually request only data within the map's viewport, which change while resizing. The amount of visualization currently requesting size specific data is rather limited (only maps so far), why I think an opt-in might make more sense, since most visualization won't need that behavior. Also this issue might resolve itself, once we have refactored the way visualization state works (including #21644), in which case all information relevant to the data needs to be inside the vis expressions, and maps need to update those explicitly if they want to use that geo bounding box for querying. In that case we can safely disable this behavior, since every visualization needs to modify it's expression to change data fetching behavior.

[JacobBrandt]

visualizations emit a request when resizing. That is actually desired behavior, because visualizations might need different data depending on their size, e.g. the map visualizations actually request only data within the map's viewport, which change while resizing.

This ticket is not about a map. I don't see anything desirable about the behavior described in this ticket.

Also, the behavior you described as desirable for a map doesn't even happen. Resizing a map does not cause a request. So either that's another regression or that's not how it works.

[lukasolson]

I believe this has been addressed by #41204.