Able to crash web browser with bad choice of histogram interval

[jimmyjones2]

I've got a field with a maximum value of around 5,000,000. If I do a histogram agg with an interval of 1 my web browser dies (ES returns the data fine as there are only a handful of documents)

[rashidkpc]

This is tough to handle, elasticsearch can return really huge JSON responses, and we don't really know how big its going to be. We could possibly check the content-length and bail before trying to JSON parse it.

[jimmyjones2]

In this case, elasticsearch wasn't returning much data at all, so I think it was the code trying to draw all the bars, but this is probably an unusual scenario. Is it worth getting elasticsearch to calculate the max and min first, then can figure out how many buckets it'll end up creating and if that's over a threshold, refuse? Date histogram already does this, but I guess it can figure it out without having to go to elasticsearch.

[aaronleesmith]

Why not try and solve this at the Kibana ES proxy layer? You could watch the content-length there and refuse to send the client the response if it's too large.

[Hossrod]

I've run into this a few times due to an occasional bad value of a single record getting into the system. I don't have control of data coming in. I started a discussion about this on elastic search site and they suggested to update this issue. https://discuss.elastic.co/t/add-some-safety-around-out-of-control-graphs/55202

As @jimmyjones2 says, I believe this is an issue with graph rendering and not size of query results.

Maybe a configurable timeout for graph generation would be good. As it stands now, when this happens, it breaks entire dashboards that make use of that graph. I can't even try to edit the graph or dashboard because they hang just trying to load them to edit.

[epixa]

Just an update to say that this is still happening on the latest master.