Memory leak in Chrome on auto-refreshing dashboard

[marius-dr]

Kibana version: 6.0.0-rc2

Elasticsearch version: 6.0.0-rc2

Server OS version: Windows 10

Browser version: Chrome 61, Chrome 62

Browser OS version: Windows

Original install method (e.g. download page, yum, from source, etc.): BC1 build

Description of the problem including expected versus actual behavior:
There seems to be a memory leak on a auto-refreshing dashboard in Kibana.
The settings I used were:

• Metricbeat processes dashboard
• Dark theme
• Full screen mode and F11 in the browser (tested the second time without this and still experiencing the leak, so might not be related)
• Auto-refresh on 5 seconds
• Time picker: Last 15 minutes

First time I ran out of memory and Chrome crashed the tab automatically after about 8h for about 7GB of free RAM at the start of the test.
Initial memory usage just for Chrome with only this tab starts at about 250MB. After 1h it got all the way to 892MB. Chrome Tab Manager reports that as javascript only memory.

You can see the evolution of Chrome memory usage in the actual metricbeat visualization on Memory usage per process.

Chrome is the Dark Green one. You can see that the test started at about 2AM and at about 11:30 AM the tab crashed on the test machine.

On Firefox, after about the same time (16h now since the start of the test), the tab is at about 384MB of RAM broken down like this:

I will update this issue as I get more information and run more tests. Next one on the list is Chrome on Ubuntu 16.04

cc: @epixa @LeeDr @Rasroh @bhavyarm

[stacey-gammon]

I believe this has been a long standing issue (#7427) that I fixed for 6.1 only (#13871). Was a bit nervous of the courier changes required for the fix to backport to 6.0.

[stacey-gammon]

... although I am curious why firefox didn't exhibit the same issue. Could you run this on the 6.x branch to see if that has the same crash?

[marius-dr]

I'll give it a go on 6.x today and see if the memory increase rate is similar. After setting the refresh rate on 1 minute instead of 5 seconds the rate of RAM increase has significantly declined,down to about 1gb per 24h from 8 gb per 8h.

[marius-dr]

It looks fine on 6.1.0-Snapshot so it seems to be the same issue that you fixed there. Same dashboard, same settings and it's hovering aroun 170mb-200mb for the past hour, no noticeable increase. With RC2 it was going from 200MB to 800mb in 1h. @stacey-gammon

[stacey-gammon]

sweet!

[chrisronline]

We have a request to backport this to 5.5 by @smassarik.

[stacey-gammon]

Unfortunately there are a ton of merge conflicts attempting to backport to 5.x because of a big visualization refactoring done in 6.x. Since there is a workaround (hard refresh the page, which I know isn't ideal), and I haven't heard any other requests to backport to 5.x, I don't think we can prioritize this at the moment.

It happens on the auto-refresh. The longer you can make that interval, the longer it will take for the memory leak to cause an issue.

[chrisronline]

@stacey-gammon

Is this line the offender? https://github.com/elastic/kibana/blob/5.5/src/ui/public/courier/fetch/request/request.js#L17

or https://github.com/elastic/kibana/blob/5.5/src/ui/public/doc_table/doc_table.js#L126

or both?