I couldn't immediately repro this on Linux M59, so we'll need a bit more information.

Could you record and upload trace of this occurring by navigating to about:tracing? Click "Manually select settings" and make sure at least "navigation", "loading", "loader", "v8", "net", "netlog", "toplevel", "ipc", and "blink" are selected (more is ok too).

It helps if the about:tracing tab and the tab you are recording are the only two tabs open (to make the trace easier to read).

Trace attached with those settings enabled.

Deleted: trace_Thu_Jun_08_2017_11.26.09_AM.json.gz 1.6 MB

trace_Thu_Jun_08_2017_11.26.09_AM.json.gz 1.6 MB Download

It looks like a navigation issue, but there are a ton of extension requests here so it is hard to track what's going on exactly. Would you mind re-doing this without any extensions turned on?

It also might help to check all the categories on the left panel, if it doesn't fill the buffer usage to 100%

will do -- upload a trace shortly

Since the issue does not happen in incognito mode, I figured it would be more useful to upload a trace where I open the url in non-incognito mode with all extensions manually disabled.  

If this still causes too much noise, I'll do another in incognito mode in a minute, but just note the reload problem doesn't happen there.  

Deleted: trace_Thu_Jun_08_2017_11.54.45_AM.json.gz 1.1 MB

trace_Thu_Jun_08_2017_11.54.45_AM.json.gz 1.1 MB Download

Thanks, it certainly looks like these navigations are being initiated in the renderer triggered by the load event of the page, but unfortunately the page's trace data was lost. This can happen if the tab is closed before the trace is properly stopped (sorry, should have mentioned that).

In any case, I think you could take a look at your onload/load handlers to inspect if they're doing anything weird.

New trace with extensions disabled, and I stopped the trace before I closed the tab.

I have scoured the code for onload/load and set event bp but I'm not seeing anywhere we we force a reload?

Deleted: trace_Thu_Jun_08_2017_12.37.50_PM.json.gz 4.1 MB

trace_Thu_Jun_08_2017_12.37.50_PM.json.gz 4.1 MB Download

I've set breakpoints on all Load events, and added my own handlers for all Load events.  When I hit the handlers, the stack is anonymous function that I passed, so I have no insight into how I got there.

The load event fires, the page renders, and then right after the beforeunload event fires, and it proceeds to unload, etc.  

So, this app is bundled with WebPack -- I'm debugging local with webpack-dev-server and I noticed they do a reload on hot code update, but I've breakpointed that and I never hit it.

What else can I try to determine what is the root cause of the unload / reload?

Hm, load flags are NORMAL | VERIFY_EV_CERT | MAIN_FRAME which indicates that this is *not* a reload, but is a normal navigation.

I'm not sure exactly how that could happen, as I believe window.location changes will set the load flags to revalidate the main resource.

You can try debugging further by recording a performance run with devtools, which afaik gives you stack traces over time.

Sorry I can't be of more help!

What are the paths that trigger a normal navigation?  

a) Click event on a <a>
b) Auto setting window.location = "some url"

What other possibilities are there for normal navigation that I can look for?

The only other I can think of is meta refresh.

e.g. <meta http-equiv="Refresh" content="1">

I was able to track it down to a change in event handling that was triggered by KeyCodes that changed from Chrome 58 to 59.  You can close the issue.  Thanks again for the help.

Thanks for the followup!