ng: reload now only reloads current dashboard

[stephanwlee]

Previously, we were reloading all stamped dashboards which is not the
behavior of the tf-tensorboard.

Side-effect: because we were reloading scalars dashboard when it was in
the background, the charts could more frequently appear tiny when the
reload kicked in. This change removes that bug.

[psybuzz]

Looks good!

1. Can the original bug (small charts) still technically occur, if an inactive dashboard's reload() is still pending?

2. This behavior now matches the Polymer tf-tensorboard, but I'd like to question the ideal UX. Feel free to defer/not answer this part, if it is beyond the scope.

Let's say a user opens Scalars at t = 0, opens Images, and autoreloads until t = 5000s, then returns to Scalars. For up to 30s, they'll experience 2 dashboards with 5000s difference in data freshness.

Would it make sense to mark dashboards as "dirty" when they are inactive during an autoreload, and explicitly reload() upon re-opening them? That way, the data freshness gap between any 2 dashboards should be about < 1 reload period.

[stephanwlee]

1. Can the original bug (small charts) still technically occur, if an inactive dashboard's reload() is still pending?

Correct

2. This behavior now matches the Polymer tf-tensorboard, but I'd like to question the ideal UX. Feel free to defer/not answer this part, if it is beyond the scope.
   Let's say a user opens Scalars at t = 0, opens Images, and autoreloads until t = 5000s, then returns to Scalars. For up to 30s, they'll experience 2 dashboards with 5000s difference in data freshness.

Beyond scope and the same behavior is what I am trying to achieve here. Your assessment is correct and it is out of the scope.