Having plugins appear automatically as soon as is_active() checks is done

• TensorBoard version
  1.14.0a20190319
• TensorFlow version if different from TensorBoard
  nightly
• OS Platform and version (e.g., Linux Ubuntu 16.04)
  Linux
• Python version (e.g. 2.7, 3.5)
  3.5

When loading profile from gs://cloud-tpu-tools-tf13, the profile tab doesn't show up immediately. A few extra manual refreshes are required to have the profile dashboard appear. The user may not expect the extra refreshes.

The change in behavior is a consequence of https://github.com/tensorflow/tensorboard/pull/1931 which changed the is_active() computation to be asynchronous, so it's now possible that the profile dashboard shows up as inactive initially and either reloading the page or clicking the TensorBoard "refresh" icon once or twice will be necessary to have the plugin show up as active. This is the same behavior as the projector and text plugin.

I think the asynchronous computation in PR/1931 is the right way to go.
As @nfelt point out, the right answer is to change the is_active() checks to be done with 1 request per plugin (rather than all at once), and having plugins appear automatically as soon as those requests finish. Therefore opening a feature request for this here.

Start by reading PR 1931 and tracing the is_active() checks for the profile, projector, and text plugins. Determine how the frontend handles each asynchronous result and the TensorBoard refresh action. Done means each plugin appears automatically when its own activity request finishes, without manual page or toolbar refreshes.