Perceptilabs browser loses state?

Issues/Feedback
#1

After rebuilding the Seeing-in-the-dark model (today), I had work to do, so the tab got lonely.

  • On returning to Perceptilabs it had logged me out.
  • I logged in again only to be told it seemed I had not created any models.
  • Checked the hub and there was 1 model - where there should have been ~8
  • Ctrl+F5 on the tab and models reappear

Not a major issue, but disconcerting (and what would have happened if I had not saved one of the models, suppose I were unable to get back to it to save it.)

Please could you state the criteria for auto-logoff and explain how the models are protected?

(Consider FMEA - Failure Mode Effect Analysis: not “how could that happen?”, but “what if it did?”

In this case, model is not saved but I get logged out automatically, then - internet down, your server dies, whatever - I can’t log in again. How is my data protected against loss? How could the model be saved?

I see the value in logon for you, but how can both parties interests be better served?)

#2

Hi Julian,

I’m sorry to hear what happened to you and I hope you’ll be willing to hear what we’ve done in response:

  • started testing with larger session timeouts in our development environment
  • opened a bug for the items in the hub not being shown until a browser refresh (work item # 1710)

Regarding the criteria for auto-logoff, it happens when the session has been idle for 30 minutes (up to 10 hours). The intention is to ensure that your work is not easily alterable by others when you’ve been idle. I know that some services have multi-day idle periods, one choice that those services have implemented is the ability to track activity from all logged in devices.
This is currently not implemented by us (yet?).

Regarding the question about how your data is protected, I’d like to explain where the information is stored. There are actually two copies of all the model saved; one is the files on the disk and another is within your browser’s IndexedDB. Unless one deletes the model from the ModelHub explicitly (or file system) and clears the browser cache, your model will still be on the computer.

#3

Hi @david.h

Now’ I’ve looked into what indexedDb is and how it is used, I can imagine it’s quite useful for running the Perceptilabs app, I could even imagine that the whole model is managed as db objects as a simple graph of nodes and edges stored in that database.

In which case, I can also imagine that the the model is not “saved” there - as in more durable copy from a memory resident master - but rather embodied in indexedDb, which is then then serialised to create the model json.

So my comment in slack was a little off-base except to the extent that if internet access for login is still required, there is an unnecessary barrier to updating the image saved to disk as model.json (even though indexedDB is also secretly stored on disk)

I guess I’m uncomfortable with the idea that it seems I can’t do anything with my models unless I’m logged in - so there’ll be no heading off to the middle of nowhere to create my humanity saving/destroying AI in peace and quiet (unless I take a starlink base station with me, your servers aren’t being temperamental, etc.)

That said, many, many thanks for the really, really clear statements that you noted what was said and are taking specific steps in response. That kind of clarity and transparency really builds confidence. Great team/corporate ethic going there!