V0.12.22 - where did my models go?

Began as 1 thing… then multiplied.

Just updated to V0.12.22 (from 20 I think) and my carefully crafted and laid out models are gone - nothing in the model hub.

Logged out. Quit server. Restarted in Incognito mode (Chrome) still nothing.

So… it doesn’t feel like a caching issue, but why?

Anyway, they’re saved on disk so I’ll just reload them… the model navigator remains unfriendly so I went straight to Agent Ransack to find my models and then paste in the path. Oh. Can’t paste.

OK. After loading the first model I navigate to where the 2nd model is (in a completely different folder, with different name… turns out I misfiled it on creation) and from this…

I get the error “The chosen model is already in the project”, which I do not believe to be true.

Unfortunately (bit of an understatement there) since the model I really want is the one that is rejected as a duplicate, I need to remove the first one from the model hub… hmmm… no remove, only delete - and yes, it does exactly what it says, deletes the model.

Unfortunately #2, the other model is still rejected.

There’s nothing useful in the -v=3 output that was running, but here’s an example of trying to load “Photo-Z (minimal, 2-stage concat)” (I think it is the last attempt made)

INFO:perceptilabs.applogger:Request to endpoint 'session_proxy' took 0.0001s
[07/Sep/2021 08:38:23] "GET /json_models?path=A:/My%20Documents/IT/Programming/Python/Perceptilabs/Models/V%200.11%20Models/Photo-Z%20(minimal,%202-stage%20concat)//model.json&token=6gDXlbji3dNbOw1GpcvPNLwspd2aRWrgSkaCGoLJo5U HTTP/1.1" 200 97003
2021-09-07 08:38:23,127 - INFO - app.py:1884 - Request to endpoint 'session_proxy' took 0.0002s
INFO:perceptilabs.applogger:Request to endpoint 'session_proxy' took 0.0002s
[07/Sep/2021 08:38:23] "GET /models/70/?token=6gDXlbji3dNbOw1GpcvPNLwspd2aRWrgSkaCGoLJo5U HTTP/1.1" 200 329
[07/Sep/2021 08:38:23] "GET /models/71/?token=6gDXlbji3dNbOw1GpcvPNLwspd2aRWrgSkaCGoLJo5U HTTP/1.1" 200 318
[07/Sep/2021 08:38:23] "GET /models/73/?token=6gDXlbji3dNbOw1GpcvPNLwspd2aRWrgSkaCGoLJo5U HTTP/1.1" 200 266

Just to be complete in my reporting, having used Chrome and Incognito mode in Chrome and having restarted the server again I thought… let’s try a different browser… and this is what I find in Opera (NB I restored the deleted model from zip backup)

So, my models are still there - wherever “there” is - but Chrome is sulking.


  • Is there something I can do to regain access to the models as they were before the update - in Chrome?
  • If not, should I downgrade - or just wait a little bit?
  • Please add explicit warning that Delete deletes the model folder from disk and please provide a Remove from Hub option

Hey @JulianSMoore,
Really sorry to hear that the update caused issues, although I’m unsure if it’s related to the update or something else that’s going on, as we have not touched that area for a long time in the code :confused:
You can simulate this effect by creating a model in incognito and then going to the normal browser at which point you won’t see it any longer. We will be working on improving these kind of synching issue the upcoming few weeks.

Anyway, here are some answers as well as tips on how to fix this issue:

Please add explicit warning that Delete deletes the model folder from disk and please provide a Remove from Hub option

Noted, we will add it to the Delete Models warning.

As for removing from the Hub, the issue here was that the frontend for some reason did not show the models that were registered internally. (although the feature you suggest could still be useful)

You can clear out the internal database and re-import them by stopping the tool, removing the ~/.perceptilabs (Users/username/.perceptilabs for Windows) and then starting the tool again. You will now be able to import the models without the tool complaining.

I don’t think downgrading would do much, let’s start with clearing the internal database :slight_smile:

Hope that helps!

Thanks @Robert - I’ll try the db clear, that seems sensible :pray:

UPdate 14:26 - db Cleared, was able to add back both models. Problem gone :smiley:

Update #2 14:38

@robert I was wrong; issue reappears after server restart, and I quit and restated with -v = 3 in the same chrome browser. SO I removed the db again and after adding models again this time merely logged out and in again.

Even if I only add 1 model back, after logging out the model hub is empty until the db is deleted.

However I can only delete the db when the server is stopped so I have double checked that it seems to be logging out that causes the issue (on one occasion I logged out and just logged in again in a new incognito browser tab and the hub was empty.

I first thought it was only cause of incognito tab, but just tested as well and you are completely right. At every logout (and by the sounds of it, system restart) it seems to forget the existing models. Thanks for highlighting this!

Will push out a fix for this asap, expect something the upcoming day(s).

That’s a relief! I was just telling @Birdstream that ad hoc adjustments to environments (e.g. to run his cool inference!) sometimes have side effects and I really didn’t know whether my PL env was still in a good state. It seems it is, which is good :slight_smile:

I guess that implies another pre-release test to run… when there are thousands of daily users, “Where have all my models gone!?!” would be a support PITA :smiley:

Haha, a good pre-release test indeed :sweat_smile:

There is a new version available now that will show all your models, even if 0.12.22 didn’t, so no need to delete the database any longer :slight_smile: