Photos: Desktop client can't complete machine learning, but Android client can

[taivlam]

Description

1. Attempt to turn on machine learning on Ente Photos Desktop.
   • UX path: Open Ente Photos Desktop > Click on hamburger menu > Click on "Machine learning", under "Labs" > Toggle switch to "Enabled" for machine learning
2. Try to leave the desktop client open, only for it to stay on 0 out of total number of items.
3. Open Ente Photos on Android and navigate to the machine learning settings to turn machine learning on.
   • UX path: Open Ente Photos Android > Tap on hamburger menu > Tap on "General" > Tap on "Advanced" > Tap on "Machine learning" > Find both the "Enabled" and "Local indexing" switches toggled on already.
4. Observe Ente Photos Android get past 0 items completed and the whole initialization of machine learning come to an end.

This is despite the warning in the Ente Photos mobile apps that it's better to use desktop to initialize machine learning, instead of the mobile app. (For this, I plugged in my Android device and used Wi-Fi during this process.)

I'm not sure why the desktop client got hung up when initializing machine learning, while the Android client was fine.

System info

• Ente Photos on Android
  • GrapheneOS release 2024101200
  • Pixel 5a (barbet)
  • Ente Photos 0.9.58 from F-Droid
  • Logs: 2024-11-18.log
• Ente Photos on Linux
  • Manjaro Unstable
  • Ente Photos 1.7.6-1 from the AUR (ente-desktop-bin)
  • Logs: ente.log

Other details

On desktop, the settings are rather sparse:

while the settings on Android are more detailed (once the local models have been downloaded):

Version

1.7.6-1

What product are you using?

Ente Photos

What platform are you using?

Desktop - Linux

[mnvr]

This seems like a network issue in the routing between the ISP and Cloudflare (we use Cloudflare for proxying downloads), because the logs show that the desktop client is not able to download the images for indexing (no errors, the download just stalls until the app got restarted). The mobile client likely used a different (e.g. 5G) connection, so the download happened via a different network route.

Meanwhile, a workaround if you hit this issue again is to temporarily switch to a VPN or a different network where the desktop app is running.

[mnvr]

Also, if it happens again, it would be helpful to have a traceroute to see if there is some issue that can be spotted there

traceroute files.ente.io