Immich ML became unhealthy

[haroldfinch-oss]

Has anyone been noticing this lately? Usually happens when I am scrolling through my timeline on my Android App. When I close the app, it says it became healthy. However, docker doesnt show it as unhealthy at all. Only in the server logs as shown below.

[Nest] 23  - 02/17/2026, 9:10:24 PM     LOG [Api:MachineLearningRepository~lnmqytpk] Machine learning server became unhealthy (http://immich-machine-learning:3003).
[Nest] 23  - 02/17/2026, 9:10:40 PM     LOG [Api:WebsocketRepository] Websocket Disconnect: RHsk1HzRL_Ag918GAAAR
[Nest] 23  - 02/17/2026, 9:10:48 PM     LOG [Api:WebsocketRepository] Websocket Connect:    aqEsa82uahF-cgk5AAAT
[Nest] 23  - 02/17/2026, 9:10:52 PM     LOG [Api:MachineLearningRepository~lnmqytpk] Machine learning server became healthy (http://immich-machine-learning:3003).

[Mraedis]

No, this is not normal. What are your host system's specs?

[Mraedis]

Unless your ML container is somewhere else they should be communicating directly within the docker network so no.
If you want to see whether the proxy is doing anything you could connect through the local IP and see if that triggers anything.

[haroldfinch-oss]

Do you think having all my services on the same network inside docker be causing this issue?

[Mraedis]

It is certainly not good practice, you should have an internal network for each stack/compose and an additional proxy network if you want to connect any to the outside through a proxy.

[haroldfinch-oss]

It is certainly not good practice, you should have an internal network for each stack/compose and an additional proxy network if you want to connect any to the outside through a proxy.

Indeed, I think this might be the problem afterall, all my services are on a single internal network. I will implement this soon and test. Can you link any specific youtube videos that might be helpful to get me rolling? I am pretty new to networking.

[haroldfinch-oss]

I solved this problem by moving my thumbs folder to an SSD drive!
Thumbmnails load lightning fast and the server never becomes unhealthy unless some other service is hogging the resources. I guess its to do with Immich giving ML server less priority when its being throttled.

I didnt touch the network settings at all, a simple thumbs migration did wonders.

[Muehlima]

I have the same issue, random cpu spikes which are triggerd when the ml server becomes unhealthy and end as soon as its healthy again. I have no monitoring going so i have not yet catched what's exactly spiking. But what i can say, that i do not have a reverse proxy running.

[haroldfinch-oss]

Huh thats interesting, It thought the API requests for thumbnails might have been choking my reverse proxy. I havent changed my configuration at all yet but good that you also have faced similar issue and we can try to understand the root cause first, and maybe we can ask to turn this into an issue. Can you please share your logs and host system specs?

[haroldfinch-oss]

I solved this problem by moving my thumbs folder to an SSD drive!
Thumbmnails load lightning fast and the server never becomes unhealthy unless some other service is hogging the resources. I guess its to do with Immich giving ML server less priority when its being throttled.