Headless Ubuntu Client won't continue without restarting service

[Knutowskie]

So i set up a headless Ubuntu client. It folds on 3 of 4 CPUs so far. After finishing a WU it sometimes "gets stuck" and does: nothing.

When i notice, i logon to the machine and restart the fah-client service. Then it continues as normal.

How can one avoid that?

I mean, i could write a script, that is checking the WU folder and if empty for several minutes, restart the service. But i guess this behaviour is something like a bug.

Logs are clean in the webinterface. Are there any logs on the machine that would help?

[arisu]

Run the command "fahctl state" when it is stuck and post the output here (if you set a passkey, remember to censor it).

[Knutowskie]

Run the command "fahctl state" when it is stuck and post the output here (if you set a passkey, remember to censor it).

Missing python3-websocket library

Nice. So the installer didn't install all needed libs

[arisu]

Seems like it's a missing dependency. I have a small list of bugs that I'm going to report once I make a GitHub account, I'll add that to it.

Just install that package and fahctl should work.

[Knutowskie]

well it doesn't. now its stuck on DOWNLOAD.

Code: Select all

{
  "info": {
    "version": "8.4.9",
    "url": "https://v8-4.foldingathome.org/",
    "os": "linux",
    "os_version": "6.8",
    "cpu": "arm64",
    "cpu_brand": "Neoverse N",
    "cpus": 4,
    "gpus": {},
    "mach_name": "KNUT-02",
    "hostname": "KNUT-02",
    "mode": "Release",
    "revision": "360fe71b1bd05bb89814bfb97b73a5bda84802d6",
    "has_battery": false,
    "on_battery": false,
    "id": "75Z2m1_9wFIgwlWFPXxaRAp6efQrBwAGUEah5k-vQ0U",
    "account": "blablablablabla"
  },
  "config": {"user": "Knutowskie", "team": 70335, "passkey": "1234abc1234abc1234abc", "cause": "unspecified"},
  "groups": {
    "": {
      "config": {
        "on_idle": false,
        "on_battery": true,
        "keep_awake": true,
        "paused": false,
        "finish": false,
        "beta": false,
        "cuda": true,
        "key": 0,
        "cpus": 3,
        "gpus": {}
      },
      "wait": "2025-04-03T07:01:34Z",
      "failed_wus": 0,
      "lost_wus": 0,
      "failed": ""
    }
  },
  "units": [
    {
      "group": "",
      "number": 17,
      "cpus": 3,
      "gpus": [],
      "state": "DOWNLOAD",
      "paused": false,
      "id": "e2442sHmBtoNylq9WXqukAI9tWTRaaPB999FkWpVEvk"
    }
  ]
}

[muziqaz]

You are trying to fold on ARM 4 core CPU. I'd guess its some sort of Pi device, or something other low powered in nature.
Abandon your efforts, not worth it. We barely ever enable anything for these type of CPUs, as they are too slow for FAH

[arisu]

You are trying to fold on ARM 4 core CPU. I'd guess its some sort of Pi device, or something other low powered in nature.
Abandon your efforts, not worth it. We barely ever enable anything for these type of CPUs, as they are too slow for FAH

The Neoverse N series of ARM processors are for high-performance computing in datacenters. OP probably is running this in the cloud since most of those processors have more than 64 cores and he only has 4. It should be more than enough to fold, but maybe the assignment server is confusing it with a weaker processor?

[calxalot]

Cpus is set to 3, which could be a problem. Please try setting to 4 cpus.

[muziqaz]

You are trying to fold on ARM 4 core CPU. I'd guess its some sort of Pi device, or something other low powered in nature.
Abandon your efforts, not worth it. We barely ever enable anything for these type of CPUs, as they are too slow for FAH

The Neoverse N series of ARM processors are for high-performance computing in datacenters. OP probably is running this in the cloud since most of those processors have more than 64 cores and he only has 4. It should be more than enough to fold, but maybe the assignment server is confusing it with a weaker processor?

Servers are not confused.
We cannot distinguished between high performance or low performance 4 core CPUs. We blanket blacklist ARM CPUs on most projects because majority of users are on low power PIs

[arisu]

That is what I meant by confused. Of course servers are inanimate objects that can't feel anything but it was easier than saying that the servers made an incorrect decision due to having insufficient information and making an assumption about the device (yes, I know servers cannot assume either )

I guess the small number of high-performance ARM users out there are not worth the extra dev work needed to return information about CPU model.

[muziqaz]

Small number of users with access to high performance ARM should try and get access to more than 4 cores
But then again, we rarely bother with more specific constraints for ARM other than just outright disallowing the assignments

[arisu]

In theory you could spoof the number of CPUs by hardcoding the return value of sysconf(_SC_NPROCESSORS_ONLN) to something higher than 4. That could give you a larger WU, and it wouldn't violate the EULA since you aren't tampering with the client. You'd want to configure it to use just 4 threads though because it would still be inefficient to fold on more threads than you have CPU cores.

I wouldn't recommend anyone do it unless they know what they're doing, because it would be easy to give yourself a WU that you could never possibly fold within the deadline (that is why I'm not describing exactly how to adjust _SC_NPROCESSORS_ONLN).

[calxalot]

cpu_brand is the cpu model.
But I don’t expect there is a desire to further complicate the constraints.

[muziqaz]

No, spoofing would be first class ticket to a blacklist

[arisu]

It's no different than starting on a system with many cores and then changing the slider to limit it to fewer (ie cannot interfere with the research).

I suppose the EULA should be updated, since it is an operating system configuration change and not a change to the code of the client or cores, nor tampering with communication with the server (which is what the current EULA covers).