[SUPPORT] blakeblackshear - Frigate

[Lunch]

5 minutes ago, Bruceflix said:

Have you added "--runtime=nvidia" as extra parameter?

Yes sir!

 

And re your second reply I will try again now as didnt want to run before, and will report back. Thank you very much for your help

[Lunch]

OK, well I tried adding that to the top, and unfort i wont even load up at all now(I also removed the individual qsv commands under each camera)

[yayitazale]

5 hours ago, Lunch said:

OK, well I tried adding that to the top, and unfort i wont even load up at all now(I also removed the individual qsv commands under each camera)

Please read the full documentation. You are using old format hardware accel presets.

 

Are you planning to use nvidia GPU just to HW accel (case A) or/and you want to detect (case B)?

 

If this is case A:

• Did you install the nvidia driver plugin?
• Can you see the GPU doing nvidia-smi in the unraid console?
• Did you passtrought correctly the GPU using the GPUID (taken from the nvidia driver plugin)?
• Did you see the GPU inside the docker container of frigate doing nvidia-smi?
• Just in case, are you using the GPU in any VM on unraid?

If this is case B , all the questions of the case A plus:

• Did you select the nvidia branch at the frigate app install?
• Did you created the models?

Please also recheck your config file to meet the the docs: https://docs.frigate.video/

[flyize]

Is there an easy way to view the 24/7 recordings? Preferably something inside of Home Assistant (or maybe Frigate itself)? I'm looking to replace the ability that my Nest cameras have easily scroll back to a specific date and time. Or do I need external software for this?

[Lunch]

11 hours ago, yayitazale said:

Please read the full documentation. You are using old format hardware accel presets.

I have read through to the best of my(layman!) knowledge, but the old format hardware accel presets....they're at frigate.video under hardware acceleration in the documentation so I assumed these were right?

11 hours ago, yayitazale said:

 

Are you planning to use nvidia GPU just to HW accel (case A) or/and you want to detect (case B)?

 

If this is case A:

• Did you install the nvidia driver plugin?

Yes I did this

• Can you see the GPU doing nvidia-smi in the unraid console?

Yes when I type nvidia-smi it sees the Quadro GPU

• Did you passtrought correctly the GPU using the GPUID (taken from the nvidia driver plugin)?

Yes got the guid and put in template

• Did you see the GPU inside the docker container of frigate doing nvidia-smi?

No, I loaded up frigate aand best I got was the blank screens of all 6 cameras without any vision

• Just in case, are you using the GPU in any VM on unraid?

No I have another AMD passed through to my Windows 10 gaming VM only, and am not attempting to touch this

If this is case B , all the questions of the case A plus:

• Did you select the nvidia branch at the frigate app install?

No I didnt, because I was planning on doing both I thought not to was correct?

• Did you created the models?

No I hadnt got this far at all yet

Please also recheck your config file to meet the the docs: https://docs.frigate.video/

Roger, Ill try and have another look, but for quicksync with the iGPU of the Intel CPU it works well, just wanted to try and decrease the load on the CPU with what I had on hand

Hi there, thanks for taking the time, to be honest Case A would have been fine, case B even better...Ive replied under each of your line replies to make it easier

 

Thanks again for your support, and this amazing software!

 

Gareth

[yayitazale]

1 hour ago, Lunch said:

I have read through to the best of my(layman!) knowledge, but the old format hardware accel presets....they're at frigate.video under hardware acceleration in the documentation so I assumed these were right?

Hi there, thanks for taking the time, to be honest Case A would have been fine, case B even better...Ive replied under each of your line replies to make it easier

 

Thanks again for your support, and this amazing software!

 

Gareth

Open the terminal of the container and do nvidia-smi to check if the passthrought of the GPU is correct. There some issues on systems with 2 GPU dedicated cards so maybe this is what is happening to you. In any case, you can check the logs of the container or the logs inside the webUI to see what is wrong. About the HW accel presets, I suggest you to use this for nvidia:

 

ffmpeg:
  hwaccel_args: preset-nvidia-h264

 

For case B (I suggest you to first have the case A running well with nvidia GPU), you must select the option 2 at the deployment (nvidia version), but first you must create the models using the same GPU that you are going to use in frigate and the APP from store called "tensorrt-models" (follow the instructions and requirements of the template).

 

And for you info, I just maintain the frigate templates, I'm not a core developer so give the cheers to them on github not to me.

[Flubster]

@Lunch I had the same issue as you, but in my case (and I suspect this isn't the best solution although worked for me)

 

I checked config had the hwaccel line

 

I added "--gpus all" into extra parameters

 

And removed nvidia_visible_devices key in the template.

 

I do question whether i had tried --runtime=nvidia but in the end the above options worked for me so stopped there. (GPU now shows in frigate Web ui and smi)

 

I needed to get frigate back up quick so didn't have time to experiment with parameters.

 

YMMV etc.

 

Flub.

 

[yayitazale]

1 hour ago, Flubster said:

@Lunch I had the same issue as you, but in my case (and I suspect this isn't the best solution although worked for me)

 

I checked config had the hwaccel line

 

I added "--gpus all" into extra parameters

 

And removed nvidia_visible_devices key in the template.

 

I do question whether i had tried --runtime=nvidia but in the end the above options worked for me so stopped there. (GPU now shows in frigate Web ui and smi)

 

I needed to get frigate back up quick so didn't have time to experiment with parameters.

 

YMMV etc.

 

Flub.

 

In theory, using the GPUID variable and the extra --runtime=nvidia parameter should work. If there are problems with certain GPUs we can talk with @ich777 to see if the driver is loading everything OK or maybe we have to open a issue on github.

[Lunch]

On 4/11/2023 at 3:28 AM, flyize said:

Is there an easy way to view the 24/7 recordings? Preferably something inside of Home Assistant (or maybe Frigate itself)? I'm looking to replace the ability that my Nest cameras have easily scroll back to a specific date and time. Or do I need external software for this?

Bud its in Frigate itself...and if you want to watch back in home assistant al you need is the Frigate Proxy linked to your main Frigate entity wherever it is on your home network

[Lunch]

On 4/11/2023 at 3:15 PM, yayitazale said:

Open the terminal of the container and do nvidia-smi to check if the passthrought of the GPU is correct. There some issues on systems with 2 GPU dedicated cards so maybe this is what is happening to you. In any case, you can check the logs of the container or the logs inside the webUI to see what is wrong. About the HW accel presets, I suggest you to use this for nvidia:

 

ffmpeg:
  hwaccel_args: preset-nvidia-h264

 

For case B (I suggest you to first have the case A running well with nvidia GPU), you must select the option 2 at the deployment (nvidia version), but first you must create the models using the same GPU that you are going to use in frigate and the APP from store called "tensorrt-models" (follow the instructions and requirements of the template).

 

And for you info, I just maintain the frigate templates, I'm not a core developer so give the cheers to them on github not to me.

Thanks mate!  I will try tmrw as run out of time tonight with family commitments...I know its not you as the developer but I am also aware of just hoe much work you do so to you both I am so grateful for!

[Lunch]

6 hours ago, Flubster said:

@Lunch I had the same issue as you, but in my case (and I suspect this isn't the best solution although worked for me)

 

I checked config had the hwaccel line

 

I added "--gpus all" into extra parameters

 

And removed nvidia_visible_devices key in the template.

 

I do question whether i had tried --runtime=nvidia but in the end the above options worked for me so stopped there. (GPU now shows in frigate Web ui and smi)

 

I needed to get frigate back up quick so didn't have time to experiment with parameters.

 

YMMV etc.

 

Flub.

 

Flub...love that name lol

Thanks so much bud will be trying all this asap...and will report back...thanks so much bud!

[Lunch]

9 hours ago, Flubster said:

@Lunch I had the same issue as you, but in my case (and I suspect this isn't the best solution although worked for me)

 

I checked config had the hwaccel line

 

I added "--gpus all" into extra parameters

 

And removed nvidia_visible_devices key in the template.

 

I do question whether i had tried --runtime=nvidia but in the end the above options worked for me so stopped there. (GPU now shows in frigate Web ui and smi)

 

I needed to get frigate back up quick so didn't have time to experiment with parameters.

 

YMMV etc.

 

Flub.

 

Hey sorry Flub, was this extra parameters BEFORE we list the individual cameras in the config, ie after putting in the mqqt settings?  Im getting so many red herrings at the moment I getting very confused....so near yet so far?!!!

[Flubster]

extra params are in docker advanced in unraid.

 

I think like yayitazale said, you need to concentrate on one part first.

 

get the docker frigate container up, web interface working, GPU accel showing on web internface, then worry about mqtt / homeassistant reporting / alerts / integration next.

 

Flub

[Lunch]

On 4/12/2023 at 10:40 PM, Flubster said:

extra params are in docker advanced in unraid.

 

I think like yayitazale said, you need to concentrate on one part first.

 

get the docker frigate container up, web interface working, GPU accel showing on web internface, then worry about mqtt / homeassistant reporting / alerts / integration next.

 

Flub

Could you please post a pic of your config with the extra parameters part.  

Now whenI set it up and then in Frigate console write nvidia-smi all i get it: sh: 1: nvidia-smi: not found yet the nvidia plugin is installed and my GPU has its identifying code from there in the config too...this is so frustrating

[flyize]

On 4/12/2023 at 7:39 AM, Lunch said:

Bud its in Frigate itself...and if you want to watch back in home assistant al you need is the Frigate Proxy linked to your main Frigate entity wherever it is on your home network

No, I would like to be able to scroll through the last 24 hours of a camera feed. Is that easily possible?

[yayitazale]

On 4/14/2023 at 9:21 AM, Lunch said:

Could you please post a pic of your config with the extra parameters part.  

Now whenI set it up and then in Frigate console write nvidia-smi all i get it: sh: 1: nvidia-smi: not found yet the nvidia plugin is installed and my GPU has its identifying code from there in the config too...this is so frustrating

The extra parameter field will show up if you enable Advanced view in the upper corner of the configuration page of the containers template. This is something common in unraid there is plenty info on internet about that. You have to add the "--runtime=nvidia" there and not in the config file.

Edited April 16, 2023 by yayitazale

[yayitazale]

On 4/14/2023 at 4:36 PM, flyize said:

No, I would like to be able to scroll through the last 24 hours of a camera feed. Is that easily possible?

Please read the docs https://docs.frigate.video/configuration/record

[Kaldek]

Hi folks.  About to upgrade to Frigate 0.12.0 in case that answers this but I have an issue with my Frigate install on unRAID.

That is, my Coral TPU busID changes after Frigate loads the Tensor code into it.  At boot time this device is at /dev/bus/usb/004/003 but it then flips to /dev/bus/usb/004/002 after Frigate loads the Tensor software into it.

 

I'll be damned if I'm going to run this container as privileged and expose all of /dev/bus/usb to it, just to deal with that issue.  Right now I'm in the middle of a rather long chat with GPT4 about how we can get around this issue (currently, dumping the JSON config of the running container, finding the BusID, changing it and then restarting the container with the Google-ified BusID using the same JSON code method).

Total pain in the ass of course just to flip a couple of bits and then restart a container.

 

Don't suppose 0.12.0 resolves that issue with the Coral TPU by any chance?

Edited April 17, 2023 by Kaldek

[yayitazale]

1 hour ago, Kaldek said:

Hi folks.  About to upgrade to Frigate 0.12.0 in case that answers this but I have an issue with my Frigate install on unRAID.

That is, my Coral TPU busID changes after Frigate loads the Tensor code into it.  At boot time this device is at /dev/bus/usb/004/003 but it then flips to /dev/bus/usb/004/002 after Frigate loads the Tensor software into it.

 

I'll be damned if I'm going to run this container as privileged and expose all of /dev/bus/usb to it, just to deal with that issue.  Right now I'm in the middle of a rather long chat with GPT4 about how we can get around this issue (currently, dumping the JSON config of the running container, finding the BusID, changing it and then restarting the container with the Google-ified BusID using the same JSON code method).

Total pain in the ass of course just to flip a couple of bits and then restart a container.

 

Don't suppose 0.12.0 resolves that issue with the Coral TPU by any chance?

Hi, I don't think so. You can open a issue on github if you want to push on this to the main development team. Currently th only way to use the USB coral is to launch the container in privileged mode.

[Kaldek]

13 minutes ago, yayitazale said:

Currently th only way to use the USB coral is to launch the container in privileged mode.

Not technically true since I have it working as long as I boot the container twice with the two different USB bus entries.

[Kaldek]

For what it's worth, here's the code from my discussion with ChatGPT.  This is - as yet - untested.  But knock yourselves out if you want to see what was generated.

Note that this script is intended to be run "After start of array".

 

#!/bin/bash

CONTAINER_NAME="frigate"

# Wait for 2 minutes for container to start
sleep 120

# Get the current container configuration
CONFIG_JSON=$(docker inspect --type container --format '{{json .}}' ${CONTAINER_NAME})

# Extract the current command and entrypoint
CMD=$(echo ${CONFIG_JSON} | jq -r '.Config.Cmd | join(" ")')
ENTRYPOINT=$(echo ${CONFIG_JSON} | jq -r '.Config.Entrypoint | join(" ")')

# Extract all options from the HostConfig property
HOST_CONFIG_OPTIONS=$(echo ${CONFIG_JSON} | jq -r '.HostConfig | to_entries | map(select(.key != "Devices")) | map("--" + .key + "=\"" + (.value | tostring) + "\"") | join(" ")')

# Replace the USB device option with the new bus ID
HOST_CONFIG_OPTIONS=$(echo ${HOST_CONFIG_OPTIONS} | sed 's@--device="/dev/bus/usb/004/002@--device="/dev/bus/usb/004/003@g')

# Extract the image name
IMAGE=$(echo ${CONFIG_JSON} | jq -r '.Config.Image')

# Build the new container run command
NEW_CMD="docker run --name ${CONTAINER_NAME} ${HOST_CONFIG_OPTIONS} ${ENTRYPOINT} ${CMD}"

# Stop the existing container
docker stop ${CONTAINER_NAME}

# Run the container with the new configuration
eval ${NEW_CMD}

 

Edited April 17, 2023 by Kaldek

[Lunch]

16 hours ago, yayitazale said:

The extra parameter field will show up if you enable Advanced view in the upper corner of the configuration page of the containers template. This is something common in unraid there is plenty info on internet about that. You have to add the "--runtime=nvidia" there and not in the config file.

OMFG I am a goose...I was looking for the Extra PArameters section at the bottome where you can add a device, path etc...what a fool I am so sorry, I havent progressed yet as have the kids on school holidays but now I know a 1% more than I did before, sorry for you having to talk to me as if I was a pre-teen!  Thanks again Yayitazale(and Flubster!)

[Lunch]

On 4/11/2023 at 2:06 AM, yayitazale said:

Please read the full documentation. You are using old format hardware accel presets.

 

Are you planning to use nvidia GPU just to HW accel (case A) or/and you want to detect (case B)?

 

If this is case A:

• Did you install the nvidia driver plugin?
• Can you see the GPU doing nvidia-smi in the unraid console?
• Did you passtrought correctly the GPU using the GPUID (taken from the nvidia driver plugin)?
• Did you see the GPU inside the docker container of frigate doing nvidia-smi?
• Just in case, are you using the GPU in any VM on unraid?

If this is case B , all the questions of the case A plus:

• Did you select the nvidia branch at the frigate app install?
• Did you created the models?

Please also recheck your config file to meet the the docs: https://docs.frigate.video/

OK, further to before, Ive finally managed to get my Quaddro P600 as HW accel.  

Ive got a Coral on order coming in next week or two so I wont be doing case B.  I just wanted to again say thanks so much for your help on here for myself and many others.  I know you werent Frigate creator, but massive kudos to you for all that you've done in the support for us

[danimal86]

i wish there was a config creator tool....something where you plug in all the necessary values (camera address/storage/settings) and it would spit out the cofig file.  

 

I've got the m.2 coral installed but i have no idea where to start as far as setting up that config file.  I tried reading up on the frigate page, but since you created the docker for Community Apps, i'm not sure where to start with the docs.  

 

is there any guides with pictures and colors for the notepad++ illiterate like me?  

[Bruceflix]

7 hours ago, danimal86 said:

i'm not sure where to start with the docs.  

https://docs.frigate.video/configuration/

 

I recommend using vscode over notepad++

and following from the documentation:
 

VSCode Configuration Schema

VSCode (and VSCode addon) supports the JSON schemas which will automatically validate the config. This can be added by adding # yaml-language-server: $schema=http://frigate_host:5000/api/config/schema.json to the top of the config file. frigate_host being the IP address of Frigate or ccab4aaf-frigate if running in the addon.