BrandonG777

Well, looks like there is some sort of overall Tailscale issue with my unRAID server. Even the built in containers are experiencing the lag. I suppose playing with my Tailscale options/routes is first order of business but will need to wait until I get home tomorrow before I do all that.

all client machines have Tailscale. I understand how Tailscale works and use it everyday without issue. TSDProxy or how I have it configured is adding so much latency, the containers are unusable but they do connect.. if I give them 10 minutes to load up. If I connect a container via unRAID Tailscale option, lag is minimal. Has to be some sort of routing/DNS issue I'm thinking but I'm struggling to identify the cause.

My connections are crazy slow. Not sure what is happening but I use tailscale all the time without issue except in this case. Haven't been able to determine a cause but I can connect via unRAID tailscale option and I don't see the insane amount of lag I'm seeing with TSDProxy.

same. also found all of the tools still worked... but if they're aren't being updated, that will be a problem sooner or later. seems like a simple script could keep this stuff up to date and maybe even streamline the install process. I'd do it if I didn't already have too many irons in the fire and having to prioritize.

@shinji257 vim please? 🙏 EDIT: nm, located via https://pkgs.org/ Seems to work fine.

Yeah, sorry I know better. smh

Pardon my ignorance but prior to removing the plugin I had more to up to date packages than what is provided in the Slackware respositories (https://slackware.pkgs.org/15.0/slackware-x86_64/)... After a bit of searching I found your github repository but I haven't had much luck with those packages as installpkg <pkg> results in "gzip: stdin: not in gzip format" But the bigger issue I see here that without something like Nerdtools to maintain the current versions, we are putting ourselves at risk with potentially unmaintained versions of these packages. Again there is much about that process I don't fully understand so maybe I'm missing something but without us manually checking for new versions on a regular basis we are opening the door to unpatched vulnerabilities. On this note, couldn't we implement some sort of script we could throw a list of packages at and do the same magic Nerdtools was doing, just minus a pretty GUI. With all that being said, thank you for work on this. I'm sure what you're doing is helping push this forward, despite my lack of understanding.

Best solution I've found is the limit the memory of the container, --memory=4G, under extra parameters. Not perfect but best I can come up with. Obviously, your memory usage will vary with number of cameras/resolution/etc.

Other than the occasional push notification that fails to send the attached image, it works much better than RTMP streams.

Did you setup a substream (720p 5fps or lower resolution) I see around 20% constant usage on my i5-12500 but I have 8 cameras being monitored and also using a Coral USB. Also something to think about, your wifi doorbell will have a constant video stream running across your wifi which could adversely affect wifi performance. I originally did this with my Amcrest/Wyze doorbells. Ended up just adding another analog camera through my Amcrest DVR for $25. This also prevents a blind spot for my packages being put directly under the doorbell cams. When I started this project I thought I had to convert all my cams to POE digital cams but analog video at 720p still works great and pricing on a good analog DVR is less than $100 these days. Highly recommend Amcrest but full disclosure, I’ve reviewed their devices for years which they have provided at no cost. Sent from my iPhone using Tapatalk

You will need to change your config though I’ve not worked through getting an AMD cpu/gpu setup with Frigate. However, there are example config you can copy from the frigate site. I believe what you are looking for is VAAPI or something like that. Sent from my iPhone using Tapatalk

So after running both the Coral with 12th Gen Intel QSV and Nvidia Quadro RTX 4000 for Frigate processing for some time, the Coral/Intel setup significantly uses less power and generates much less heat than the Nvidia method. Overall, this resulted in 100watt less power usage and Quadro RTX 4000 is no longer running a constant 60C. Unrelated to unRAID but kind of wondering how the HA Yellow/RP4 would handle Frigate with Coral attached, I know others are doing it but just unsure about how many cameras and what not they are able to handle.

Good looking out! I bet this was my issue all along.

Just ran into this myself... trying to find a solution to move the dataset rather than just moving the files themselves. I'd like to keep my snapshots if possible.

These hard lock ups are killing me. @yayitazale Thoughts on using the yolov8n model? Looks like others are having better luck with it, reading through the gitlub issue link you posted. Also, I attempted limit the memory this container has to 4GB but that didn't help. I have also verified that it is limiting the memory. I thought this would keep it from using too much and killing off critical system processes but no. If anyone could please explain this behavior I would greatly appreciate it. The other attempt I made was restarting the container on a schedule but this most recent lock up happened within 4 hours so don't see this bandaid working without getting in the way of the events within frigate. I have to figure this out, I love Frigate. I'm finally able to accomplish what I was after when I first started purchasing security cams, to know whats happening around my home but not get notified everytime a leaf blows through my yard or the sun decides to come out. UPDATE: Nevermind, got home and realized there must have been some sort of power outage because one of my UPSs was screaming. Will continue with this 4GB limit and see what happens.

So turns out my server locks up were not tied to the USB Coral device I was trying to use but rather out of memory issues that seem to be related to the ffmpeg process. Not sure where I've gone wrong, it works for days/weeks and then at the most inconvenient times it hard locks. mqtt: host: 10.0.0.3 port: 1883 topic_prefix: frigate client_id: frigate user: frigate password: **CENSORED** stats_interval: 60 ffmpeg: hwaccel_args: preset-nvidia-h264 detectors: cuda: type: tensorrt model: path: /config/model_cache/tensorrt/yolov4-tiny-416.trt labelmap_path: /labelmap.txt input_tensor: nchw width: 416 height: 416 record: enabled: true retain: days: 7 mode: motion events: retain: default: 5 mode: active_objects pre_capture: 3 post_capture: 10 snapshots: enabled: true retain: default: 30 timestamp: true bounding_box: true birdseye: enabled: false restream: true width: 1280 height: 720 quality: 8 mode: motion go2rtc: streams: driveway: - rtsp://frigate:**CENSORED**@10.0.0.8:554/cam/realmonitor?channel=1&subtype=0 driveway_sub: - rtsp://frigate:**CENSORED**@10.0.0.8:554/cam/realmonitor?channel=1&subtype=2 front: - rtsp://frigate:**CENSORED**@10.0.0.6:554/cam/realmonitor?channel=1&subtype=0 front_sub: - rtsp://frigate:**CENSORED**@10.0.0.6:554/cam/realmonitor?channel=1&subtype=2 front_door: - rtsp://frigate:**CENSORED**@10.0.0.5:554/cam/realmonitor?channel=2&subtype=0 garage: - rtsp://frigate:**CENSORED**@10.0.0.7:554/cam/realmonitor?channel=1&subtype=0 garage_sub: - rtsp://frigate:**CENSORED**@10.0.0.7:554/cam/realmonitor?channel=1&subtype=2 livingroom: - rtsp://homeassistant:**CENSORED**@10.0.0.9:554/cam/realmonitor?channel=1&subtype=0 north_deck: - rtsp://frigate:**CENSORED**@10.0.0.5:554/cam/realmonitor?channel=4&subtype=0 south_deck: - rtsp://frigate:**CENSORED**@10.0.0.5:554/cam/realmonitor?channel=3&subtype=0 cameras: driveway: ffmpeg: input_args: preset-rtsp-restream inputs: - path: rtsp://127.0.0.1:8554/driveway_sub roles: - detect - path: rtsp://127.0.0.1:8554/driveway roles: - audio - record detect: width: 1280 height: 720 fps: 5 objects: track: - person - bicycle - car - cat - dog filters: car: mask: - 1280,0,1280,440,992,296,569,331,708,720,0,720,0,0 birdseye: mode: continuous zones: driveway_top: coordinates: 687,720,561,321,955,294,1280,481,1259,528,1160,720 motion: mask: - 1243,659,1243,701,831,699,831,663 front: ffmpeg: input_args: preset-rtsp-restream inputs: - path: rtsp://127.0.0.1:8554/front_sub roles: - detect - path: rtsp://127.0.0.1:8554/front roles: - audio - record detect: width: 1280 height: 720 fps: 5 motion: mask: - 1243,659,1243,701,831,699,831,663 objects: track: - person - bicycle - cat - dog filters: person: mask: - 728,720,931,720,887,615,746,469,632,499 zones: driveway_top: coordinates: 238,227,580,161,463,148,218,194 front_lawn: coordinates: 364,255,256,290,745,511,925,720,1280,720,1280,345,1112,276,844,194,637,145,564,165,355,210,312,227 front_patio: coordinates: 0,720,894,720,747,507,233,305,0,412 front_porch: coordinates: 0,407,344,253,231,254,195,105,0,178 birdseye: mode: continuous front_door: ffmpeg: input_args: preset-rtsp-restream inputs: - path: rtsp://127.0.0.1:8554/front_door roles: - detect - record detect: width: 960 height: 480 fps: 5 objects: track: - person - cat - dog zones: front_lawn: coordinates: 588,144,734,221,795,0,351,0,365,44,506,50 front_porch: coordinates: 223,480,960,480,960,160,743,202,588,138,541,81,291,70 motion: mask: - 377,16,383,53,16,45,17,17 garage: ffmpeg: input_args: preset-rtsp-restream inputs: - path: rtsp://127.0.0.1:8554/garage_sub roles: - detect - path: rtsp://127.0.0.1:8554/garage roles: - audio - record detect: width: 1280 height: 720 fps: 5 objects: track: - person - bicycle - car - cat - dog zones: garage_floor: coordinates: 21,263,182,619,428,673,463,720,1280,720,1280,216,967,21,437,26 objects: - person - bicycle - car - cat - dog birdseye: mode: continuous motion: mask: - 430,13,432,44,12,45,12,8 north_deck: ffmpeg: input_args: preset-rtsp-restream inputs: - path: rtsp://127.0.0.1:8554/north_deck roles: - detect - record detect: width: 1280 height: 720 fps: 5 objects: track: - person - cat - dog motion: mask: - 1188,22,1188,55,825,53,828,27 south_deck: ffmpeg: input_args: preset-rtsp-restream inputs: - path: rtsp://127.0.0.1:8554/south_deck roles: - detect - record detect: width: 1280 height: 720 fps: 5 objects: track: - person - cat - dog motion: mask: - 1188,22,1188,55,825,53,828,27 zones: backyard: coordinates: 0,442,715,171,821,133,995,93,604,0,297,0,0,46 deck: coordinates: 0,720,1280,720,1280,0,1202,0,1154,170,908,114,0,433 livingroom: ffmpeg: input_args: preset-rtsp-restream inputs: - path: rtsp://127.0.0.1:8554/livingroom roles: - detect - record detect: width: 1280 height: 720 fps: 5 objects: track: - person - cat - dog zones: kitchen: coordinates: 178,272,780,280,773,117,195,88 livingroom_area: coordinates: 0,720,1280,720,1280,60,890,77,794,115,802,282,0,340 motion: mask: - 1227,30,1227,60,906,60,906,31 birdseye: enabled: false ui: live_mode: mse z-diagnostics-20240322-0532.zip

Thank you but I wasn’t questioning your diagnosis but rather what logs/files you looked at to determine that so that I might possibly be able to diagnose these issues myself and stop relying on others for things I feel like I should know how to do. Sent from my iPhone using Tapatalk

Would you mind sharing what you looked at to determine that it's Frigate causing the OOM errors? I've dug through the logs but perhaps I'm overlooking it or just looking in the wrong place?

Is that related to these extra parameters? --shm-size=256mb --mount type=tmpfs,target=/tmp/cache,tmpfs-size=100000000 I do have a dedicated temp SSD for stuff like this. I'm thinking I should just map that to /tmp/cache? However, I'm sure the SSD isnt as fast as system memory and I have 64GB (which currently reports as 46% utilized). For performance reasons, should I try to adjust this shm to work?

Getting some out of memory errors after a handful of recent changes to one of my systems. Changes I can think of are switching to a new PC, setting up Frigate container, migrated from Plex to Emby, and upgrading to 6.12.8 unRAID. Server runs fine for days/weeks and then eventually hard locks. I originally thought it was the addition of a defective USB Coral device causing this but removing that I'm still having this issue. z-diagnostics-20240321-0518.zip

I gave up on the USB Coral. Couldn't get it to do anything useful. RMA'd for being DOA. I'll order a PCIe version and give that a go... maybe. I've just read so many problems with these devices, even if you do make it work, I've read numerous reports where it stops working after a few months. Hard for me to get excited about a device this inconsistent. Nvidia products while much more expensive actually work from my experience.

I do. I'll see what I can do with those examples but even if I am successful, I feel like I'm out of options for making it work with my target Frigate system (HP z840 workstation). Maybe it's a result of multiple USB dongles like my Zigbee & Bluetooth dongles.

Well, everything has been going fine for me using GPU detector but I have not been able to get my USB Coral device to detect within the container. Early this morning my entire unRAID hard locked on me. Something I have not seen before plugging in this sketchy USB Coral device in. I've mapped /dev/bus/usb and granted the container priviledged access. Beyond this, I can't find anything that I'm missing. Unless someone can point out something I've missed I think I'm throwing in the towel on the USB Coral. You can see my full config above.

Yeah I think you're right at the cutoff. I can't remember where I read it but somewhere I think it said v535 was required.

This was my exact issue on my original post. Somewhere between there and the config I posted yesterday it started working. Just can’t get my usb coral device to recognize Sent from my iPhone using Tapatalk