Are those the MergerFS drives?
If they are… move the files to a different drive (something non-merger) and retest
Yeah, everything is on the MergerFS ‘drive’ its a hit or miss if it works. You think that MergerFS is causing some issues? I have used it for a while but its been updated so maybe there is a bug or something.
Yeah… I’ve seen mergerfs cause lots of problems.
you’re right – Hit or Miss … This might be a miss.
Copy some episodes / files you know have problems to a proper location (even the SSD if you want).
Create a small test section and try it.
Hmph ok. I have noticed that I have HDD’s that end up with filesystem issues pretty regularly. But I dont have a RAID, and it seemed this was a best option. Do you have a better suggestion for uniting drives? I am open to better / more reliable suggestion that would allow other things like Sonarr that doesn’t support multiple paths would work well with it. Which is why I moved it to MergerFS in the first place.
IMHO —
MergerFS is “poor man’s software RAID”. It’s ok for experimenting / test beds but you want production stability.
RAID is my Go-To and this is my beast. I started with 3x 4TB drives 8 years ago and it was a screamer. It’s scaled perfectly since then even with a few HDD failures along the way – never any data loss 
[chuck@glockner ~.1998]$ df -h
Filesystem Size Used Avail Use% Mounted on
udev 126G 0 126G 0% /dev
/dev/md0 233G 19G 214G 8% /
/dev/sda1 511M 5.3M 506M 2% /boot/efi
/dev/md2 1.9T 671G 1.3T 36% /mnt/vmssd
/dev/md1 931G 3.0G 928G 1% /home
/dev/md3 99T 45T 54T 46% /mnt/vol
[chuck@glockner ~.1999]$
Here are the drives which comprise it.
┌┈▶ sdj running sas 4:0:4:0 HGST_HUH721212AL W925 5000cca291ea9801 disk 10.9T linux_raid_member
├┈▶ sdk running sas 4:0:5:0 HGST_HUH721212AL W925 5000cca291e7dedd disk 10.9T linux_raid_member
├┈▶ sdl running sas 4:0:6:0 HGST_HUH721212AL W925 5000cca291e1d226 disk 10.9T linux_raid_member
├┈▶ sdm running sas 4:0:7:0 HGST_HUH721212AL W925 5000cca291dee455 disk 10.9T linux_raid_member
├┈▶ sdn running sas 4:0:8:0 HGST_HUH721212AL W925 5000cca291e7d217 disk 10.9T linux_raid_member
├┈▶ sdo running sas 4:0:9:0 HGST_HUH721212AL W925 5000cca291db0daf disk 10.9T linux_raid_member
├┈▶ sdp running sas 4:0:10:0 HGST_HUH721212AL W925 5000cca291e8089f disk 10.9T linux_raid_member
├┈▶ sdq running sas 4:0:11:0 HGST_HUH721212AL W925 5000cca2b0d294ef disk 10.9T linux_raid_member
├┈▶ sdr running sas 4:0:12:0 HGST_HUH721212AL W925 5000cca291e1dbd6 disk 10.9T linux_raid_member
├┈▶ sds running sas 4:0:13:0 HGST_HUH721212AL W925 5000cca291eaddc3 disk 10.9T linux_raid_member
└┬▶ sdt running sas 4:0:14:0 HGST_HUH721212AL W925 5000cca291e8040e disk 10.9T linux_raid_member
└┈┈md3 raid6 98.2T xfs /mnt/vol 98.2T 53.8T 44.4T 45% raid6 98.2T xfs /mnt/vol
Here’s the kicker.
- Sonarr writes directly to this storage
- When I rip a disc, It writes over the network to it.
- I get 1.5 GB/sec throughput from this array.
I also have a QNAP NAS and a Synology NAS.
If you’re thinking of RAID / a better I/O solution
- Gigabit from the PMS host to the NAS – works fine for now (117 MB/sec) which will give you 800 Mbps of video streaming (that enough) ?
- The load is balanced and data protected (RAID 5/RAID 6).
- You can expand it over time (grow as you need if you plan ahead a bit)
My PMS hosts are two NUC8-i7-HVK boxes. One for Plex testing, one for production.
My whole LAN is 10 GbE but that’s only because I need this level of performance with all the data I throw around daily.
Thats totally fair. And I have thought about moving to a RAID but it sounds like a daunting task. And also I have a 10TB disk and IIRC that means I am going to loose a bunch of “unusable space” which was something that aggravated me about RAID’s that I couldn’t mix and match drive sizes. I also figured that I dont have hundreds of users so I wasn’t too concerned with the speed gain from striping and if I did a RAID0 that would support different drive sizes I get no parity so I felt that I didnt gain that much from RAID. But maybe its time to start looking into it.
Tell me what you have for disks in this?
RAID == Redundant Array of Independent Disks (R.A.I.D)
I’ll help you craft this out but you can’t use a single 10TB drive
(Notice how many I have?)
You can do RAID 0 (concatination) , with no protection but it’s striped. Lose one and you lose it all.
I have 4 disks:
NAME MOUNTPOINT LABEL SIZE FSTYPE UUID
sdc 14.6T
└─sdc1 /Volumes/Media 4 Media 4 14.6T ext4 e97685d9-d13d-416e-ba8b-25ebf7afce23
sdd 12.8T
└─sdd1 /Volumes/Media Media 12.8T ext4 2fdf80cd-851e-42fe-a4a4-b08f5000f501
sde 14.6T
└─sde1 /Volumes/Media 3 Media 3 14.6T ext4 f062d7fd-37f1-4d73-8c0a-9d116d360c56
sdf 9.1T
└─sdf2 /Volumes/Media 2 Media 2 9.1T ext4 31c264c2-ef31-45e0-a28b-0739571f86f8
2 16TB drives
1 14TB drive
1 10TB drive
Something that just happened, I was in PMS and noticed that a friend was buffering an episode so out of curiosity I went to play it as well, it loaded instantly for me but he kept buffering (verified via text) so I thought that is kinda interesting that I could load it and he could not. It was a direct play at 16Mbps (I have 1Gbps fiber so bandwidth shouldn’t be an issue ever)
Just because you have 1 Gbps fiber… What does he have ?
What’s end-to-end throughput?
Yeah this what I am hesitant to do… I would love to concat the drives but I dont want to be in a total loss situation.
Where I have a problem with MergerFS.
It’s a layer on top of a bunch of single drives.
- Is there any protection?
- Is there any performance gain?
- what does it really give you? (just a big honking volume or are files limited to whatever fits on the target?)
Checking… He lives near me in SF so im pretty sure that its plenty but I asked to be sure.
Use iperf3 .
Open the port … run it both ways
1 Like
- No protection, which I get. But worst case I lose one drive data and can recover what I want. Probably a ton of stuff that never gets watched anyone. And 25% loss is much better than 100% which I would have in RAID0
- Not really I suppose. I mean different files from different drives I guess would be faster theoretically I guess.
- What I liked about it is well pretty much what you said. I can have similar structured folders on drives together and it makes them all concat’d which I liked.
If there is a way to get something pretty similar to this that wouldn’t waste a ton of space, even if its with a parity, I am open to it I think. Just will need to probably figure out how to get the data moved off and back on without losing anything. Which sounds like a big task haha
Haha hes a non techie…- lets see what I can get done.
FWIW he tried a different episode and it worked and went back and it started working fine so hard to say exactly there I guess.
One side question, do you see any problem with having my transcodes go to RAM (/dev/shm)?
To save wear on the SSD — Use RAM. PMS will trim as it needs
I have 64 GB in the NUC. 32 GB is the default /dev/shm amount. More than enough for 4 transcodes
Awesome- good to know. Yeah SSD wear is why I moved it there and I never saw any issues but wasn’t sure if there was something I wasn’t considering. I have 64GB in the server as well and typically dont use over 10GB so should have plenty for my capacity I need.
exactly. The OS always runs minimal. PMS itself is lean except for spikes during scanning.
Need to call it a night here.
Here is some testing info. Good to start learning overnight.
mtr will give you a basic level of data… from his IP to the IP of your server.
https://www.ateam-oracle.com/post/testing-latency-and-throughput
Appreciate all the help as always, Chuck. I owe ya a Starbucks drink or something sometime.
