Will Limetech ever fix the persistent cache writes?

[BillR]

I've been using UNRAID for a few years now and I have burnt through so many SSDs, that I have stopped using them for cache and now just use a pair of spinning rust drives.  I want redundancy in my cache, so I pool, and that means I have to use btrfs, which I understand is more susceptible to the problem.  I know a few versions ago (6.9.0, I believe), there was a supposed fix to stop excessive writes to cache.  At the time, I blew away my cache pool and re-created it, but this has been an ongoing problem for me regardless.

 

After burning out my last pair of NVMe drives after just a few months, I ordered and installed a new pair.  After about a week they showed hundreds of GB of writes - and this was without caching any data shares, just the default used for Docker containers.

 

I tried the suggested fix to send log files to RAMDisk here:

But this didn't seem to stop the incessant writes to cache that seem to occur if I have pretty much any Docker container running - and not just the known troublesome ones like Plex.

 

We get cool new features with new Unraid builds, but show-stopping issues like this seem to persist for years.  I would love to see some more effort into fixing problems like this, that actually cost me significant money in having to replace SSDs.  As said, I've now given up on using SSDs for cache and have a pair of hard drives on cache duty, sucking more power and not performing as well.  And yes, they are still constantly being written to by just having docker running, but fortunately, don't wear out and die like SSDs do.

 

 

Edited October 18, 2022 by BillR
Added version number when cache partition changes were made

[fwiler]

Following thread, as I'm hoping this can be fixed in OS and not rely on specific Docker fixes.  I've actually stopped using all Dockers for now.

[BillR]

1 hour ago, fwiler said:

Following thread, as I'm hoping this can be fixed in OS and not rely on specific Docker fixes.  I've actually stopped using all Dockers for now.

Wow, that's a big sacrifice to make.  If I couldn't use dockers at all, I would have to move away from Unraid, which would be a real shame, as I really do love the functionality and ease of use of Unraid.  May I suggest doing what I've done and cache on a couple of hard drives instead?  I do notice that it takes much longer for all my dockers to start up on spinning rust, but once they are up, the performance isn't too bad.  The disks I'm using are a couple of 2 TB WD Reds I happened to have lying about, but I'm tempted to buy a couple of newer drives with higher performance, as the 2TB Reds are quite slow.

[kizer]

I've tried this with my Drive formatted to XFS. Seems to help alot ..

 

https://forums.unraid.net/bug-reports/stable-releases/683-unnecessary-overwriting-of-json-files-in-dockerimg-every-5-seconds-r1079/?tab=comments#comment-15472

 

[BillR]

On 10/18/2022 at 5:23 PM, kizer said:

I've tried this with my Drive formatted to XFS. Seems to help alot ..

 

https://forums.unraid.net/bug-reports/stable-releases/683-unnecessary-overwriting-of-json-files-in-dockerimg-every-5-seconds-r1079/?tab=comments#comment-15472

 

Yep, that's the same thing I tried (see above).  Unfortunately, I need btrfs because I want redundant disks and it didn't seem to help.

[kizer]

9 hours ago, BillR said:

Yep, that's the same thing I tried (see above).  Unfortunately, I need btrfs because I want redundant disks and it didn't seem to help.

 

Exactly. I was running XFS switched to BTFFS and tried this too. Seemed to help some, but then I messed up my second drive in the pool tinkering around and switched back to single XFS. Switching back to single XFS seems to reduce the writes a lot. 

 

From my experience its been the docker container 25GB file and its internal logs that seems to be doing most of all the writes. I think anything limetech does will be more or less be a hack a fix than a true solution, but only time will tell.