Following Spaceinvaderone's excellent YouTube video, I installed a cache drive and gave write access to it to my shares appdata, domains, and system (and one other share to which I have yet to write anything).  The 120GB SSD cache drive is now completely full (466MB free), even though I only have only three dockers (and two of those are ones I have just now installed).  According to my User Share data the appdata share is 113GB in size.  When my mover runs, it takes a long time to finish (hours).  In fact, when I left it last night it was running as it was this morning.  Drives seem to be spinning for hours at a time, as well.  When running Fix Common Problems I was getting an error about /var/logs (too many) as well as about my docker img (size or corruption).  After rebooting now (the server was unresponsive to refreshes from my brower), I increased the Docker image size to 40GB and increased the min. free space on the cache drive from 2MB to 20MB.  Running Fix Common Probs now I get a "Call Traces found on your server" error, along with a suggestion to post to the forum.  I have attached the Diagnostics file.  Thank you.

tower-diagnostics-20170722-1117.zip

just an update:  I just now attempted to start the docker (after doubling its size, as above) and it won't start up.  Also, disks are spinning like mad again.  Revised diagnostics attached.

tower-diagnostics-20170722-1159.zip

Your docker image file is corrupt so it will need deleting so a new image can be created.

 

you should not have had to increase the size of the docker image file unless you have a docker misconfigured so that it is writing its data into the image file rather than to a location that is external to the image file.   Perhaps you need to provide more detail on the docker configuration?

thanks.  what kind of information can I provide?  Is 120GB too small for a cache drive?  It seems just about all of the space is being taken up by one app (Plex).  That seems surprising to me....(?)

14 hours ago, smellthebean said:

thanks.  what kind of information can I provide?  Is 120GB too small for a cache drive?  It seems just about all of the space is being taken up by one app (Plex).  That seems surprising to me....(?)

Not if you have thumbnails being generated.

 

Additionally, if you have ever copied the Plex appdata over the network to make a backup, and then restored it (say replaced the cache drive) then the size explodes because Plex makes extensive use of hardlinks to save storage space which aren't understood over a network so one file could get duplated 10's if not 100's of times in that circumstance.

Edited July 23, 20178 yr by Squid

will deleting and reinstalling the image file reset the hardlinks?  This was my first use of a cache drive and I did not do any copying of the Plex appdata, to the best of my knowledge.

Just now, smellthebean said:

will deleting and reinstalling the image file reset the hardlinks?  This was my first use of a cache drive and I did not do any copying of the Plex appdata, to the best of my knowledge.

Nope  Probably you have thumbnail generation enabled in Plex

Edited July 23, 20178 yr by Squid

I have just deleted and reinstalled the docker image.  On running Fix Common Problems I am still getting "Call Traces found on your server" and "/var/log is getting full (currently 100% used)".  Mover is running right now.

tower-diagnostics-20170723-1038.zip

Your docker image is corrupted.  Got corrupted right away.

 

Personally, if you're not using a cache-pool, then you're going to be better off with the cache drive formatted as XFS instead of BTRFS.  Far more forgiving of a filesystem.

 

Some of the underlying problem could be that the cache drive is full, formatted as BTRFS, and BTRFS does not respond well when its 100% full.

 

You can limit the logging by going to Settings - Scheduler - Disable Mover Logging.  But, you're going to have to reboot to clear out the existing log.

OK.  The mover is now running and I will not make any changes until it finishes.  I was leaning toward a cache pool but my parity, cache, and array drives have maxed out my SATA ports.  Is there any reason not to use a PCI SSD?

 

Update:  I have rebooted and, before bringing unRAID up, changed the cache drive format to xfs.  as i have not made any app changes to data for days (docker has been unavailable) I am now looking at PLEX to confirm that the Mover did in fact successfully move over the cache data. Of course after starting the array the cache drive is now unmountable:  I assume that I must initialize it if I want it to mount it as xfs(?).  However, Squid said above that unless I were to use a cache pool I should switch to xfs.  Does that mean that I should stay w BTRFS if indeed I do add a second cache drive?  And can that cache drive be a PCI SSD?  Thanks. 

 

General consensus from the forum:

 

XFS is more robust, and handles things like unclean shutdowns, etc far better than BTRFS.   But, if you do intend to run a cache pool, you have no choice but BTRFS.  

 

AFAIK a PCI SSD (m.2 or NVMe) work perfectly with unRaid.

 

And yes, after switching the format type to be XFS, you will have to format the drive.  (There's a button on the Array Operations to do that)

well, I installed a PCI SSD (256GB) and removed the old 120GB one.  I formatted the new drive as xfs.  I set the shares using cache ("Preferred") to system, appdata, and domains.  I monitored things for a few days and all was fine.  I then enabled a share named rename to use cache (Preferred).  This I will use in conjunction with, e.g. FileBot to rename music/movies/tv files.  Again no problems.  Next (yesterday) I changed my "movieshare" to use cache, once again "Preferred".  This has caused things once again to go south:  Mover has been running all day, the cache has less than 1MB free space,  two of the docker apps I try to launch, FileBot and Krusader, fail with execution errors, my var/log is huge, and Fix Common Errors tells me my docker img. is full and can't be written to.  Am I, perhaps, confused about what a cache is for?  My understanding is that it expedites writes to the unRAID server.  While data reside on the cache they are unprotected (unless there is a cache pool) but on a periodic basis the Mover moves the files from cache to one or more of the drives in the array.  My mover is set to run hourly.  I did, upon enabling cache for movieshare, move two ripped and compressed movie files (approx 15GB each) to this share. 

 

Should I disable the docker, delete the docker image, and restart?  If I can't get Krusader to launch, how can I move files off of movieshare and on to the array?  Do I need to do this?  Drives are now spinning, Mover says it is running, but I see no increase in the free space on the cache drive.

 

More importantly, what do I do going forward?  I'd like to take advantage of the performance increase provided by a cache drive but I am obviously doing something wrong.  Thanks.

tower-diagnostics-20170802-1237.zip

22 minutes ago, smellthebean said:

Next (yesterday) I changed my "movieshare" to use cache, once again "Preferred"

 

That's the wrong setting for that share:

 

Change to cache "Yes" and run the mover again.

 

 

thanks much, johnnie.black.  Mover was still running before i made the change to "Yes" (from "Preferred") and is still running now.  Can I force it to restart? 

 

Thanks

 

IIRC:

 

mover stop

 

Thanks johnnie.black.  I was able to get the mover to stop.  After making the change from "Yes" to "Preferred", as above, everything seemed to be working well:  the mover seemed to be appropriately moving files on a regular basis, and the pct of cache used seemed about right.  Everything, in short, appeared to be in order.  But then... I started watching movies.  Over the years, I have watched over 100, at least, via PLEX and the experience has been close to flawless:  no stuttering or dropped frames, even on 12GB 1080p video.  But now (confirmed after watching three movies, two of which were SD, under 1GB files):  stuttering and jerkiness, especially on pans or scenes with movement.  I really have no idea why this might be?  The only change I have made is to install the cache and make choose these settings:  put appdata, domains, and system as "Preferred" cache and some other shares as "Yes" in cache usage.  Checking disk speeds, using the CA plugin, cache reads are, as expected, much faster (1.6 GB/s) than array reads.  I'd prefer not to lose cache functionality, but I may have to go back to my previous config if I can't figure this out.  Baffled...   Thanks.

tower-diagnostics-20170811-1252.zip