Renegade605

This week I replaced the parity drive in my server with a larger one. I had a preclear running as a stress test for a while, which I interrupted mid-cycle when adding the device to the array. The following line appears in the logs when starting the array, which seems like it shouldn't be there. Feb 21 19:51:13 Global-Dynamics preclear.disk: Resuming preclear of disk 'sdr' Also, it is listed in the historical devices section as "standby" and the option to delete it is disabled. I don't know if these two things are connected. Should I be concerned? How can I remove it from historical devices?

Huh. Today I learned SAS drives control the lights themselves. Looks like it's configurable too so that's perfect cause it would have really bothered me if this one drive was different than all the rest. Thanks!

Today, I've noticed that the sorting algorithm for this plugin is lexicographic. Switching to a more intelligent sort method in a future release would be desirable. EDIT: Also, when creating a new dataset, the options for "Mount" are Yes/No. But, "Yes" is actually more like inherit, since it will not be mounted if the parent dataset is not mounted. The options should probably be Inherit/Yes/No, with "Yes" enabling (and requiring) the text box for a mountpoint and the other options leaving that text box disabled.

I replaced the parity disk in one of my systems this week with a larger one. The parity rebuild finished yesterday, but today I noticed that the hard drive activity light for that disk is on all the time (hot-swap bays with individual lights for each disk). However, there is no activity on the drive in the GUI. The light does flash correctly when there is real activity. The following line appears in the logs when starting the array, which seems like it shouldn't be there. I was running a pre-clear on the disk as a stress test even though it wasn't necessary for a parity disk. I did interrupt this operation partway through one of the cycles. Feb 21 19:51:13 Global-Dynamics preclear.disk: Resuming preclear of disk 'sdr' Also, the unassigned devices plugin has it listed as "in standby" and the option to remove it from the historical devices list is disabled. I'm concerned that my parity is reporting as okay when there's actually some silent error going on in the background.

All this plugin really does is give you buttons that run the same commands you would type in the shell. The shell command would also fail in many circumstances. Perhaps you should review the OpenZFS documentation to learn how it works.

You can either - rename the current folder - remount the dataset - move the files - delete the folder Or - mount the dataset somewhere else (eg /mnt/cache/appdata-new) - move the files - delete the folder - change the dataset mount point back to default ZFS will let you mount a dataset anywhere you want, which has many practical uses. If you wanted you could create a dataset on the cache and then mount it to /mnt/disk1, although that would not be a good idea. At minimum, every top level folder on a zpool should be a dataset. Unraid will do this automatically for user shares that don't exist yet, but doesn't convert folders that already exist. It's up to you if you want to create more. For example, you can see my "cache/domains" dataset has a "cache/domains/pfSense" child dataset.

Yes.

It isn't gone, just hidden. The old appdata folder exists but the new appdata dataset is mounted over top. Change the mountpoint of the appdata dataset to "none" and the folder will be there.

Correct. There's generally no reason to have a child dataset if it's the only child. The benefit to child datasets is different properties, attributes, snapshots, etc. for each. "Task" may have been the wrong word choice. For example, another thing I've done is create a separate child dataset in appdata for each container. They all get their own snapshots, so if an update borks one container, I can rollback the appdata for that one container with a single button. Postgres performs better when the zfs record size matches the database page size, so I've done that for only those containers. Once you start playing with ZFS tuning, you may find yourself going down a rabbit hole, but the sky is the limit. I just let it handle itself. My 24 containers use 8.7G. And again, there's a quota in place just in case.

Now that you mention it, I did typo that. But it worked anyway, because "docker/" matches all child datasets of docker (ie. "cache/system/docker/01d13...." matches "docker/" and the + 'one or more times' is irrelevant). Looking closer, I suspect it's because your docker folder is just a folder, not a dataset. The exclusion pattern only matches datasets, not file/folder structure. (Type "zfs list" in the terminal to see what datasets there are on your system.) You can either a) make a child dataset under 'system' named 'docker' (you'll have to delete your existing directory first, and reinstall your containers after), or b) exclude "system/.+" If you do the latter, and later make a child dataset for system, it will also not appear. Unless you make your exclusion pattern something like "system/[a-f0-9]{64}". EDIT: Nevermind, looks like Lua doesn't support {} notation. I suggest the former, as creating new datasets for differing tasks is a big part of zfs design philosophy and you can do more later on. (For example: I have a reservation and quota of 20G on the docker dataset, so it's guaranteed to always have 20G available no matter how full the rest of my cache pool is, and to never take up more than 20G if something weird should happen.)

That's how Docker handles using a directory for image storage. It's normal. Try "/docker/+" as the exclusion pattern, or similar for your dataset structure.

Bump

global-dynamics-diagnostics-20240108-1215.zip

After a recent hardware upgrade (new mobo/CPUs) I found that a number of my docker containers were performing poorly. I've also noticed that my cache pool is seeing constant reads @ 10-40 MB/s per drive. I don't know for sure this is the cause, but it seems a likely place to start. Cache configuration is 6x250GB SSD in raidz2. File Activity and Active Streams plugins show no activity at all. lsof | grep /mnt/cache Shows a couple log files open by rsyslogd and znapzend, and a bajillion lines of "lsof: no pwd entry for UID 1883/999" 999 lines extend below this screenshot for more than 1000 lines (estimated). iostat doesn't seem to exist on Unraid (command not found). Can someone assist with tracking down this issue?

What makes you think znapzend is abandoned? For aliasing, definitely a niche case; I don't think many people run zfs on their array disks. I'm not even sure that I think it's the greatest idea, but it allowed for some interesting experiments with zfs that I wouldn't have had the available hardware to try otherwise. Overall, I think it's still worth it, detecting silent corruption, transaction groups, and caching metadata off the spinning rust without needing plugins for any of that are big pluses. As an example, right now I have a lot of scripts and snapshots set up for the pool disk6. It would be nice if I could just refer to that pool as backup and have a single place where I could tell the OS "backup" = "disk6", for readability and ease of change later on. Again, definitely niche, but I figured that would be built in to zfs from the start.

One thing not related to new features is that I'd suggest is that sending notifications through Unraid's notification handler is largely unnecessary. Especially since some are sent as warning importance, I get an email for "cannot unmount x; dataset busy" (for example), even though I already knew that from the popup as soon as I attempted it. Notifications should be saved for things that the system does while unattended that I'd otherwise miss.

One big thing would be znapzend integration with a nice GUI. I have znapzend but it's CLI only. Which is, fine, but it can be a complex enough task that GUI would be nice. Configuring your Samba to use shadow copy for snapshots as well, especially if you know of a way to make that work with nested datasets and recursive snapshotting, cause I have yet to figure that one out. I don't know if that can work with datasets on multiple pools/disks after they've been through the FUSE system, but also would be great. Configuring ZFS properties in the Share page and then auto-creating datasets with those props on all applicable disks/pools and keeping them updated. Maybe with a tunables for array vs pool to optimize properties for (eg.) an all SSD cache pool and array disks separately. Some way to alias ZFS pools so you could say "this disk is called 'backup'" so it can be in the array but if you change the disk number later you (or automated tools) won't be running commands on the wrong disk. I assumed that would exist natively in ZFS but hell if I can find it. Auto (and/or offer to) create datasets and zvols in the Docker container and VM setup pages, with property tuning. (I keep a separate child dataset for every container in appdata to allow rolling back or cloning one appdata folder from a snapshot without affecting the rest.) Perhaps a one-click way to exclude certain subfolders from snapshots on their parent datasets. I've done this manually and it isn't the worst but it's kind of a drag especially if you forget to do it before there's data in there. I'm sure there's more I'm forgetting, but that comes to mind at the moment.

I know. That's why I haven't made any formal feature request at this time nor gone full send on switching away.

I've got a fair number of ideas for more advanced zfs functionality I wish unraid (or a plugin like this) would do. They may be far too advanced for average Unraid user; in fact I'm leaning towards abandoning Unraid at this point because I feel like it's holding me back. But, if you'd want to chat about them anyway, just let me know.

I notice the directory listing feature doesn't list top level directories for entire pools, if enabled for them. This is where I'd personally find that most useful if it could. In fact I would suggest it do so by default, as there's basically no reason to have a top level folder in a zpool.

Oops just seeing this. My new request is related: https://forums.unraid.net/topic/149858-maintenance-mode-boot-start/

Currently, when array / container / VM autostarts are enabled, booting the server will of course start all these services. If we are rebooting to do maintenance (ie. add or remove hardware, etc.) we might not want the array to start so we can adjust settings that need the array stopped. At present, this means remembering to disable array autostart before shutting down. Forgetting means waiting for the array to start just to stop it again (can be quite long, depending on filesystem and how many disks you have). Additionally, one could forget to turn array autostart back on, and then an unexpected reboot never brings the server back up. Other maintenance tasks involving disk configuration might require starting the array, but we don't want containers or VMs to start. Similarly, we must disable the Docker and VM engines before starting the array, or wait for them all to start just to stop them again. We may also need to edit container or VM settings and want to do that before they start (perhaps we've changed disk configuration and need to change path mappings). Array must be started and Docker engine must be enabled to edit containers, but they will autostart when we do that and we can't stop it. Proposed Solution A checkbox next to the reboot/shutdown buttons (like the "Reboot in Safe Mode" one) that will disable array autostart for the next boot only. When array autostarts, continue to autostart containers and VMs, but when the array is manually started offer a dialog asking if containers, VMs, both, or neither should start with it. Once we've completed our task(s), a button like "Start All" for containers/VMs that respects if they are set to autostart could be clicked. (Perhaps we could just always have such a button?) This could save users a lot of time over the course of an Unraid machine's lifetime.

Cache is going to vary wildly depending on what's happening at any given moment, and my Media is huge because I have all possible thumbnails generated. And my actual libraries are into the dozens of TB at this point.

I've since switched to using ZFS snapshots instead of this plugin, but here's my exclusion list: Cache excluded on the basis that it's meant to be temporary anyway, all others because they are easily and automatically re-downloaded/-generated if lost.

Yes that and it deduplicates notifications so you can run it more often (I schedule it hourly) and be alerted sooner without getting spammed with warnings.