#Autobalance

1 messages · Page 1 of 1 (latest)

simple lily
#

Hello Everyone,

I recently configured a new system with a FlexGroup with 8 constituent volumes. I’ve realized that autobalance is not enabled so I’d like to take advantage of that feature so I don’t need to manually monitor and balance the volumes within the FlexGroup. I’ve found two similar commands and I’m unsure which one is the right choice:

 Autobalance aggregagate config modify -is-enabled true
 Storage aggregate modify -is-autobalance-eligible true

What I’m also a little confused about is that these commands are at the aggregate level – seeing the volumes within the FlexGroup are what I want to have balanced, why isn’t the command at the volume level? I guess the command is just saying “all volumes within this aggregate, if they are within the same FlexGroup we will balance them accordingly”? Additionally, is there a recommended best practice for balanced/unbalanced threshold? It looks like 40%/70% is the default.

rugged obsidian
#

hm, I know that feature exists but I have honestly never seen the need for it. All it does is try to keep aggregate utilizazion equally distributed across aggregates. Especially with FlexGroups (which have all constituents of the same, pre-determined size) I don't think there will be much balancing at all to be done

#

it is also completely undocumented outside of manual pages so there are no real recommendations on how to configure it that I can find

simple lily
#

I only started looking into it as i was investigating something else and realized my constituent volumes are pretty heavily unbalanced, favoring the 1st of the 8. Not sure if its really a problem tbh but was curious

late vortex
#

In 9.7 elastic sizing helped and on top of that, 9.8 introduced the proactive resizing.

rugged obsidian
#

yeah, autobalance will do nothing if the constituents are unbalanced. wait for 9.16 which will do some awesome things with FlexGroup rebalancing 🙂

lethal palm
simple lily
rugged obsidian
#

yes. but the options you mentioned balance volumes across aggregates, which is a different thing

#

that's why it's called FlexGroup "rebalancing" and not "autobalancing" (two different things)

simple lily
#

Ah makes sense!

wispy pumice
#

I’m still hearing it’s not a good plan to use ONTAP to rebalance flexgroup members.
Second, member imbalances happen for a number of reasons. Remember a single file is only ever on one member and if that file is large it will make the member look larger. Also when files are appended to they remain on the same member. Over time this gets better. As stated elastic resizing will dynamically shrink and grow individual members while maintaining the flexgroup size and if auto resize is enabled to can grow the fg to prevent out of space errors

rugged obsidian
#

Remember a single file is only ever on one member

I think this refers to FlexGroups? If so, you might want to look into the 9.16.1 EAP, especially the "Advanced Capacity Balancing" feature and multipart-Inodes

wispy pumice
#

Yes yes! Sorry. I’ll correct

#

And ain’t no way I’m putting any of my customers 9.16 or c even 9.15. Most are air gapped and don’t have auto support connected

rugged obsidian
#

fair enough 🙂 just saying that the old "one file is on one and only one constituent" rule will not be true for much longer

wispy pumice
#

That scares me. I used coral a long time ago. I remember…

acoustic creek
#

Jumping in very late here. Lots of changes to FlexGroup volumes with some (perhaps) poor names. So FlexGroup volumes will place data across all member volumes. So think of it as "auto balance" is always on. You should not have to do anything. Now this does NOT mean that all member volumes will fill at the same rate. Thats not the intent. ONTAP places data for optimal performance and by doing that some volumes will be more full than others. Thats totally OK and actually a good thing!! No what you don't want is any one volume filling up. There are enhancements in ONTAP 9.14.1 and again something big coming in 9.16.1 that will further help this. The "auto rebalance" in System Manager is manually moving files from memeber volume to member volume. You have to manually kick this off. The "auto" part is ONTAP will automatically select the files to move. This should only be done if you have one volume very very full. Ideally you let ONTAP sort it out and not do the rebalance.

simple lily
# acoustic creek Jumping in very late here. Lots of changes to FlexGroup volumes with some (perha...

Great explanation, thank you for that! So I first realized that my member volumes weren’t equally balanced when I was investigating an error that was referring to individual member volumes within my FlexGroup. Initially I thought that while I fix the other problem, I may as well look into balancing the volumes to make them as even as possible. Now as you mentioned above my unequal volumes are not a bad thing, however I do think my errors may be more related than I first thought. The errors that I am receiving are only on member volumes that are above the “ideal distribution” threshold that System Manager has labeled for those member volumes, basically the most full volumes that have surpassed the “balanced” threshold that it seems was created when I created the FlexGroup.

I work at a secure facility so I’m in the parking lot relaying this. I’m going to report back with the exact error I’m getting on those member volumes.

acoustic creek
#

We are removing that "ideal distribution" line. 🙂 It was a silly metric. ONTAP will group files, ideally by directory, in the same member volume. This makes things like "ls" and searches fast. It is tottaly normal to have one or more volumes much "fuller" than others. The only danger is if one volume fills up. Ideally you have auto grow enabled or the option to "Flex" capacity where ONTAP will grow one vol by shrinking another. That makes sure you don't fill the Aggr. Doing a rebalance will move files around but it can hurt performance and if you have snapshots, it will consume MORE space till the snaps expire so....not ideal. We are real close to not needing to rebalance at all (or at least not recommending it anymore). Let us know the errors, will help you source them out.

lethal palm
acoustic creek
#

My session at Insight touched on it and John Lantz did a full session on FlexGroup Volumes which goes into more detail. Hopefully the replays posted soon. Will have some more blogs and details out closer to the ONTAP 9.16.1 release

rugged obsidian
# lethal palm Where can I learn more about this? 🙂

there's a (moderated) group in the NetApp community about the Early Access Program (EAP) for 9.16.1. You can apply to join and get the EAP test plans and access to an X-Release. I think you can just search for 9.16.1 EAP or something like that

simple lily
#

The error I'm getting is "fp.est.scan.start.failed: Volume footprint estimator scan fails to start on "volumename__0001" - CDE disabled."

#

Best I could find was a KB linking this error to a problem with DP volume, however my volume isnt set up for data protection

#

So I started looking at the volume and realized it was unbalanced, which is how I landed here.

kind plume
late vortex
late vortex
rugged obsidian