Solved

ECX error & disk id & svmotion

  • 27 June 2016
  • 4 replies
  • 3695 views

Userlevel 1
Badge +14
Hello Folks,
I`m busy installing a 6 node cluster, some errors or doubt comes out along.
my environment : 3 block, 6 nodes (2 nodes/block) AOS 4.6.2

一. When I create a new container, make the Enable Erasure coding button checked, error pops up"enable EC-X will break the current block awareness !" seems mean ECX will disable block level awareness ?

二. As many of us know, if need locate the physical slot location of a disk, you can simply led on that drive. but there is possibility led on function inaccurate or disabled, in this scenario, you must locate the disk by disk id rule, e.g shelf id.slot/bay ID, but the DISK ID in prism hardware section seems ramdom/ruleless number, I can`t locate physical location for a disk.

三. When I do the test about removing a node from nutanix cluster, A doubt appears: Do I need or is it the best practice to execute storage vmotion(other then only vmotion vm`s ) before removing the node ? I think the diffecrence of two methods is who is the data mover, one is ESXI svmotion, the other is CVM RF ? perhaps I`m wrong totally,even if I choose svmotion, RF recovery still occurs ? Thanks you guys valuable time and help in advance !

四. I enable RF3 first time, and after somes days, user would like to revert to RF2, do nutanix support this ? I know I can online change RF2 to RF3;
another case for RF: when I image node using Foundation tool, I choose default RF2, and Can I enable RF3 support by NCLI command ? (I notice there is related command in NCLI)
similarly, Can I enable EC-X and disable it again ? I`d like to know common features that can be changed twice or many times, other than one change is permanent.

Thanks community friends !
icon

Best answer by Jon 27 June 2016, 18:20

RE ECX
Yes, ECX and block awareness do not mix. There is a long technical explanation of why, but the short of it is that the majority of customers who would use ECX wouldn't have enough physical blocks to actually enforce block awareness on ECX data (due to the different way that ECX stores blocks on disk), so we have delayed working on making the two work together as of now.

RE Physical disk LED / finding a disk
The disk ID is really an "internal" construct, do not pay much attention to it.

If the LED on a disk was not working, or perhaps a HDD was 100% "dead" and the disk was just 100% offline, you would just simply look for the node that it is attached to (which is visibile in the Prism hardware display diagram) and go to that physical node and remove it.

Alternatively, You could flash the LED on that node as well, which would help, OR perhaps flash all of the LED's of the disks around the dead disk, so that the dead disk stands out as "not flashing"

If there was any doubt, you would compare the serial number from the Disk in Prism to the physical one printed on the drive to double check.

This is the general approach for almost all storage systems.

RE Removing a node
You do not need to svMotion anything at all. The CVM's will move the data off that node.

Even if you did "svMotion" all of those VM's, you'd still have data on that CVM before it is removed from the cluster, as it is a part of the clusters overall capacity, so the svMotion literally doesn't do anything but waste time.

Just regular vMotion the VM's to another node first, then remove the node using Prism and wait for the task to complete.

RE RF3 to RF2 change
Are you talking at the cluster level? or the container level?

Cluster level can not be changed back. I *think* the container level can be changed back, but I've never tried it. Absolute worst case, just provision another container at RF2 and storage vMotion the data.

That said, this question is one of those that you should really think through before arbitrarily changing. This is why the "change RF" command is NCLI only, and not in the GUI.

RE Foundation RF2
This is setting up the "cluster level" RF, but regardless of foundation, if you have a cluster that is RF2, you can go to RF3 with NCLI, yes.

RE Enable/Disable ECX
Sure, you can change that all you want. Just note that encoded data will stay encoded until the data is overwritten.

This is slightly different behavior than enable/disable compression, which does actively compression/decompress data when the setting is changed.
View original

4 replies

Userlevel 7
Badge +30
RE ECX
Yes, ECX and block awareness do not mix. There is a long technical explanation of why, but the short of it is that the majority of customers who would use ECX wouldn't have enough physical blocks to actually enforce block awareness on ECX data (due to the different way that ECX stores blocks on disk), so we have delayed working on making the two work together as of now.

RE Physical disk LED / finding a disk
The disk ID is really an "internal" construct, do not pay much attention to it.

If the LED on a disk was not working, or perhaps a HDD was 100% "dead" and the disk was just 100% offline, you would just simply look for the node that it is attached to (which is visibile in the Prism hardware display diagram) and go to that physical node and remove it.

Alternatively, You could flash the LED on that node as well, which would help, OR perhaps flash all of the LED's of the disks around the dead disk, so that the dead disk stands out as "not flashing"

If there was any doubt, you would compare the serial number from the Disk in Prism to the physical one printed on the drive to double check.

This is the general approach for almost all storage systems.

RE Removing a node
You do not need to svMotion anything at all. The CVM's will move the data off that node.

Even if you did "svMotion" all of those VM's, you'd still have data on that CVM before it is removed from the cluster, as it is a part of the clusters overall capacity, so the svMotion literally doesn't do anything but waste time.

Just regular vMotion the VM's to another node first, then remove the node using Prism and wait for the task to complete.

RE RF3 to RF2 change
Are you talking at the cluster level? or the container level?

Cluster level can not be changed back. I *think* the container level can be changed back, but I've never tried it. Absolute worst case, just provision another container at RF2 and storage vMotion the data.

That said, this question is one of those that you should really think through before arbitrarily changing. This is why the "change RF" command is NCLI only, and not in the GUI.

RE Foundation RF2
This is setting up the "cluster level" RF, but regardless of foundation, if you have a cluster that is RF2, you can go to RF3 with NCLI, yes.

RE Enable/Disable ECX
Sure, you can change that all you want. Just note that encoded data will stay encoded until the data is overwritten.

This is slightly different behavior than enable/disable compression, which does actively compression/decompress data when the setting is changed.
Userlevel 1
Badge +14
Dear Jon,
Thanks for your detailed reply. I`d like to confirm these further more:
--RE ECX--
You mean ECX and block awareness function can`nt be used together as of now.
But I think most users will need block level fault-tolerant and decrease RF data( more usable capacity) in the same time.
also 3 block is a conventional config, need more block to enable ECX together ?

--RE RF3 to RF2 change & RE Foundation RF2--
Just want confirm difference and relationship between Cluster-level and Container Level RF?
In particular, if Cluster-level set to RF3, can container-level be set to RF2 and RF3 (I mean different containers with different RF, I know a container can be set only one RF type simultaneously ) ?
As a opposite, If Cluster-level set to RF2, can container-level be set to RF2 and RF3 ?

In my Current understanding, Cluster-level controls RF of some cluster components , e.g. Zookeeper
and Container-Level for usual vm data.
Also RF of Some cluster components even can be set 5, I`m curious if I can change it.

--RE Enable/Disable ECX--
From what you said, I guess if I disable ECX, No addtional capacity is required ? as data still encoded.
But if I disable compression, Additional capacity must be satisfied for decompression data ? otherwise I may fail to disable it ? if so, how to calculate the required free space ?

TKS. BR
Userlevel 7
Badge +30
RE ECX, Block Awareness
The majority of our customer base views block awareness as a nice to have, but not a need to have, and either way ... is not always possible in customer environments, especially when they are pretty small.

Even when you have block awareness, it is always best effort, so again, a nice to have, but not a need to have. We have had, literally, quantity one customer demand strict block awareness, where they wanted the system to STOP writing data if it could not be block aware. I actually just had this conversation with engineering last week.

Anyhow ... The simple answer here is block awareness and ECX don't mix, which in my opinion really isn't that big of a deal, especially with the system reliability and failure statistics we see global (failure rates are very low, and stability is very high from what we see in call home data)

RE Cluster vs Container RF
If cluster RF is RF3, container can be RF2 or RF3.

If cluster RF is RF2, container can ONLY be RF2, as there would be zero point of container RF3, since we wouldn't be RF3 on the cluster level (for things like metadata, zookeeper, etc)

completely ignore anything that says RF5, that has nothing to do with containers. Thats an internal only construct for metadata, you'll find it in the nutanix bible, and it is 100% purely academic.

RE ECX
Yes, encoded data, if you never touched it ever again, will be encoded for life, so just leave it as is.

If you REALLY want to unencode it, you can svMotion to another container.

RE Compression
There is a counter in the GUI that tells you the pre- and post- reduction space for compression, so thats how you would calculate space needed to decompress (at a high level).

That said, compression is a very mature feature, so we do recommend the vast majority of deployments enable it.
Userlevel 1
Badge +14
Hi Jon,
gotcha ! Thanks very much for you help

Reply