vdev replication

[TorbenMartin]

hello.

i have a pool:

    NAME                        STATE     READ WRITE CKSUM
    storage                     ONLINE       0     0     0
      mirror-0                  ONLINE       0     0     0
        pci-0000:2c:00.0-ata-1  ONLINE       0     0     0
        pci-0000:2c:00.0-ata-2  ONLINE       0     0     0

mirror-0 is full 70%

when i add more vdevs like:

    NAME                        STATE     READ WRITE CKSUM
    storage                     ONLINE       0     0     0
      mirror-0                  ONLINE       0     0     0
        pci-0000:2c:00.0-ata-1  ONLINE       0     0     0
        pci-0000:2c:00.0-ata-2  ONLINE       0     0     0
      mirror-1                  ONLINE       0     0     0
        pci-0000:3c:00.0-ata-1  ONLINE       0     0     0
        pci-0000:3c:00.0-ata-2  ONLINE       0     0     0
      mirror-2                  ONLINE       0     0     0
        pci-0000:4c:00.0-ata-1  ONLINE       0     0     0
        pci-0000:4c:00.0-ata-2  ONLINE       0     0     0

will zfs automatic rebalance the data between the vdevs?

70%/3 = 23,3

mirror-0 = 23,3%
mirror-1 = 23,3%
mirror-2 = 23,3%

?

and if not... ho can i rebalance the data?

[amotin]

will zfs automatic rebalance the data between the vdevs?

No. ZFS does not do it. Newly written data should gradually fix imbalance with time.

and if not... ho can i rebalance the data?

You'd need to read and rewrite the data and make sure that old data are not held by snapshots. Also the trick won't work if dedup or block cloning is enabled unless data are complete unloaded and deleted, since their goal is to avoid extra copies.

[TorbenMartin]

will zfs automatic rebalance the data between the vdevs?

No. ZFS does not do it. Newly written data should gradually fix imbalance with time.

and if not... ho can i rebalance the data?

You'd need to read and rewrite the data and make sure that old data are not held by snapshots. Also the trick won't work if dedup or block cloning is enabled unless data are complete unloaded and deleted, since their goal is to avoid extra copies.

Thanks.

1. Newly written data should gradually fix imbalance with time.

what will happen when i do this:

mirror-0 = 70%
mirror-1 = 100%
mirror-2 = 100%

when i write more data on it... it will be stored on mirror-0 ?

2. read and rewrite the data

That means:

mirror-0 = 70% - File A
mirror-1 = 0%
mirror-2 = 0%

I just have to copy File A as File A.1 , delete File A and have something like this?

mirror-0 = 68% - File A.1/3
mirror-1 = 1% - File A.1/3
mirror-2 = 1% - File A.1/3

[amotin]

Something like that, though it is not trivial to predict now. I have rethinking of it on my radar, but haven't got to it yet.

[IvanVolosyuk]

If there is no dedup, you can just send / receive the data locally, if it changes you can do that with a snapshot, then with a few incremental send / receives, in the end you will have a brief downtime when you have to stop the writes and move old dataset away and new in place of the old one. The smaller the granularity of datasets the better with files being a smaller unit of work second only to individual blocks, but it works only if you don't have snapshots.

…

On Wed, Aug 7, 2024, 08:58 Alexander Motin ***@***.***> wrote: Something like that, though it is not trivial to predict now. I have rethinking of it on my radar, but haven't got to it yet. — Reply to this email directly, view it on GitHub <#16420 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABXQ6HNLDFQDGPSYEL7XUZ3ZQI72DAVCNFSM6AAAAABMEVSGHKVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTAMRWGY2DQNA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>