Change Record Size?

[jasonsansone]

I want to change my pool record size to 1MB. I have used zfs replication to move the data off the pool, nuked the pool, and replicated it back. The results of zdb show the record size as unchanged... it is still the default 128KB. I was under the impression that to change compression or record size you copy data off and back... but that isn't working. Can I not use zfs send recv to change record size?

 

[ornias]

pool record size? record size is a per-dataset property....

 

[jasonsansone]

pool record size? record size is a per-dataset property....

Sorry, you are correct. I changed root level dataset to 1MB block size. All child datasets inherit the block size. I believe I figured out that the -R flag is causing the 128 block size to persist. I am testing using -L on send and -o recordsize=1M on receive.

 

[HoneyBadger]

The zfs send/recv commands send records, not files; ergo, the recordsize will not change.

You'll have to use something "above" the ZFS level - rsync, robocopy, Ctrl-C + Ctrl-V - in order to have the files written as new records with the new recordsize=1M.

 

[jasonsansone]

The zfs send/recv commands send records, not files; ergo, the recordsize will not change.

You'll have to use something "above" the ZFS level - rsync, robocopy, Ctrl-C + Ctrl-V - in order to have the files written as new records with the new recordsize=1M.

Tested... confirmed... drat

 

[jasonsansone]

For anyone that ever comes across this via search:

zfs send receive won't expand record size to fit a dataset with a larger record size, but it will shrink if you are migrating to from one dataset with a larger size to a dataset with a smaller size. Omitting the L flag from send (if using larger then 128Kb record size) or specifying "-o recordsize=" on the receiving end will force smaller records.

 

[Johnnie Black]

Omitting the L flag from send (if using larger then 128Kb record size)

You need to be careful with that, or risk running it a known bug that can result in silent data corruption:

FAQ — OpenZFS documentation

openzfs.github.io

 

[jasonsansone]

Interesting and good to know. The particular data I tested with wasn’t critical; it’s a test setup.

 

[Johnnie Black]

Yeah, and that bug only applies to incremental sends, still good to be aware of it.