On 5/18/19 2:02 PM, Sean Fagan wrote:
On May 18, 2019, at 1:47 AM, Allan Jude <[email protected]> wrote:
Writing a file of all zeros doesn't test the compression, since if compress !=
off, we check for the 'all zeros' case and write a hole, rather than actually
allocating space.
That has not been my experience, on FreeBSD or ZoL:
root@linux-vm:/tank# stat */bigfile
File: DS1/bigfile
Size: 1073741824 Blocks: 1 IO Block: 131072 regular file
Device: 33h/51d Inode: 2 Links: 1
Access: (0644/-rw-r--r--) Uid: ( 0/ root) Gid: ( 0/ root)
Access: 2019-05-17 19:15:05.755556266 +0000
Modify: 2019-05-17 19:15:08.069527862 +0000
Change: 2019-05-17 19:15:08.069527862 +0000
Birth: -
File: DS2/bigfile
Size: 1073741824 Blocks: 2097365 IO Block: 131072 regular file
Device: 34h/52d Inode: 2 Links: 1
Access: (0644/-rw-r--r--) Uid: ( 0/ root) Gid: ( 0/ root)
Access: 2019-05-18 18:00:37.259314881 +0000
Modify: 2019-05-18 18:00:46.794556728 +0000
Change: 2019-05-18 18:00:46.794556728 +0000
Birth: -
(DS2 was "zfs create -o compression=off tank/DS2", and bigfile was created the
same way on both datasets).
It looks like I expect there, DS1 uses 1 block, and DS2 uses 2 million
blocks. It would be easier to see by comparing 'du file' and 'du -A file'.
If you write something else compressible, you should be able to use zdb to see
if it ended up using lzjb or lz4.
How? :)
cat /usr/src/sys/kern/*.c /usr/src/sys/kern/*.c > DS1/srcfile
or even write a file of all 1s instead of all 0s.
Sean.
--
Allan Jude
------------------------------------------
openzfs: openzfs-developer
Permalink:
https://openzfs.topicbox.com/groups/developer/Tcfdc8da2da71e1c0-M93b508b74b540c48194256b0
Delivery options: https://openzfs.topicbox.com/groups/developer/subscription
