Join GitHub today
GitHub is home to over 28 million developers working together to host and review code, manage projects, and build software together.Sign up
Zpool zol fixes #47224
What does this PR do?
Was going though some of the zfs/zpool tickets, looks liked I did not get tagged on #38671
This is also because the ZFS on Linux version on Ubuntu 16.04 is rather old.
Then manually strip 'name'.
What issues does this PR fix or reference?
The call to
We do not use
Tests update to include new command output.
Commits signed with GPG?
IMHO, this is a bad idea, as not limiting properties names makes zfs states/module quite slow on systems with lots of datasets (mor specifically snapshots), as I already stated at issue #47225.
A better option would be to only ask for specific property names if/when the installed zfs utilities allow for it (something which can be tested by pseudo-parsing "zfs/zpool -h" output).
@pruiz we're still limiting property names!
The older ZFS does not allow limiting of the returned fields for the property we request. This should not introduce a noticable performance impact.
@sjorge yeah, I understand older zfs versions do not allow for '-o' argument. However, on new zfs versions supporting it, not passing '-o' (and sometimes -s too) when querying/filtering datasets with lots of snapshots, the command takes minutes to complete.
So my request was: instead of removing the '-o' usage here. Try to guess wether the installed zfs command supported it or not (for example by parsing zfs help output with a regex like '^\slist .[-o .*'), so you can dynamically decide wether to pass '-o' or not.
@pruiz Do you have any examples I have not seen any performance impact of this. I only had up to about a thousand datasets though.
Difference would be nice data to have. (Now mind you dropping the 'size' one and fetching all properties does have an impact but that is not what this particular change does.
My main concern is not with 'zpool', but with 'zfs' command, see this for example:
This machine is somewhat old (Xeon Server from 2012), with 4xSAS disks on raidz1. But still not uncommon, nor specially slow on anything else, but listing zfs datasets..
This PR does not touch the zfs list. I went over all the commands and zpool list was the only one that is currently broken on ZoL due to the use of
Not as large as your pool, but here is a largish one. Same as you it is slowish when using zfs list.