Saso Kiselkov
2013-08-02 11:14:55 UTC
Here's a quick review. ATM when strictly bind data compression in the
L2ARC to the 'compression' setting on the dataset, even though metadata
is always compressed on the pool regardless of this setting. As a
result, unless compression is enabled on the dataset, we won't compress
it in L2ARC, leading to some potentially suboptimal corner cases (e.g.
compress=off and secondarycache=metadata means we won't compress
anything in the L2ARC, even though it is all metadata which is
compressible). This patch changes the behavior so that metadata blocks
are always compressed in the L2ARC (in accordance with how we deal with
them on the main pool) and user-data blocks are compressed according to
the 'compress' property setting on the dataset.
Issue: https://www.illumos.org/issues/3964
Webrev: http://cr.illumos.org/~webrev/skiselkov/3964/
Cheers,
L2ARC to the 'compression' setting on the dataset, even though metadata
is always compressed on the pool regardless of this setting. As a
result, unless compression is enabled on the dataset, we won't compress
it in L2ARC, leading to some potentially suboptimal corner cases (e.g.
compress=off and secondarycache=metadata means we won't compress
anything in the L2ARC, even though it is all metadata which is
compressible). This patch changes the behavior so that metadata blocks
are always compressed in the L2ARC (in accordance with how we deal with
them on the main pool) and user-data blocks are compressed according to
the 'compress' property setting on the dataset.
Issue: https://www.illumos.org/issues/3964
Webrev: http://cr.illumos.org/~webrev/skiselkov/3964/
Cheers,
--
Saso
Saso