Thanks to everyone who participated. Sorry for the delay in getting this
email sent!
Meeting recording is now posted: https://www.youtube.com/watch?v=tyCk376bC9A
Notes (thanks Christian):
-
Encryption Bugs (Rich Ercolani)
-
Rich’s spreadsheet is now publicly accessible
<https://docs.google.com/spreadsheets/d/1OfRSXibZ2nIE9DGK6swwBZXgXwdCPKgp4SbPZwTexCg/edit#gid=1560550070>
-
One new bug was reported that doesn’t involve replication: #1293
<https://github.com/openzfs/zfs/issues/12931>1 (dereferencing a wild
pointer in the checksum code with encryption+dedup on under heavy load)
-
Compressors (Rich Ercolani)
-
Issue that sparked the discussion: #12840
<https://github.com/openzfs/zfs/issues/12840>
-
Summary: Replacing the compressor code causes problems if the
output is not bit-for-bit identical.
-
Problem #1: Uncompressed ARC + Compressed On-Disk + L2ARC
-
Assume uncompressed ARC, which holds the data of a block that
is compressed on disk.
-
The L2ARC invariant is that it stores a bit-for-bit identical
copy of the on-disk data.
-
So, the L2ARC write must recompress the uncompressed copy.
-
But if the compressor changed, the recompressed copy will not
be bit-for-bit identical to the on-disk data, and the
L2ARC read will fail
with a checksum error, resulting in an L2ARC miss.
-
Problem #2: NOP writes
-
TODO explain
-
Problem #3: Dedup
-
Hashing is against the compressed blocks. Thus, blocks
compressed with the new compressor will not dedup with
blocks compressed by
the old compressor.
-
Initial consensus: all three problematic cases are very minor. We
can just replace-upgrade decompressor and compressor.
-
We already did that inadvertently with the different gzip
implementations (QAT, Linux, FreeBSD).
-
Uncompressed ARC in general should be considered an edge case.
Users should expect that it might not interact perfectly
with all other
features (such as L2ARC).
-
For L2ARC, we can add a small check that marks recompressed
buffers which are not bit-for-bit identical ineligible for L2ARC.
-
LZ4 (Rich) and ZSTD are two concrete upgrades that are planned.
-
Special treatment of the LZ4 compressor upgrade?
-
LZ4 is the default on many systems.
-
A replacement-upgrade of the LZ4 compressor would thus affect
many users of dedup.
-
So, should we require users to opt-in for LZ4 upgrades, e.g.,
via a feature flag? This would require keeping multiple
LZ4 compressor
versions around.
-
Decision: Wait for 1 month to see whether someone comes up with
a scheme that requires user opt-in. If there is no
solution, go ahead with
a replacement-upgrade.
-
NB: For ZSTD, which changes more frequently, we might not want to
maintain multiple compressor versions. So, whatever will be
decided for LZ4
does not set a precedent for ZSTD.
On Mon, Jan 3, 2022 at 3:52 PM Matthew Ahrens <mahr...@delphix.com> wrote:
> The next OpenZFS Leadership meeting will be held tomorrow, January 4, 1pm
> Pacific time.
> Everyone is welcome to attend and participate, and we will try to keep the
> meeting on agenda and on time. The meetings will be held online via Zoom,
> and recorded and posted to the website and YouTube after the meeting.
>
> For more information and details on how to attend, as well as notes and
> video from the previous meeting, please see the agenda document:
>
>
> https://docs.google.com/document/d/1w2jv2XVYFmBVvG1EGf-9A5HBVsjAYoLIFZAnWHhV-BM/edit
>
> --matt
>
------------------------------------------
openzfs: openzfs-developer
Permalink:
https://openzfs.topicbox.com/groups/developer/Td6ebc6df2e2a14bd-Md73cea97245493e922c7e523
Delivery options: https://openzfs.topicbox.com/groups/developer/subscription
