Yes, it was very annoying that I couldn't "fix" the issue by turning
off all the feature@ things as once they are in use, you can't turn
them off. Had to rebuild from scratch. Thanks for fixing! I will grab a daily image with partman 43 in it and give it a go. Regards, Mike On 02/27/2015 04:34 PM, Steven
Chamberlain wrote:
tags 775395 + confirmed user 775395 debian-bsd@lists.debian.org usertags 775395 + kfreebsd thanks Hi Mike, On 09/02/15 03:02, Michael Milligan wrote:This may help... 10.1 added a bunch of feature flags that grub may need to now support in the ZFS code and are probably tripping this all up.You're absolutely right, some of the new feature flags are breaking GRUB; I just confirmed it in a fresh test-install with ZFS mounted on /boot This is my fault; I was so sure that GRUB in jessie supported these, that I requested that `zpool create` no longer use "-d": http://anonscm.debian.org/cgit/d-i/partman-zfs.git/commit/?id=46b42715da4ee17c95762129600958ff75f6d02d GRUB *does* support feature@lz4_compress, but otherwise, only features that are 'read-only compatible' according to zpool-features(7): http://man.freebsd.org/zpool-features Here is the list of "not read-only compatible' features that are supported by GRUB - currently only lz4_compress: http://anonscm.debian.org/cgit/pkg-grub/grub.git/tree/grub-core/fs/zfs/zfs.c#n277 The most problematic ones are hole_birth and embedded_data, because they go "active" as soon as enabled. Some of the others wouldn't initially be a problem, until that functionality is used, but it's not safe to enable those by default. I think most of the new features are valuable and we should keep them if possible, most certainly lz4_compress (which can be used to compress metadata, even if you don't set compression=on for a dataset). So I suggest we use "-d" but still enable lz4 and all read-only compatible features. (`zpool create` without "-d" would enable *all* features by default). Thanks for reporting this! Regards, |