[LU-205] Build on master broken with ext3

• Status: Resolved
• Project: Lustre
• Type: Bug
• Priority: Major
• Reporter: Zhenyu Xu
• Assignee: Zhenyu Xu
• Resolution: Won't Fix
• Severity: 3

Description

The build on master with ext3-based ldiskfs is broken due to the RHEL6 landings:
1) shadow's ext3-export-64bit-name-hash patch conflicts with the no-mb-cache one.
2) *_discard_preallocations() only makes sense for ext4, not for ext3.
3) In file included from lustre/lvfs/fsfilt-ldiskfs.c:58:
ldiskfs/ldiskfs/linux/ldiskfs_jbd.h:39: error: "LDISKFS_SINGLEDATA_TRANS_BLOCKS_HAS_SB" redefined
config.h:707: note: this is the location of the previous definition

Comments

[Zhenyu Xu]

patch posted at http://http://review.whamcloud.com/418

[Brian Murrell (Inactive)]

Do you have a URL to a build that fails? I see the last couple of builds at https://build.whamcloud.com/job/lustre-master/ have passed.

[Andreas Dilger]

I domt think this needs to be fixed. We need a bunch if features that ate only available in the ext4 code, and dropping ext3 support will reduce the number of patches that we have to maintain.

[Johann Lombardi (Inactive)]

> Do you have a URL to a build that fails?

ext4 is the default, that's why builds don't fail.

> I don't think this needs to be fixed. We need a bunch if features that ate
> only available in the ext4 code, and dropping ext3 support will reduce the
> number of patches that we have to maintain.

2.0 was already intended to use ext4 as a default and this was changed at the
last minute due to performance & eviction issues.
Since 2.1 has still not been fully tested, i think it makes sense to keep the
ext3 option opened, all the more since it is very easy to fix (it should not have
been broken in the first place imo).
Also, if we have a performance problem, we can easily retest with ext3-based ldiskfs.

[Brian Murrell (Inactive)]

If we are going to want to build both ext3 and ext4 based ldiskfses then the building of them needs to be fixed once and for all (as I had described way back when we started building dual sets of packages for ext3 and ext4) so that one build run builds both and as RPMs they are interchangeable (i.e. allowing the user to decide which to use simply by selecting a different ldiskfs RPM). Having an explosion of build jobs that builds a full set of packages for all combinations of options doesn't scale.

[Peter Jones]

I think that this is now obsolete as the ext3 patches were recently removed from master