LFSCK Phase 2 - MDT-OST Consistency Scope Statement

From Lustre Wiki
Revision as of 20:12, 4 February 2019 by Adilger (talk | contribs) (Key Deliverables: add catefory link)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

Problem Statement

MDT-OST consistency will implement functionality for distributed verification and repair of the MDT inode to OST object layout. This will add functionality while the MDT is iterating over the inodes (see LFSCK Phase 1 - Inode Iterator and OI Scrub) to check the file layout (LOV EA) to verify the ostid therein, that the OST objects referenced by the file layout exist, and that each OST object has a back reference to the correct MDT inode. Incorrect or missing back pointers on the OST objects will be corrected, and missing objects will be recreated when detected.

The UID and GID of OST objects will also be verified to match that of the MDT inode to ensure correct quota allocation. After the MDT iteration is complete, any unreferenced OST objects will be linked into the Lustre lost+found directory.

The Inode Iterator and OI Scrub and MDT-OST Consistency together constitute a complete, scalable replacement of the existing e2fsprogs-based lfsck utility. This will allow the distributed checking and repair of Lustre inter-server state for non-DNE file systems while the file system is on-line.

Project Goals

- Distributed verification and repair of MDT inode to OST object mappings. - Measure the resulting code performance characteristics.


  • User-space control code for MDS-OST consistency tool.
  • The detected/repaired inconsistencies will be reported to the administrator.
  • Administrative documentation in the form of a man page and update to Lustre 2.x manual.
  • Documentation for user interface.

Out of Scope

  • Design and implementation will strive to be agnostic to the back-end OSD file system. Only ldiskfs OSDs will be tested and landed for this phase, pending the development of a ZFS object iterator.
  • Only support verification of currently supported LOV_MAGIC_V1 and LOV_MAGIC_V3 RAID-0 file layouts. Verifying new layouts should be done by the projects adding the new layouts.
  • OST iteration will skip object orphan cleanup if the parent FID is on a MDT that has not been processed.
  • Detection and resolution of internal file system inconsistencies is provided by e2fsck and is not within scope of this phase.
  • Additional Lustre specific attributes, for example SOM, are not considered for verification.

Key Deliverables

  • Feature test plan.
  • Source code that meets feature requirements and runs with Lustre 2.x Master branch on customer’s site.
  • Source code for new test cases.
  • MDT-OST consistency code landed in on the Lustre 2.x Master branch.
  • LU tickets for public visibility of MDT-OST Consistency development.