[LUDOC-87] OSD-Mount Doc Changes

• Status: Resolved
• Project: Lustre Documentation
• Type: Improvement
• Priority: Critical
• Reporter: Jodi Levi (Inactive)
• Assignee: Alex Zhuravlev
• Resolution: Fixed
• Project: Orion

Description

Changes to format and mount options

Issue Links

• is related to: LU-1711 Server Stack

Comments

[Andreas Dilger]

Update the user manual to describe changes to the user tools:

• formatting, mounting, usage, configuration, etc.
• potentially some of the example output (if relevant to the examples)
• /proc tunables, module parameters, etc.

Since we use a single manual for all of the Lustre releases, when changing any significant sections of text the content should include the version in which the change was made (i.e. 2.4) and leave the old comments in place (at least back to 2.1, older 1.x sections can be removed), for example:

Up to Lustre 2.3 running writeconf was via tunefs.lustre --writeconf /dev/mdtdev. Starting with Lustre 2.4 running writeconf should be use ...

[Alex Zhuravlev]

I've gone through the manual and got the following TODO:

13.14.�Identifying To Which Lustre File an OST Object Belongs – check

14.2.� Finding Nodes in the Lustre File System

cat /proc/fs/lustre/lov/lustre-mdtlov/target_obd – lustre-MDTXXXX-mdtlov

14.8.5.� Restoring OST Configuration Files – check

14.12.� Separate a combined MGS/MDT – check

17.1.1.1.�Using Lustre_rsync – check

24.3.1.�Testing Local Disk Performance – load ofd module since 2.4

24.3.4.2.�Visualizing Results - brw_stats in /proc/.../osd-*/brw_stats

30.1.4.�OST Failure (Failover) - s/LOV/LOD/

31.1.5.�Free Space Distribution – lustre-MDTXXXX-mdtlov

31.2.6.�Watching the OST Block I/O Stream – in OSD now

31.2.8.1.�Using OSS Read Cache - in OSD now

36.14.3.�Examples – should we specify MDT index always?

36.15.3.�Options – check md_stripe_cache_size still works

36.18.3.�Options – check --mountfsoptions=opts

• • describe LOV->LOD, OSC->OSP change on MDS ?
  • new parameters in LOD (under /proc/fs/lustre/osc/fsname-OST*-osc-MDT*/:
    sync_changes, sync_in_flight, sync_in_progress
  • ZFS support

any comments?

[Alex Zhuravlev]

– 13.14.�Identifying To Which Lustre File an OST Object Belongs – check
probably we need to mention "0" in /O/0/d$((34976 % 32))/34976" is a sequence number and with DNE it won't be 0, instead it should be taken properly from lfs getstripe output

– 14.2.� Finding Nodes in the Lustre File System

cat /proc/fs/lustre/lov/fsname-mdtlov/target_obd - s/fsname-mdtlov/fsname-MDT<index>-mdtlov/target_obd

– 14.8.5.� Restoring OST Configuration Files

should be OK

– 14.12.� Separate a combined MGS/MDT – check

should be OK

– 17.1.1.1.�Using Lustre_rsync – check

should be OK

– 24.3.1

starting from 2.4 we load ofd.ko instead of obdfilter.ko

24.3.4.2.�Visualizing Results

since 2.4 brw_stats should be checked in /proc/.../osd-*/brw_stats

30.1.4.�OST Failure (Failover) - s/LOV/LOD/

since 2.4 MDS uses LOD instead of LOV

31.1.5.�Free Space Distribution

cat /proc/fs/lustre/lov/fsname-mdtlov/qos_prio_free – it's actually /.../fsname-MDTXXXX-mdtlov/ since some previous version

31.2.6.�Watching the OST Block I/O Stream – in OSD now

oss# lctl get_param obdfilter.testfs-OST0000.brw_stats - osd-*.testfs-OST0000.brwstats since 2.4

31.2.8.1.�Using OSS Read Cache

lctl set_param osd-ldiskfs..read_cache_enable=0 – osd-.*.read_cache_enable since 2.4, doesn't apply to osd-zfs

36.18.3.�Options

--mountfsoptions=opts should be OK

[Alex Zhuravlev]

with the latest changes to master branch, we can keep using obdfilter.* to access brw_stats. we should handle read_cache_enable similarly.

[Alex Zhuravlev]

http://review.whamcloud.com/5873 – with the patch read_cache_enable, readcache_max_filesize and writethrough_cache_enable can be accessed via obdfilter.*, so we wouldn't need to change the manual.

[Alex Zhuravlev]

compatibility bits are landed