[LU-17995] In-memory OSD device

• Type: New Feature
• Resolution: Unresolved
• Priority: Minor
• Severity: 3
• Assignee: Tim Day
• Reporter: Tim Day

Description

Implement a simple in-memory OSD device for Lustre. This enables Lustre servers to be run without any backing filesystem, such as ldiskfs or openZFS.

The main goal of this OSD is to make it easier to test Lustre on kernels which don't readily ldiskfs or openZFS. It will also make it simpler to set up a new Lustre dev environment, since many changes to Lustre do not depend on a particular filesystem.

 

Issue Links

• is blocked by: LU-18545 Autotest doesn't install OSD packages when using fstype=mem
• is related to: LU-18813 memory FS based OSD with write-back support
• is related to: LU-17848 OSD and dt cleanups / deduplication
• is related to: LU-18162 Improve lu_device/obd_device handling and Lustre mount state machine
• is related to: LU-18766 In-memory OSD sparse file handling

Comments

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/58174/
Subject: LU-17995 obdclass: remove obdname2fsname()
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: c40370698e32aafba5c0323f957bd700e3353cd9

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/58163/
Subject: LU-17995 osd-zfs: remove server_name_is_ost()
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 4a7b1f6ad3e3f4c0dec35210b50240826fc4e8a2

[Gerrit Updater]

"Timothy Day <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/58174
Subject: LU-17995 obdclass: remove obdname2fsname()
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 3bd06be6c011dd932f2570503b565a1189a7246f

[Gerrit Updater]

"Timothy Day <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/58163
Subject: LU-17995 osd-zfs: remove server_name_is_ost()
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: e155e4eaec94f6329ba7a280077b33c9ec7e57dd

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/57336/
Subject: LU-17995 libcfs: implement libcfs_subsystem_debug
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 627c4b2b93484818f13c3e7158593ae0c097bf6c

[Gerrit Updater]

"Timothy Day <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/57336
Subject: LU-17995 libcfs: implement libcfs_subsystem_debug
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 9ec8bfce24b204fa74086d52e18e4c71b843bc42

[Gerrit Updater]

"Timothy Day <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/57287
Subject: LU-17995 osd-mem: test
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: bbc088d5b51ded29f144b3fb5e7b5a638e71553b

[Tim Day]

Once I get to a base level of correctness, I going to spend some time speeding it up. There's going to be a lot of low hanging fruit perf wise. For persistence, it shouldn't be to hard to periodically serialize objects to disk a la Redis - https://redis.io/docs/latest/operate/oss_and_stack/management/persistence/. I think that'd be useful for debugging.

[Patrick Farrell (Inactive)]

Thanks, Tim - makes sense.  I suspect someone will eventually decide they want the in-memory OSD to be fast, too.  I can imagine someone wanting to shuffle IO off to a server and have it stored later with the cost of loss, etc.  (I'm not saying I think it's an idea with a huge window of utility, but I am saying I think someone will do it. )

[Tim Day]

The use case is narrow, but I think the amount of work is (relatively) small. I think the biggest hassles were (a) recreating the necessary boilerplate to get the OSD to start correctly and (b) outdated documentation. Now that I'm more familiar with the code, it shouldn't be too hard to improve the API and docs - writing OSDs could be made much simpler.

There might be other use-cases outside of testing and improving/understanding the APIs, but those were my main motivators. An in-memory OSD would be handy to have; the Lustre arcana involved dissuaded people from writing one, I think.