[LU-12756] Refactor lnet_select_pathway and lnet_peer_ni ref counting

• Type: Improvement
• Resolution: Unresolved
• Priority: Minor
• Assignee: Chris Horn
• Reporter: Chris Horn

Description

lnet_select_pathway() contains a fair amount of duplicated code. I took a stab at refactoring this function and I believe I have a functionally equivalent implementation with about half as much code.

During that process I noted that the lnet_nid2peerni_locked() function takes a ref on the lnet_peer_ni
object that it returns, however most callers of this function do not need this reference as they hold a net lock when referencing the lnet_peer_ni. So some additonal code removal is possible here as well.

Comments

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/36537/
Subject: LU-12756 lnet: Refactor lnet_set_non_mr_pref_nid
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: ce442253a719a065bf85e0202546a3afd4a38524

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/36621/
Subject: LU-12756 lnet: Refactor lnet_compare_routes
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: e02287b4ef6a30e9e23a84012d8a133621ea454e

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/36620/
Subject: LU-12756 lnet: Remove unused vars in lnet_find_route_locked
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: b129f7b1f76a716c656627beb271de58e6af473c

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/36536/
Subject: LU-12756 lnet: Avoid extra lnet_remotenet lookup
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 3812c54b9ca3cc08be947f893bdf55a41aa876ed

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/36535/
Subject: LU-12756 lnet: Avoid comparing route to itself
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 2b8d9d12d182fc91d671558434cc0b652c1ade21

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/36534/
Subject: LU-12756 lnet: Refactor lnet_find_best_lpni_on_net
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 80edb2ad72baf9b096e03f2929b8b018b0a630d2

[Alexey Lyashkov]

@Chris - my vote to add some performance testing against that series.
As I see, it have touched a sort of hot code. So I think we need to add performance evaluation as similar as we have with MR routing testing.
workload is same for all tests - mdtest + IOR with small record size (4k-8k).
1) new code on server
2) new code on routers also -
3) all nodes have same code.

[Chris Horn]

FYI, I plan the following test activities for this patch series.

Port patch series to Cray's Lustre 2.12. All patches apply cleanly because our 2.12 LNet is basically identical to master.

Deploy the code to dev system with Clusterstor server and small XC system (40 compute nodes, 4 LNet routers).

Place load on system (IOR, mdtest, etc.).

Perform LNet router failure testing over 24 hours. This involves: NMI a router, wait 15 minutes, reboot router, wait 15 minutes, repeat.

Expectation is that there are no client evictions and no test failures.

I will perform the above testing for the following configurations:
1. Test code on servers, routers and clients with default settings (Edit: Passed)
2. Test code on servers, routers and clients with discovery disabled on clients and routers (Edit: Passed)
3. Test code on servers, routers and clients with discovery disabled on servers. (Edit: Test fails due to https://jira.whamcloud.com/browse/LU-12955)
4. Test code on servers, routers and clients with discovery disabled on clients servers and routers (Edit: Passed).
5. Test code on servers, clients and multi-rail routers with default settings.

[Gerrit Updater]

Chris Horn ([e-mail]) uploaded a new patch: https://review.whamcloud.com/36624
Subject: LU-12756 lnet: Take peer NI ref in lnet_discover_peer_locked
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 148184da0238d4c533c4d3e9a6ecaebaa8efa635

[Gerrit Updater]

Chris Horn ([e-mail]) uploaded a new patch: https://review.whamcloud.com/36627
Subject: LU-12756 lnet: Refactor lnet_handle_lo_send
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 30a4416bffcf7830203344d6938e1efab16d92fe

[Gerrit Updater]

Chris Horn ([e-mail]) uploaded a new patch: https://review.whamcloud.com/36626
Subject: LU-12756 lnet: Switch lnet_handle_lo_send to void function
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 788b4bd153e2fc7b67a27ca1b2fd51f5a8136651