[LU-18993] Add latency stats to rpc_stats

• Type: Improvement
• Resolution: Unresolved
• Priority: Minor
• Severity: 3
• Assignee: Patrick Farrell
• Reporter: Patrick Farrell

Description

Implement separate read and write RPC latency histograms in the OSC layer to track the time between RPC submission and completion. This provides valuable performance metrics for diagnosing client-server communication issues, similar to existing size histograms.

(Ticket description by Augment)

Issue Links

• is related to: LU-18711 Add latency counters
• is related to: LU-18934 report latency stats by IO size

Comments

[Gerrit Updater]

"Patrick Farrell <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/59175
Subject: LU-18993 osc: Add RPC latency statistics by size
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 2f10d5a4a19fc92e7afbc98084884b654961649a

[Andreas Dilger]

It may also be useful to have a stat somewhere that tracks the number of sync writes for debugging purposes.

[Andreas Dilger]

See LU-18934 for similar work that is in progress on the server side.

Note that modifying existing stats files can be problematic for tools that scrape these stats, though I suspect it is less of a concern on the client than on the server...

[Patrick Farrell]

Note this info is already available in other places, but is harder to get.

[Gerrit Updater]

"Patrick Farrell <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/59172
Subject: LU-18993 osc: Add separate read/write RPC latency histograms
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: a1c4ef8c294395a241e5edf25d3ad96b4aad2dbc