[LU-24] LNET selftest improvements

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.0.0, Lustre 2.1.0, Lustre 1.8.6
• Type: Improvement
• Priority: Minor
• Reporter: John Spray (Inactive)
• Assignee: John Spray (Inactive)
• Resolution: Incomplete
• Bugzilla ID: 15,646

Description

(Lustre bugzilla 15646 - LNET selftest to measure latency)
(Lustre bugzilla 17328 - minor issue for lnet selftest)

Intention is to:

• Bring Liang's stats cleanup (bz17328 att 21207) up to master.
• Make the ping test use the png_size parameter
• Report latency measured from the ping test in lst stat.

Also various typos in messages, and want to sort out the way that utils/lst lets you put in any old size for the brw test, but in the lnet_selftest module it gets rounded down to the nearest page.

Comments

[Liang Zhen (Inactive)]

John, I made a quick look through this patch, seems you are improving current ping_test to make it support ping_size and measure latency.
It's not a problem if all nodes in cluster upgrade to the same version with this patch in the future, but if customer can't upgrade all nodes in one batch (i.e: some nodes are using 1.8.x and some are using 2.x), then they can't use LST anymore because this patch have changed some wired data (structures with WIRE_ATTR).
Probably we can add a new command like "ping_pro" or "ping_adv" to LST, and add some wired structures (request/reply) for this new command, so even customer upgraded some of their nodes, they still can use LST with regular "lst ping"

How do you think?

[John Spray (Inactive)]

Here's a hopefully backwards-compatible version – there's quite of a lot of code, because the stats reply structure didn't have any spare space for the ping stats.

The existing stats request structure did have a 'str_type' field, but this wasn't being used. If we started using this, then old nodes would ignore it, and when querying them we wouldn't know whether they were returning something valid or not. However, the new 'XSTAT' type is quite verbose for not that much benefit, so we might want to tolerate a bit of ugliness to make the code shorter.

[John Spray (Inactive)]

Cleaning up: this has been idle for >2yrs.