ucx icon indicating copy to clipboard operation
ucx copied to clipboard

UCP/WIREUP: Handle EP Reconfiguration under traffic

Open shasson5 opened this issue 8 months ago • 3 comments

What?

Handle EP Reconfiguration under traffic

Why?

Support RC/DC switching

How?

  1. Wait for flush on outstanding messages in discarded uct EP to be completed before next wireup step
  2. Added "deferred EP" hash table to track deferred data (such as pending requests, flush requests, reply data)
  3. Defer pending messages replay until wireup completion.
  4. Fix unhandled reconfig usecase (reused AM lane is selected for new config).
  5. Added return value to usage tracker callback in order to update promote status.

shasson5 avatar Apr 21 '25 07:04 shasson5

/azp run

shasson5 avatar Apr 22 '25 16:04 shasson5

Azure Pipelines successfully started running 5 pipeline(s).

azure-pipelines[bot] avatar Apr 22 '25 16:04 azure-pipelines[bot]

@roiedanino please review

gleon99 avatar May 05 '25 06:05 gleon99