ucx icon indicating copy to clipboard operation
ucx copied to clipboard
verified publisher icon openucx

UCP/WIREUP: Support Race EP reconfiguration - Deprecated

Open shasson5 opened this issue 2 years ago • 18 comments

What

Support reconfiguration of EP in case that dest EP selected different config. This PR handles race use case.

Why ?

Solve wire compatability issues and prepare for RC/DC support.

How

  1. Remove unused assertions to allow non CM flow to be reconfigured.
  2. Lanes intersection is done by checking which lanes are already connected to a remote side (instead of comparing dst_rsc_index).
  3. Local-remote lanes mapping is keeping reused lanes with the same mapping, and only assigns new mapping to new lanes.
  4. Allow wireup_ep to be reconfigured.
  5. Fix AUX lane selection to support reconfiguration.
  6. Disable EP connection flags (LOCAL/REMOTE_CONNECTED) during reconfiguration.

shasson5 avatar Jul 23 '23 15:07 shasson5

@shasson5 In PR description:

  1. As it is not trivial, please add the "How" section.
  2. Motivation ("why") - also for RC/DC support preparation.
  3. Start sentences with capital letters.

gleon99 avatar Jul 24 '23 06:07 gleon99

pls consider adding gtests for non-draft version

brminich avatar Jul 24 '23 06:07 brminich

Please review pipeline failures, they are relevant.

gleon99 avatar Jul 24 '23 08:07 gleon99

/azp run

shasson5 avatar Aug 27 '23 15:08 shasson5

Azure Pipelines successfully started running 4 pipeline(s), but failed to run 1 pipeline(s).

azure-pipelines[bot] avatar Aug 27 '23 15:08 azure-pipelines[bot]

/azp run perf

Alexey-Rivkin avatar Aug 27 '23 16:08 Alexey-Rivkin

Azure Pipelines successfully started running 1 pipeline(s).

azure-pipelines[bot] avatar Aug 27 '23 16:08 azure-pipelines[bot]

/azp run

shasson5 avatar Sep 12 '23 11:09 shasson5

Azure Pipelines successfully started running 3 pipeline(s), but failed to run 1 pipeline(s).

azure-pipelines[bot] avatar Sep 12 '23 11:09 azure-pipelines[bot]

as we discussed, this PR is not yet ready for review, will update when it is ready

shasson5 avatar Sep 18 '23 07:09 shasson5

@gleon99 @yosefe please review

shasson5 avatar Sep 22 '23 12:09 shasson5

Looks mostly ok. @yosefe ?

gleon99 avatar Oct 03 '23 15:10 gleon99

/azp run

shasson5 avatar Nov 07 '23 09:11 shasson5

Azure Pipelines successfully started running 3 pipeline(s), but failed to run 1 pipeline(s).

azure-pipelines[bot] avatar Nov 07 '23 09:11 azure-pipelines[bot]

/azp run

yosefe avatar Jan 15 '24 09:01 yosefe

Azure Pipelines successfully started running 4 pipeline(s).

azure-pipelines[bot] avatar Jan 15 '24 09:01 azure-pipelines[bot]

@shasson5 please change the PR title so that it represents its goal and what it does (handle / fix / support / etc).

gleon99 avatar May 06 '24 19:05 gleon99

In PR description:

  1. dsr_rsc_index -> dst_rsc_index.
  2. Please add some details in the "How" part.

gleon99 avatar May 07 '24 07:05 gleon99