Skip to content

Conversation

@jsquyres
Copy link
Member

@jsquyres jsquyres commented Oct 4, 2019

@bwbarrett Some notes:

  • See the commit messages for more detail (e.g., about c:r:a)
  • I included a NEWS bullet for the CMA/vader issue (v3.1.x: Do not use CMA in user namespaces #6999), because I'm assuming we'll review+merge it before making the RC
  • This PR is marked WIP/DNM because comparing the NEWS items of v3.0.x and v3.1.x, I see that the v3.0.x branch is missing a few fixes that made it to the v3.1.x branch (I marked these as "NO" in the NEWS file, which needs to be cleaned up before we merge).
    • Specifically: we need to decide whether we want to chase down these fixes for the v3.0.x branch.
    • My $0.02: barring technical problems / significant reasons not to include these fixes in the v3.0.x branch, we should chase them down / get the v3.0.x PRs filed.
    • I did find one v3.1.x PR that can not be applied to v3.0.x: the new regx/naive component on the v3.1.x branch. There is no regx framework here on the v3.0.x branch, so that PR is moot over here.

Refs #6591

Bump up to rc1.  For shared library version numbers:

* libopen-rte changed
* libopen-pal changed
* common OMPI library changed

Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
@jsquyres
Copy link
Member Author

jsquyres commented Oct 8, 2019

Per discussions with @bwbarrett, I tried cherry-picking fixes for the "NO"-labeled items. They all required more than trivial effort except the buffered sends with the cm PML.

The following are the list of NEWS items that were not cherry-picked back, and therefore will not be fixed in the v3.0.x series:

  • Fix MPI one-sided 32 bit atomic support.
  • Allow individual jobs to set their map/rank/bind policies when running LSF. Thanks to Nick R. Papior for assistance in solving the issue.
  • Properly propagate errors to avoid deadlocks in MPI one-sided operations.
  • Fix data corruption in non-contiguous MPI accumulates over UCX.

In particular, one-sided support changed significantly between v3.0.x and v3.1.x; it would not be simple to port those OSC changes back to v3.0.x. The solution for users needing OSC fixes is to upgrade to v3.1.x or v4.0.x.

@jsquyres
Copy link
Member Author

jsquyres commented Oct 8, 2019

@bwbarrett README updated:

Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
@jsquyres
Copy link
Member Author

jsquyres commented Oct 9, 2019

@bwbarrett #7059 is now merged. Both the v3.0.x and v3.1.x README/NEWS/VERSION PRs are ready for your review so that we can make the RCs.

@jsquyres jsquyres merged commit 576df8d into open-mpi:v3.0.x Oct 15, 2019
@jsquyres jsquyres deleted the pr/v3.0.5rc1 branch October 15, 2019 17:23
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants