abouteiller commented 2 years ago

Problem

The monolithic ULFM proposal has been split in morsels so that the MPI Forum can focus on individual topics.

Main topic issue https://github.com/mpi-forum/mpi-issues/issues/20

Proposal

The first topic slice contains the following concepts for communicators:

Chapter structure and introduction material
Error codes
Error raising and semantics after errors are raised
MPI_COMM_REVOKE
MPI_COMM_GET_FAILED
MPI_COMM_ACK_FAILED

Changes to the Text

Addition of an FT chapter containing the proposed constructs

Impact on Implementations

Implementations optionally to implement fault tolerance. Implementations to add procedures MPI_COMM_REVOKE, MPI_COMM_GET_FAILED, MPI_COMM_ACK_FAILED (implementations that do not support FT can provide stubs that are not fault tolerant).

Impact on Users

Users can receive fault events, write Manager-Worker applications, and run-through (e.g., Stencil, ABFT) type workloads that use only P2P operations (slice2 and 3 will add features for repairing communicators as needed to use collective and process respawning after a fault).

References and Pull Requests

https://github.com/mpi-forum/mpi-standard/pull/665

wesbland commented 2 years ago

I believe this is the set of changes for the no-no vote on 2022-09-30:

https://github.com/mpi-forum/mpi-standard/pull/665/files/92f7596e8958dfc3a71bbc83514dec3d3b7dcc07..43864f6499e0e8fdb25534625397761e256c7479

wesbland commented 2 years ago

This issue had a "no-no" vote on 2022-09-30, which passed:

Yes	No	Abstain
28	0	1

wesbland commented 1 year ago

This passed a first vote on 2022-12-07.

Yes	No	Abstain
28	0	5