# -*- text -*-
#
# Copyright (c) 2016      Intel, Inc. All rights reserved.
# Copyright (c) 2024-2025 Nanook Consulting  All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
#
#
[listener-thread-start]
The PMIx server's listener failed to start. We cannot
continue.
#
[collection-mismatch]
Collection strategy during a collective operation (e.g., PMIx_Fence)
is not homogeneous - at least one participating node has a
collection strategy (i.e., collect vs non-collect) that differs
from the other nodes. This must be corrected in order for us
to proceed.
#
[bad-singleton]
The PMIx server was given a malformed value for the PMIX_SINGLETON
directive:

  Value: %s

The value must be a string of the form "nspace.rank", where "rank"
is a valid (non-negative) rank - for example, "singleton.0". The
server cannot register the singleton and is unable to continue.
#
[bad-singleton-type]
The PMIx server was given a PMIX_SINGLETON directive of the wrong
data type:

  Given type:    %s
  Required type: PMIX_STRING

The value must be a string of the form "nspace.rank" - for example,
"singleton.0". The server cannot register the singleton and is
unable to continue.
#
[atomic-completion-unsupported]
The host environment answered a request with PMIX_OPERATION_SUCCEEDED,
which is not a permitted return for this operation:

  Operation:   %s

This operation reports a result - the namespace of a job that was
launched, or the data that was requested - and the only channel for
that result is the callback function the PMIx server library supplied.
That library always supplies one. There is therefore no way to return
a result alongside PMIX_OPERATION_SUCCEEDED, and no way for the
requesting process to learn the answer.

A host that has completed the work immediately should invoke the
callback function - it may do so before returning - and return
PMIX_SUCCESS. The request has been failed to its requestor.
#
[bad-keepalive]
The PMIx server was given a malformed value for the PMIX_KEEPALIVE_PIPE
environment variable:

  Value: %s

The value must be the number of an open file descriptor - the read end
of a pipe held open by the process that started us, so we learn when it
goes away. The server cannot honor the request and is unable to
continue.
#
[bad-server-rank]
The PMIx server was given a malformed value for the PMIX_SERVER_RANK
environment variable:

  Value: %s

The value must be a simple non-negative number naming this server's
rank within its namespace. The server cannot determine its own
identity and is unable to continue.
#
[bad-directive-type]
The PMIx server was given a directive of the wrong data type:

  Directive:     %s
  Required type: %s
  Given type:    %s

A key states only what the directive is, never what its value holds, so
the server cannot use this one. It is unable to continue.
#
[iof-files-not-owner]
The output files of a job were not written:

  Job:            %s
  Owner's uid:    %lu
  Server's uid:   %lu

A job's output files are created as its owner, and this server runs as
another user that cannot act as the owner. The job's output is not
written to files here.

A launcher can write them itself, with its own permissions: one whose
PMIx library supports PMIX_IOF_FILE_WRITER does so for the jobs it
starts. Updating the launcher, or running it against a server that runs
as root or as the job's owner, writes the files.
#
[iof-file-not-opened]
An output file of a job was not written:

  Job:            %s
  File:           %s
  Owner's uid:    %lu

A job's output files are created as its owner, with the owner's
permissions, and this file - or a directory on the way to it - could
not be created or opened as the owner. A directory the owner cannot
write is the usual cause. The job's output is not written to this file.

A server running as root does this work in a helper process running as
the owner, and abandons it if it takes longer than the runas_timeout
MCA parameter allows (10 seconds by default) - a filesystem that does not
respond, say. When that happens, none of the job's other output files is
tried either.

This is reported once per job; any other output file of the job that
cannot be opened is not written either.
#
[iof-file-backlog]
Output for a job's output file is being dropped:

  Job:            %s
  Rank:           %lu
  Held:           %lu chunks

The PMIx server is running as root, so it writes this job's output
files from a thread of their own rather than from the thread that
serves every job on the node. That thread has fallen this far behind
the job's output - a filesystem that is slow, or that does not respond -
and output beyond it is not written to the file. Output already held is
still written if the filesystem recovers.

The amount held is set by the iof_writer_backlog MCA parameter.
#
[group-id-in-use]
The PMIx server %s:%u was told that a group formed under an ID one of
its clients' groups still holds:

  Group ID:            %s
  Members it holds:    %lu
  Members reported:    %lu

A group ID belongs to one group until that group is destructed, so the
host should have refused the new one. This server keeps the group it
holds; its clients in the new group are told the construct failed
(PMIX_ERR_EXISTS) where it reports to them.
#
[register-nspace-no-job-size]
The host asked this PMIx server to register namespace %s without
giving the job's size (PMIX_JOB_SIZE). The size is required whenever a
namespace's job data is registered - at the top level of the info array
or inside its PMIX_JOB_INFO_ARRAY - and the registration was refused
(PMIX_ERR_BAD_PARAM). A registration with PMIX_REGISTER_NODATA carries
no job data and is not affected; the update that later delivers that
namespace's job data must include the size.
