Most MPI users will probably not need to use the
bfctl
command; see mpimsg(1) and mpitask(1).
If no process selection is given on the command line, one line
of status will be printed for every process on each specified node.
In default mode, information is organized under the following headings:
NODE
the nodeid on which the process is running
INDEX
PID
the process identifier from the underlying operating system
KPRI
KSTATE
the LAM kernel execution status:
R
running - unencumbered by LAM
BR
blocked receiving - The blocked message event and type are also
displayed.
BS
blocked sending - The blocked message event and type are also
displayed.
F
PROGRAM
With the
-l
option,
state
prints signal information instead of execution status.
SIGS
all pending but undelivered LAM signals
Some event/type combinations are used by the system to implement
client/server requests or to implement other message passing libraries.
Where possible,
state
replaces these numbers with the name of the function that caused
the current execution state.
In general, the information provided by
state
is low-level and geared toward LAM system programmers.
MPI programmers are expected to use mpitask(1) to
monitor the state of MPI processes.
A process must call kenter(2) or be created by loadgo(1)
in order to be directly visible to
state .
In other cases, limited monitoring is possible in one of two ways.
If a process identifier is given, the kernel
on each specified node will be consulted for kernel information (kstate(2)).
On the local node (use node specifier h) a process can
be watched without the assistance of the process management daemon
by giving the process identifier and the -k option.