November 2021 - discuss - mailman3.cels.anl.gov

Re: [mpich-discuss] Why stuck in MPI_Finalize?
by Zhou, Hui 30 Nov '21

30 Nov '21

There is no setting to peek inside `MPI_Fianlize`. You can use debugger to check. -- Hui Zhou From: Erik Schnetter <schnetter(a)gmail.com> Date: Tuesday, November 30, 2021 at 9:54 AM To: Zhou, Hui <zhouh(a)anl.gov> Cc: discuss(a)mpich.org <discuss(a)mpich.org> Subject: Re: [mpich-discuss] Why stuck in MPI_Finalize? Is there a way to see what is happening during MPI_Finalize, maybe some kind of `verbose` setting?

1 0

Slides from SC21 MPICH BoF posted
by Raffenetti, Ken 30 Nov '21

30 Nov '21

Hi all, Slides from the SC21 MPICH BoF have been posted to the MPICH Github wiki. https://github.com/pmodels/mpich/wiki/2021-Supercomputing-BoF#presentations Additionally, the session recording is available to SC21 attendees via the Hubb platform for a limited time. https://sc21.hubb.me/fe/schedule-builder/sessions/879718 Once again, thanks to all the presenters and the MPICH community for participating! Cheers, The MPICH team

1 0

Re: [mpich-discuss] Why stuck in MPI_Finalize?
by Erik Schnetter 30 Nov '21

30 Nov '21

Hui A garbage collector can't reliably reclaim resources. When MPI_Finalize is called by the garbage collector, then there's a chance it won't be run (e.g. if things live in global variables), and then the application won't shut down completely. Therefore, MPI_Finalize is called explicitly, while e.g. MPI_Comm_free usually would be called automatically. Is there a way to see what is happening during MPI_Finalize, maybe some kind of `verbose` setting? -erik On Mon, Nov 29, 2021 at 11:18 PM Zhou, Hui <zhouh(a)anl.gov> wrote: > > Hi Erik, > > Shouldn't MPI_Finalize run by the garbage collector as well by treating the entire MPI as an object? > > -- > Hui Zhou > ________________________________ > From: Erik Schnetter <schnetter(a)gmail.com> > Sent: Monday, November 29, 2021 7:23 PM > To: Zhou, Hui <zhouh(a)anl.gov> > Cc: discuss(a)mpich.org <discuss(a)mpich.org> > Subject: Re: [mpich-discuss] Why stuck in MPI_Finalize? > > Hui > > Julia is garbage collected, and our MPI wrappers free MPI handles > automatically when the wrapped handles become garbage collected. I > thus don't have direct control over when and which handles are freed > before MPI_Finalize is called at the end of the test case. Our > approach is to let things be as they are, ensuring that no MPI calls > are made any more when MPI_Finalized returns true. That means that > different MPI processes might (or might not ) free handles before > calling MPI_Finalize. I notice that this seems to work fine on Linux, > but not always on macOS. I don't have a simple reproducer in C. > > Pointers to setting up Julia: See e.g. > <https://github.com/JuliaParallel/MPI.jl/blob/master/.github/workflows/UnitT…> > which runs the unit tests for the Julia package `MPI.jl` which wraps > MPI implementations. There are several stanzas; the one called > `test-system-brew` runs MPI.jl on a system-provided (external MPI) > installation that has been installed via Homebrew on macOS. You would > replace the step "Install MPI via homebrew" by a step that downloads > and installs MPICH into /usr/local. I'll be happy to discuss details, > possibly over Zoom. > > -erik > > > On Mon, Nov 29, 2021 at 5:12 PM Zhou, Hui <zhouh(a)anl.gov> wrote: > > > > Hi Erik, > > > > It is not just a barrier in MPI_Finalize. It also needs to make sure pending communications are completed. Sometimes this is the user's fault, such as mismatched communication or partially freed communicators. But sometimes this is due to lower-level network library. For example, some libfabric provider does not always flush send. We tried to detect and prevent all cases that result in hanging, but there always seem to be cases that escape our solution. If you can drill down some simple reproducible (even with 5% chance) cases, please create a github issue and we'll track it down. > > > > By the way, if you have pointers on setting up Julia CI testing, we can try to setup nightly testing on our end as well. Catching the errors when it appear always makes troubleshooting easier. > > > > -- > > Hui > > ________________________________ > > From: Erik Schnetter via discuss <discuss(a)mpich.org> > > Sent: Monday, November 29, 2021 12:30 PM > > To: discuss(a)mpich.org <discuss(a)mpich.org> > > Cc: Erik Schnetter <schnetter(a)gmail.com> > > Subject: [mpich-discuss] Why stuck in MPI_Finalize? > > > > I have a Julia test case on macOS where MPICH randomly gets stuck in > > MPI_Finalize (with about a 5% chance). See e.g. > > https://github.com/JuliaParallel/MPI.jl/runs/4357341818 > > > > Can you advise under what circumstances MPICH could get stuck there? > > The respective run uses 3 processes, and all 3 processes call into > > MPI_Finalize, but no process returns. > > > > I assume that MPI_Finalize contains internally the equivalent to an > > MPI_Barrier, but that should succeed here. Are there other actions > > taken in MPI_Finalize that would require some kind of consistent state > > across the application? For example, if a communicator was created on > > all processes, but freed only on some processes, could this cause such > > a deadlock? > > > > -erik > > > > -- > > Erik Schnetter <schnetter(a)gmail.com> > > http://www.perimeterinstitute.ca/personal/eschnetter/ > > _______________________________________________ > > discuss mailing list discuss(a)mpich.org > > To manage subscription options or unsubscribe: > > https://lists.mpich.org/mailman/listinfo/discuss > > > > -- > Erik Schnetter <schnetter(a)gmail.com> > http://www.perimeterinstitute.ca/personal/eschnetter/ -- Erik Schnetter <schnetter(a)gmail.com> http://www.perimeterinstitute.ca/personal/eschnetter/

1 0

Re: [mpich-discuss] Why stuck in MPI_Finalize?
by Zhou, Hui 29 Nov '21

29 Nov '21

Hi Erik, Shouldn't MPI_Finalize run by the garbage collector as well by treating the entire MPI as an object? -- Hui Zhou ________________________________ From: Erik Schnetter <schnetter(a)gmail.com> Sent: Monday, November 29, 2021 7:23 PM To: Zhou, Hui <zhouh(a)anl.gov> Cc: discuss(a)mpich.org <discuss(a)mpich.org> Subject: Re: [mpich-discuss] Why stuck in MPI_Finalize? Hui Julia is garbage collected, and our MPI wrappers free MPI handles automatically when the wrapped handles become garbage collected. I thus don't have direct control over when and which handles are freed before MPI_Finalize is called at the end of the test case. Our approach is to let things be as they are, ensuring that no MPI calls are made any more when MPI_Finalized returns true. That means that different MPI processes might (or might not ) free handles before calling MPI_Finalize. I notice that this seems to work fine on Linux, but not always on macOS. I don't have a simple reproducer in C. Pointers to setting up Julia: See e.g. <https://github.com/JuliaParallel/MPI.jl/blob/master/.github/workflows/UnitT…> which runs the unit tests for the Julia package `MPI.jl` which wraps MPI implementations. There are several stanzas; the one called `test-system-brew` runs MPI.jl on a system-provided (external MPI) installation that has been installed via Homebrew on macOS. You would replace the step "Install MPI via homebrew" by a step that downloads and installs MPICH into /usr/local. I'll be happy to discuss details, possibly over Zoom. -erik On Mon, Nov 29, 2021 at 5:12 PM Zhou, Hui <zhouh(a)anl.gov> wrote: > > Hi Erik, > > It is not just a barrier in MPI_Finalize. It also needs to make sure pending communications are completed. Sometimes this is the user's fault, such as mismatched communication or partially freed communicators. But sometimes this is due to lower-level network library. For example, some libfabric provider does not always flush send. We tried to detect and prevent all cases that result in hanging, but there always seem to be cases that escape our solution. If you can drill down some simple reproducible (even with 5% chance) cases, please create a github issue and we'll track it down. > > By the way, if you have pointers on setting up Julia CI testing, we can try to setup nightly testing on our end as well. Catching the errors when it appear always makes troubleshooting easier. > > -- > Hui > ________________________________ > From: Erik Schnetter via discuss <discuss(a)mpich.org> > Sent: Monday, November 29, 2021 12:30 PM > To: discuss(a)mpich.org <discuss(a)mpich.org> > Cc: Erik Schnetter <schnetter(a)gmail.com> > Subject: [mpich-discuss] Why stuck in MPI_Finalize? > > I have a Julia test case on macOS where MPICH randomly gets stuck in > MPI_Finalize (with about a 5% chance). See e.g. > https://github.com/JuliaParallel/MPI.jl/runs/4357341818 > > Can you advise under what circumstances MPICH could get stuck there? > The respective run uses 3 processes, and all 3 processes call into > MPI_Finalize, but no process returns. > > I assume that MPI_Finalize contains internally the equivalent to an > MPI_Barrier, but that should succeed here. Are there other actions > taken in MPI_Finalize that would require some kind of consistent state > across the application? For example, if a communicator was created on > all processes, but freed only on some processes, could this cause such > a deadlock? > > -erik > > -- > Erik Schnetter <schnetter(a)gmail.com> > http://www.perimeterinstitute.ca/personal/eschnetter/ > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss -- Erik Schnetter <schnetter(a)gmail.com> http://www.perimeterinstitute.ca/personal/eschnetter/

1 0

Re: [mpich-discuss] Why stuck in MPI_Finalize?
by Erik Schnetter 29 Nov '21

29 Nov '21

Hui Julia is garbage collected, and our MPI wrappers free MPI handles automatically when the wrapped handles become garbage collected. I thus don't have direct control over when and which handles are freed before MPI_Finalize is called at the end of the test case. Our approach is to let things be as they are, ensuring that no MPI calls are made any more when MPI_Finalized returns true. That means that different MPI processes might (or might not ) free handles before calling MPI_Finalize. I notice that this seems to work fine on Linux, but not always on macOS. I don't have a simple reproducer in C. Pointers to setting up Julia: See e.g. <https://github.com/JuliaParallel/MPI.jl/blob/master/.github/workflows/UnitT…> which runs the unit tests for the Julia package `MPI.jl` which wraps MPI implementations. There are several stanzas; the one called `test-system-brew` runs MPI.jl on a system-provided (external MPI) installation that has been installed via Homebrew on macOS. You would replace the step "Install MPI via homebrew" by a step that downloads and installs MPICH into /usr/local. I'll be happy to discuss details, possibly over Zoom. -erik On Mon, Nov 29, 2021 at 5:12 PM Zhou, Hui <zhouh(a)anl.gov> wrote: > > Hi Erik, > > It is not just a barrier in MPI_Finalize. It also needs to make sure pending communications are completed. Sometimes this is the user's fault, such as mismatched communication or partially freed communicators. But sometimes this is due to lower-level network library. For example, some libfabric provider does not always flush send. We tried to detect and prevent all cases that result in hanging, but there always seem to be cases that escape our solution. If you can drill down some simple reproducible (even with 5% chance) cases, please create a github issue and we'll track it down. > > By the way, if you have pointers on setting up Julia CI testing, we can try to setup nightly testing on our end as well. Catching the errors when it appear always makes troubleshooting easier. > > -- > Hui > ________________________________ > From: Erik Schnetter via discuss <discuss(a)mpich.org> > Sent: Monday, November 29, 2021 12:30 PM > To: discuss(a)mpich.org <discuss(a)mpich.org> > Cc: Erik Schnetter <schnetter(a)gmail.com> > Subject: [mpich-discuss] Why stuck in MPI_Finalize? > > I have a Julia test case on macOS where MPICH randomly gets stuck in > MPI_Finalize (with about a 5% chance). See e.g. > https://github.com/JuliaParallel/MPI.jl/runs/4357341818 > > Can you advise under what circumstances MPICH could get stuck there? > The respective run uses 3 processes, and all 3 processes call into > MPI_Finalize, but no process returns. > > I assume that MPI_Finalize contains internally the equivalent to an > MPI_Barrier, but that should succeed here. Are there other actions > taken in MPI_Finalize that would require some kind of consistent state > across the application? For example, if a communicator was created on > all processes, but freed only on some processes, could this cause such > a deadlock? > > -erik > > -- > Erik Schnetter <schnetter(a)gmail.com> > http://www.perimeterinstitute.ca/personal/eschnetter/ > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss -- Erik Schnetter <schnetter(a)gmail.com> http://www.perimeterinstitute.ca/personal/eschnetter/

1 0

Re: [mpich-discuss] Why stuck in MPI_Finalize?
by Zhou, Hui 29 Nov '21

29 Nov '21

Hi Erik, It is not just a barrier in MPI_Finalize. It also needs to make sure pending communications are completed. Sometimes this is the user's fault, such as mismatched communication or partially freed communicators. But sometimes this is due to lower-level network library. For example, some libfabric provider does not always flush send. We tried to detect and prevent all cases that result in hanging, but there always seem to be cases that escape our solution. If you can drill down some simple reproducible (even with 5% chance) cases, please create a github issue and we'll track it down. By the way, if you have pointers on setting up Julia CI testing, we can try to setup nightly testing on our end as well. Catching the errors when it appear always makes troubleshooting easier. -- Hui ________________________________ From: Erik Schnetter via discuss <discuss(a)mpich.org> Sent: Monday, November 29, 2021 12:30 PM To: discuss(a)mpich.org <discuss(a)mpich.org> Cc: Erik Schnetter <schnetter(a)gmail.com> Subject: [mpich-discuss] Why stuck in MPI_Finalize? I have a Julia test case on macOS where MPICH randomly gets stuck in MPI_Finalize (with about a 5% chance). See e.g. https://github.com/JuliaParallel/MPI.jl/runs/4357341818 Can you advise under what circumstances MPICH could get stuck there? The respective run uses 3 processes, and all 3 processes call into MPI_Finalize, but no process returns. I assume that MPI_Finalize contains internally the equivalent to an MPI_Barrier, but that should succeed here. Are there other actions taken in MPI_Finalize that would require some kind of consistent state across the application? For example, if a communicator was created on all processes, but freed only on some processes, could this cause such a deadlock? -erik -- Erik Schnetter <schnetter(a)gmail.com> http://www.perimeterinstitute.ca/personal/eschnetter/ _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

Why stuck in MPI_Finalize?
by Erik Schnetter 29 Nov '21

29 Nov '21

I have a Julia test case on macOS where MPICH randomly gets stuck in MPI_Finalize (with about a 5% chance). See e.g. https://github.com/JuliaParallel/MPI.jl/runs/4357341818 Can you advise under what circumstances MPICH could get stuck there? The respective run uses 3 processes, and all 3 processes call into MPI_Finalize, but no process returns. I assume that MPI_Finalize contains internally the equivalent to an MPI_Barrier, but that should succeed here. Are there other actions taken in MPI_Finalize that would require some kind of consistent state across the application? For example, if a communicator was created on all processes, but freed only on some processes, could this cause such a deadlock? -erik -- Erik Schnetter <schnetter(a)gmail.com> http://www.perimeterinstitute.ca/personal/eschnetter/

1 0

Re: [mpich-discuss] MPI_Comm_spawn and SLURM
by Mccall, Kurt E. (MSFC-EV41) 27 Nov '21

27 Nov '21

Hui, Looks like the child process failed to launch, so it may be a process manager issue. I ran "scontrol show hosts" and it showed a number of possible errors in the compute node SLURM configuration (like "CoresPerSocket=1", which is wrong). Looks like our compute nodes are not configured correctly for SLURM. Thanks, Kurt From: Zhou, Hui <zhouh(a)anl.gov> Sent: Friday, November 26, 2021 10:13 PM To: discuss(a)mpich.org Cc: Mccall, Kurt E. (MSFC-EV41) <kurt.e.mccall(a)nasa.gov> Subject: [EXTERNAL] Re: MPI_Comm_spawn and SLURM Kurt, Could you verify whether the children process is launched? The way `MPI_Comm_spawn_multiple` works is - 1. Parent root open a port 2. Parent root call PMI routine to spawn children processes - here the process manager is SLURM. If the children processes failed to spawn, then it is likely the process manager issue. 3. Children processes run MPI_Init, and at the end internally call MPI_Comm_connect 4. Meanwhile, the parent root process wait in `MPI_Comm_accept` If root process is hanging in `MPI_Comm_accept`, then either the process manager failed to launch children processes, or children processes are misconfigured by process manager and didn't call `MPI_Comm_connect`. Let's determine where children are at. -- Hui Zhou From: Mccall, Kurt E. (MSFC-EV41) via discuss <discuss(a)mpich.org<mailto:[email protected]>> Date: Friday, November 26, 2021 at 5:05 PM To: discuss(a)mpich.org<mailto:[email protected]> <discuss(a)mpich.org<mailto:[email protected]>> Cc: Mccall, Kurt E. (MSFC-EV41) <kurt.e.mccall(a)nasa.gov<mailto:[email protected]>> Subject: [mpich-discuss] MPI_Comm_spawn and SLURM I am attempting to run MPICH under SLURM for the first time, and there could be a lot of things I am doing wrong. All processes are getting launched but the master process is freezing in MPI_Comm_spawn. The stack trace is below, followed by the SLURM command I use to start the job. Even though all 20 processes are running, one per node as desired, salloc reports that "Requested node configuration is not available". Not sure if that is why MPI_Comm_spawn is frozen. Thanks for any help. #0 0x00007f0183d08a08 in poll () from /lib64/libc.so.6 #1 0x00007f0185aed611 in MPID_nem_tcp_connpoll (in_blocking_poll=<optimized out>) at ../mpich-4.0b1/src/mpid/ch3/channels/nemesis/netmod/tcp/socksm.c:1765 #2 0x00007f0185ad8186 in MPID_nem_mpich_blocking_recv (completions=<optimized out>, in_fbox=<synthetic pointer>, cell=<synthetic pointer>) at ../mpich-4.0b1/src/mpid/ch3/channels/nemesis/include/mpid_nem_inline.h:947 #3 MPIDI_CH3I_Progress (progress_state=progress_state@entry=0x7ffcc4839760, is_blocking=is_blocking@entry=1) at ../mpich-4.0b1/src/mpid/ch3/channels/nemesis/src/ch3_progress.c:360 #4 0x00007f0185a8811a in MPIDI_Create_inter_root_communicator_accept (vc_pptr=<synthetic pointer>, comm_pptr=<synthetic pointer>, port_name=<optimized out>) at ../mpich-4.0b1/src/mpid/ch3/src/ch3u_port.c:417 #5 MPIDI_Comm_accept (port_name=<optimized out>, info=<optimized out>, root=0, comm_ptr=0x7f0185ef1540 <MPIR_Comm_builtin+832>, newcomm=0x7ffcc4839bb8) at ../mpich-4.0b1/src/mpid/ch3/src/ch3u_port.c:1176 #6 0x00007f0185ac3d45 in MPID_Comm_accept ( port_name=port_name@entry=0x7ffcc4839a00 "tag#0$description#n001$port#55907$ifname#172.16.56.1$", info=info@entry=0x0, root=root@entry=0, comm=comm@entry=0x7f0185ef1540 <MPIR_Comm_builtin+832>, newcomm_ptr=newcomm_ptr@entry=0x7ffcc4839bb8) at ../mpich-4.0b1/src/mpid/ch3/src/mpid_port.c:130 #7 0x00007f0185a73285 in MPIDI_Comm_spawn_multiple (count=<optimized out>, commands=0x7ffcc4839b78, argvs=0x7ffcc4839b70, maxprocs=0x7ffcc4839b6c, info_ptrs=<optimized out>, root=<optimized out>, comm_ptr=0x7f0185ef1540 <MPIR_Comm_builtin+832>, intercomm=0x7ffcc4839bb8, errcodes=<optimized out>) at ../mpich-4.0b1/src/mpid/ch3/src/ch3u_comm_spawn_multiple.c:258 #8 0x00007f0185abec99 in MPID_Comm_spawn_multiple (count=count@entry=1, array_of_commands=array_of_commands@entry=0x7ffcc4839b78, array_of_argv=array_of_argv@entry=0x7ffcc4839b70, array_of_maxprocs=array_of_maxprocs@entry=0x7ffcc4839b6c, array_of_info_ptrs=array_of_info_ptrs@entry=0x7ffcc4839b60, root=root@entry=0, comm_ptr=0x7f0185ef1540 <MPIR_Comm_builtin+832>, intercomm=0x7ffcc4839bb8, array_of_errcodes=0x7ffcc4839c98) at ../mpich-4.0b1/src/mpid/ch3/src/mpid_comm_spawn_multiple.c:49 #9 0x00007f0185a34895 in MPIR_Comm_spawn_impl (command=<optimized out>, command@entry=0x995428 "NeedlesMpiMM", argv=<optimized out>, argv@entry=0x993f50, maxprocs=<optimized out>, maxprocs@entry=1, info_ptr=<optimized out>, root=root@entry=0, comm_ptr=comm_ptr@entry=0x7f0185ef1540 <MPIR_Comm_builtin+832>, p_intercomm_ptr=0x7ffcc4839bb8, array_of_errcodes=0x7ffcc4839c98) at ../mpich-4.0b1/src/mpi/spawn/spawn_impl.c:168 #10 0x00007f0185953637 in internal_Comm_spawn (array_of_errcodes=0x7ffcc4839c98, intercomm=0x7ffcc4839d7c, comm=1140850689, root=0, info=-1677721600, maxprocs=1, argv=<optimized out>, command=0x995428 "NeedlesMpiMM") at ../mpich-4.0b1/src/binding/c/spawn/comm_spawn.c:83 #11 PMPI_Comm_spawn (command=0x995428 "NeedlesMpiMM", argv=0x993f50, maxprocs=1, info=-1677721600, root=0, comm=1140850689, intercomm=0x7ffcc4839d7c, array_of_errcodes=0x7ffcc4839c98) at ../mpich-4.0b1/src/binding/c/spawn/comm_spawn.c:169 #12 0x000000000040cec3 in needles::NeedlesMpiMaster::spawnNewManager (this=0x995380, nodenum=0, host_name="n001.cluster.pssclabs.com", intercom=@0x7ffcc4839d7c: 67108864) at src/NeedlesMpiMaster.cpp:1432 #13 0x00000000004084cb in needles::NeedlesMpiMaster::init (this=0x995380, argc=23, argv=0x7ffcc483a548, rank=0, world_size=20) at src/NeedlesMpiMaster.cpp:246 #14 0x0000000000406799 in main (argc=23, argv=0x7ffcc483a548) at src/NeedlesMpiManagerMain.cpp:96 Here is my salloc command to start the job. I want one task per node, reserving the rest of the cores on the node for spawning of additional processes. $ salloc -ntasks=20 --cpus-per-task=24 -verbose Here is what salloc reports: salloc: -------------------- -------------------- salloc: cpus-per-task : 24 salloc: ntasks : 20 salloc: verbose : 1 salloc: -------------------- -------------------- salloc: end of defined options salloc: Linear node selection plugin loaded with argument 4 salloc: select/cons_res loaded with argument 4 salloc: Cray/Aries node selection plugin loaded salloc: select/cons_tres loaded with argument 4 salloc: Granted job allocation 34311 srun: error: Unable to create step for job 34311: Requested node configuration is not available

1 0

Re: [mpich-discuss] MPI_Comm_spawn and SLURM
by Zhou, Hui 26 Nov '21

26 Nov '21

Kurt, Could you verify whether the children process is launched? The way `MPI_Comm_spawn_multiple` works is – 1. Parent root open a port 2. Parent root call PMI routine to spawn children processes – here the process manager is SLURM. If the children processes failed to spawn, then it is likely the process manager issue. 3. Children processes run MPI_Init, and at the end internally call MPI_Comm_connect 4. Meanwhile, the parent root process wait in `MPI_Comm_accept` If root process is hanging in `MPI_Comm_accept`, then either the process manager failed to launch children processes, or children processes are misconfigured by process manager and didn’t call `MPI_Comm_connect`. Let’s determine where children are at. -- Hui Zhou From: Mccall, Kurt E. (MSFC-EV41) via discuss <discuss(a)mpich.org> Date: Friday, November 26, 2021 at 5:05 PM To: discuss(a)mpich.org <discuss(a)mpich.org> Cc: Mccall, Kurt E. (MSFC-EV41) <kurt.e.mccall(a)nasa.gov> Subject: [mpich-discuss] MPI_Comm_spawn and SLURM I am attempting to run MPICH under SLURM for the first time, and there could be a lot of things I am doing wrong. All processes are getting launched but the master process is freezing in MPI_Comm_spawn. The stack trace is below, followed by the SLURM command I use to start the job. Even though all 20 processes are running, one per node as desired, salloc reports that “Requested node configuration is not available". Not sure if that is why MPI_Comm_spawn is frozen. Thanks for any help. #0 0x00007f0183d08a08 in poll () from /lib64/libc.so.6 #1 0x00007f0185aed611 in MPID_nem_tcp_connpoll (in_blocking_poll=<optimized out>) at ../mpich-4.0b1/src/mpid/ch3/channels/nemesis/netmod/tcp/socksm.c:1765 #2 0x00007f0185ad8186 in MPID_nem_mpich_blocking_recv (completions=<optimized out>, in_fbox=<synthetic pointer>, cell=<synthetic pointer>) at ../mpich-4.0b1/src/mpid/ch3/channels/nemesis/include/mpid_nem_inline.h:947 #3 MPIDI_CH3I_Progress (progress_state=progress_state@entry=0x7ffcc4839760, is_blocking=is_blocking@entry=1) at ../mpich-4.0b1/src/mpid/ch3/channels/nemesis/src/ch3_progress.c:360 #4 0x00007f0185a8811a in MPIDI_Create_inter_root_communicator_accept (vc_pptr=<synthetic pointer>, comm_pptr=<synthetic pointer>, port_name=<optimized out>) at ../mpich-4.0b1/src/mpid/ch3/src/ch3u_port.c:417 #5 MPIDI_Comm_accept (port_name=<optimized out>, info=<optimized out>, root=0, comm_ptr=0x7f0185ef1540 <MPIR_Comm_builtin+832>, newcomm=0x7ffcc4839bb8) at ../mpich-4.0b1/src/mpid/ch3/src/ch3u_port.c:1176 #6 0x00007f0185ac3d45 in MPID_Comm_accept ( port_name=port_name@entry=0x7ffcc4839a00 "tag#0$description#n001$port#55907$ifname#172.16.56.1$", info=info@entry=0x0, root=root@entry=0, comm=comm@entry=0x7f0185ef1540 <MPIR_Comm_builtin+832>, newcomm_ptr=newcomm_ptr@entry=0x7ffcc4839bb8) at ../mpich-4.0b1/src/mpid/ch3/src/mpid_port.c:130 #7 0x00007f0185a73285 in MPIDI_Comm_spawn_multiple (count=<optimized out>, commands=0x7ffcc4839b78, argvs=0x7ffcc4839b70, maxprocs=0x7ffcc4839b6c, info_ptrs=<optimized out>, root=<optimized out>, comm_ptr=0x7f0185ef1540 <MPIR_Comm_builtin+832>, intercomm=0x7ffcc4839bb8, errcodes=<optimized out>) at ../mpich-4.0b1/src/mpid/ch3/src/ch3u_comm_spawn_multiple.c:258 #8 0x00007f0185abec99 in MPID_Comm_spawn_multiple (count=count@entry=1, array_of_commands=array_of_commands@entry=0x7ffcc4839b78, array_of_argv=array_of_argv@entry=0x7ffcc4839b70, array_of_maxprocs=array_of_maxprocs@entry=0x7ffcc4839b6c, array_of_info_ptrs=array_of_info_ptrs@entry=0x7ffcc4839b60, root=root@entry=0, comm_ptr=0x7f0185ef1540 <MPIR_Comm_builtin+832>, intercomm=0x7ffcc4839bb8, array_of_errcodes=0x7ffcc4839c98) at ../mpich-4.0b1/src/mpid/ch3/src/mpid_comm_spawn_multiple.c:49 #9 0x00007f0185a34895 in MPIR_Comm_spawn_impl (command=<optimized out>, command@entry=0x995428 "NeedlesMpiMM", argv=<optimized out>, argv@entry=0x993f50, maxprocs=<optimized out>, maxprocs@entry=1, info_ptr=<optimized out>, root=root@entry=0, comm_ptr=comm_ptr@entry=0x7f0185ef1540 <MPIR_Comm_builtin+832>, p_intercomm_ptr=0x7ffcc4839bb8, array_of_errcodes=0x7ffcc4839c98) at ../mpich-4.0b1/src/mpi/spawn/spawn_impl.c:168 #10 0x00007f0185953637 in internal_Comm_spawn (array_of_errcodes=0x7ffcc4839c98, intercomm=0x7ffcc4839d7c, comm=1140850689, root=0, info=-1677721600, maxprocs=1, argv=<optimized out>, command=0x995428 "NeedlesMpiMM") at ../mpich-4.0b1/src/binding/c/spawn/comm_spawn.c:83 #11 PMPI_Comm_spawn (command=0x995428 "NeedlesMpiMM", argv=0x993f50, maxprocs=1, info=-1677721600, root=0, comm=1140850689, intercomm=0x7ffcc4839d7c, array_of_errcodes=0x7ffcc4839c98) at ../mpich-4.0b1/src/binding/c/spawn/comm_spawn.c:169 #12 0x000000000040cec3 in needles::NeedlesMpiMaster::spawnNewManager (this=0x995380, nodenum=0, host_name="n001.cluster.pssclabs.com", intercom=@0x7ffcc4839d7c: 67108864) at src/NeedlesMpiMaster.cpp:1432 #13 0x00000000004084cb in needles::NeedlesMpiMaster::init (this=0x995380, argc=23, argv=0x7ffcc483a548, rank=0, world_size=20) at src/NeedlesMpiMaster.cpp:246 #14 0x0000000000406799 in main (argc=23, argv=0x7ffcc483a548) at src/NeedlesMpiManagerMain.cpp:96 Here is my salloc command to start the job. I want one task per node, reserving the rest of the cores on the node for spawning of additional processes. $ salloc –ntasks=20 --cpus-per-task=24 –verbose Here is what salloc reports: salloc: -------------------- -------------------- salloc: cpus-per-task : 24 salloc: ntasks : 20 salloc: verbose : 1 salloc: -------------------- -------------------- salloc: end of defined options salloc: Linear node selection plugin loaded with argument 4 salloc: select/cons_res loaded with argument 4 salloc: Cray/Aries node selection plugin loaded salloc: select/cons_tres loaded with argument 4 salloc: Granted job allocation 34311 srun: error: Unable to create step for job 34311: Requested node configuration is not available

1 0

MPI_Comm_spawn and SLURM
by Mccall, Kurt E. (MSFC-EV41) 26 Nov '21

26 Nov '21

I am attempting to run MPICH under SLURM for the first time, and there could be a lot of things I am doing wrong. All processes are getting launched but the master process is freezing in MPI_Comm_spawn. The stack trace is below, followed by the SLURM command I use to start the job. Even though all 20 processes are running, one per node as desired, salloc reports that "Requested node configuration is not available". Not sure if that is why MPI_Comm_spawn is frozen. Thanks for any help. #0 0x00007f0183d08a08 in poll () from /lib64/libc.so.6 #1 0x00007f0185aed611 in MPID_nem_tcp_connpoll (in_blocking_poll=<optimized out>) at ../mpich-4.0b1/src/mpid/ch3/channels/nemesis/netmod/tcp/socksm.c:1765 #2 0x00007f0185ad8186 in MPID_nem_mpich_blocking_recv (completions=<optimized out>, in_fbox=<synthetic pointer>, cell=<synthetic pointer>) at ../mpich-4.0b1/src/mpid/ch3/channels/nemesis/include/mpid_nem_inline.h:947 #3 MPIDI_CH3I_Progress (progress_state=progress_state@entry=0x7ffcc4839760, is_blocking=is_blocking@entry=1) at ../mpich-4.0b1/src/mpid/ch3/channels/nemesis/src/ch3_progress.c:360 #4 0x00007f0185a8811a in MPIDI_Create_inter_root_communicator_accept (vc_pptr=<synthetic pointer>, comm_pptr=<synthetic pointer>, port_name=<optimized out>) at ../mpich-4.0b1/src/mpid/ch3/src/ch3u_port.c:417 #5 MPIDI_Comm_accept (port_name=<optimized out>, info=<optimized out>, root=0, comm_ptr=0x7f0185ef1540 <MPIR_Comm_builtin+832>, newcomm=0x7ffcc4839bb8) at ../mpich-4.0b1/src/mpid/ch3/src/ch3u_port.c:1176 #6 0x00007f0185ac3d45 in MPID_Comm_accept ( port_name=port_name@entry=0x7ffcc4839a00 "tag#0$description#n001$port#55907$ifname#172.16.56.1$", info=info@entry=0x0, root=root@entry=0, comm=comm@entry=0x7f0185ef1540 <MPIR_Comm_builtin+832>, newcomm_ptr=newcomm_ptr@entry=0x7ffcc4839bb8) at ../mpich-4.0b1/src/mpid/ch3/src/mpid_port.c:130 #7 0x00007f0185a73285 in MPIDI_Comm_spawn_multiple (count=<optimized out>, commands=0x7ffcc4839b78, argvs=0x7ffcc4839b70, maxprocs=0x7ffcc4839b6c, info_ptrs=<optimized out>, root=<optimized out>, comm_ptr=0x7f0185ef1540 <MPIR_Comm_builtin+832>, intercomm=0x7ffcc4839bb8, errcodes=<optimized out>) at ../mpich-4.0b1/src/mpid/ch3/src/ch3u_comm_spawn_multiple.c:258 #8 0x00007f0185abec99 in MPID_Comm_spawn_multiple (count=count@entry=1, array_of_commands=array_of_commands@entry=0x7ffcc4839b78, array_of_argv=array_of_argv@entry=0x7ffcc4839b70, array_of_maxprocs=array_of_maxprocs@entry=0x7ffcc4839b6c, array_of_info_ptrs=array_of_info_ptrs@entry=0x7ffcc4839b60, root=root@entry=0, comm_ptr=0x7f0185ef1540 <MPIR_Comm_builtin+832>, intercomm=0x7ffcc4839bb8, array_of_errcodes=0x7ffcc4839c98) at ../mpich-4.0b1/src/mpid/ch3/src/mpid_comm_spawn_multiple.c:49 #9 0x00007f0185a34895 in MPIR_Comm_spawn_impl (command=<optimized out>, command@entry=0x995428 "NeedlesMpiMM", argv=<optimized out>, argv@entry=0x993f50, maxprocs=<optimized out>, maxprocs@entry=1, info_ptr=<optimized out>, root=root@entry=0, comm_ptr=comm_ptr@entry=0x7f0185ef1540 <MPIR_Comm_builtin+832>, p_intercomm_ptr=0x7ffcc4839bb8, array_of_errcodes=0x7ffcc4839c98) at ../mpich-4.0b1/src/mpi/spawn/spawn_impl.c:168 #10 0x00007f0185953637 in internal_Comm_spawn (array_of_errcodes=0x7ffcc4839c98, intercomm=0x7ffcc4839d7c, comm=1140850689, root=0, info=-1677721600, maxprocs=1, argv=<optimized out>, command=0x995428 "NeedlesMpiMM") at ../mpich-4.0b1/src/binding/c/spawn/comm_spawn.c:83 #11 PMPI_Comm_spawn (command=0x995428 "NeedlesMpiMM", argv=0x993f50, maxprocs=1, info=-1677721600, root=0, comm=1140850689, intercomm=0x7ffcc4839d7c, array_of_errcodes=0x7ffcc4839c98) at ../mpich-4.0b1/src/binding/c/spawn/comm_spawn.c:169 #12 0x000000000040cec3 in needles::NeedlesMpiMaster::spawnNewManager (this=0x995380, nodenum=0, host_name="n001.cluster.pssclabs.com", intercom=@0x7ffcc4839d7c: 67108864) at src/NeedlesMpiMaster.cpp:1432 #13 0x00000000004084cb in needles::NeedlesMpiMaster::init (this=0x995380, argc=23, argv=0x7ffcc483a548, rank=0, world_size=20) at src/NeedlesMpiMaster.cpp:246 #14 0x0000000000406799 in main (argc=23, argv=0x7ffcc483a548) at src/NeedlesMpiManagerMain.cpp:96 Here is my salloc command to start the job. I want one task per node, reserving the rest of the cores on the node for spawning of additional processes. $ salloc -ntasks=20 --cpus-per-task=24 -verbose Here is what salloc reports: salloc: -------------------- -------------------- salloc: cpus-per-task : 24 salloc: ntasks : 20 salloc: verbose : 1 salloc: -------------------- -------------------- salloc: end of defined options salloc: Linear node selection plugin loaded with argument 4 salloc: select/cons_res loaded with argument 4 salloc: Cray/Aries node selection plugin loaded salloc: select/cons_tres loaded with argument 4 salloc: Granted job allocation 34311 srun: error: Unable to create step for job 34311: Requested node configuration is not available

1 0