[CP2K-user] [CP2K:19487] Memory Leak on CP2k 9.1

Krack Matthias matthias.krack at psi.ch
Thu Nov 9 16:04:14 UTC 2023


Dear Quentin

This can’t be accomplished just by a change of the docker file. The configure command for the plumed build in the CP2K toolchain has to be changed too. I just applied such a change in my personal fork of the current CP2K master version (see https://github.com/mkrack/cp2k):
https://github.com/cp2k/cp2k/commit/cbe099e098ced65b0aeda353db8eb4c5e391f27f
You can do the same in your personal fork of CP2K and then use the URL of that github repository in one of the master_* docker files for the “git clone” command.

HTH

Matthias

From: cp2k at googlegroups.com <cp2k at googlegroups.com> on behalf of Quentin Pessemesse <q.pessemesse at gmail.com>
Date: Thursday, 9 November 2023 at 16:19
To: cp2k at googlegroups.com <cp2k at googlegroups.com>
Subject: Re: [CP2K:19486] Memory Leak on CP2k 9.1

Dear Matthias,

Do you know if there is a easy way to use plumed modules beyond the default ones with these docker images ? Some modules I need are deactivated by default. I have no experience in editing docker images, and this may rather be a question for the plumed forum...

Thank you again,

Q.
On 09/11/2023 15:57, Krack Matthias wrote:
Dear Quentin

I am glad to read that you could solve the problems by using the provided container images.

For future reference, I would like the mention that the README and the docker files for CP2K production containers have been moved to a separate repository on GitHub called cp2k-containers<https://github.com/cp2k/cp2k-containers>, recently. Therefore, the links in my previous mail are no longer valid. The CP2K production containers can be download from DockerHub here<https://hub.docker.com/r/cp2k/cp2k/tags/>.

Best

Matthias

From: cp2k at googlegroups.com<mailto:cp2k at googlegroups.com> <cp2k at googlegroups.com><mailto:cp2k at googlegroups.com> on behalf of Quentin Pessemesse <q.pessemesse at gmail.com><mailto:q.pessemesse at gmail.com>
Date: Thursday, 9 November 2023 at 15:33
To: cp2k <cp2k at googlegroups.com><mailto:cp2k at googlegroups.com>
Subject: Re: [CP2K:19485] Memory Leak on CP2k 9.1
Dear Matthias,
Thank you very much for your help, we were able to solve the memory leak issue by using the mpi that is compiled with the image, with some modifications to the command you provided to account for specificities of the cluster.
Best,
Quentin

Le vendredi 6 octobre 2023 à 15:16:09 UTC+2, Krack Matthias a écrit :
Dear Quentin

These containers are built differently following the usual cp2k toolchain installation process. There is no /opt/cp2k-toolchain/ folder but the folder /opt/cp2k/tools/toolchain/install/. There is no need, however, to source that setup file, because the entrypoint.sh<https://github.com/mkrack/cp2k/blob/master/tools/docker/production/Dockerfile.2023.2_mpich_generic_psmp#L83> script takes already care of that. You should be able to run the container as described in the READ.me<https://github.com/mkrack/cp2k/blob/master/tools/docker/production/README.md>.

Best

Matthias

From: cp... at googlegroups.com<mailto:cp... at googlegroups.com> <cp... at googlegroups.com<mailto:cp... at googlegroups.com>> on behalf of Quentin Pessemesse <q.pess... at gmail.com<mailto:q.pess... at gmail.com>>
Date: Friday, 6 October 2023 at 14:59
To: cp2k <cp... at googlegroups.com<mailto:cp... at googlegroups.com>>
Subject: Re: [CP2K:19312] Memory Leak on CP2k 9.1
Dear Matthias,
Thank you kindly for your advice, I will try these different versions as soon as possible.
I've built the docker image for an openmpi version of CP2k on the cluster. With version 2023.1, I used to source the environment variables using " source /opt/cp2k-toolchain/install/setup". This does not work anymore. Is it a problem on the image's end or on the cluster end ?
Best,
Quentin


Le vendredi 6 octobre 2023 à 11:27:26 UTC+2, Krack Matthias a écrit :
Hi Quentin

There are some more cp2k 2023.2 docker containers for production<https://github.com/mkrack/cp2k/tree/master/tools/docker/production> available (build with MPICH or OpenMPI) which can also be pulled with apptainer (see READ.me<https://github.com/mkrack/cp2k/blob/master/tools/docker/production/README.md> for details). Maybe, you have more luck with one of these.

Best

Matthias

From: cp... at googlegroups.com<mailto:cp... at googlegroups.com> <cp... at googlegroups.com<mailto:cp... at googlegroups.com>> on behalf of Quentin Pessemesse <q.pess... at gmail.com<mailto:q.pess... at gmail.com>>
Date: Friday, 6 October 2023 at 10:47
To: cp2k <cp... at googlegroups.com<mailto:cp... at googlegroups.com>>
Subject: Re: [CP2K:19310] Memory Leak on CP2k 9.1
Dear all,
The cluster staff has moved to using a docker with a CP2k image, with CP2k 2023.1 (https://hub.docker.com/r/cp2k/cp2k/tags). The program experiences serious memory leaks (out-of-memory crash after less than 24 hours on AIMD with a system less than 100 atoms with 256 GB). The cluster cannot use intelmpi versions older than intelmpi 20. Is there a more recent version of CP2k which is stable and does not experience this type of large memory leaks?
We've tried to compile our own versions of CP2k with multiple versions of openMPI to no avail. The only stable CP2k version we have is CP2k 6.1, which is used with intelMPI 18 but it is on a legacy container where no new software can be installed.
Has anyone managed to use this docker image succesfully, and if so, which MPI package/version have you used ? If necessary, we can downgrade down to CP2k 9.1.
Best,
Quentin

Le mercredi 5 octobre 2022 à 13:19:26 UTC+2, Krack Matthias (PSI) a écrit :
Hi Quentin

It seems that you are using OpenMPI which is known to have leaks in some versions. Check this issue<https://github.com/cp2k/cp2k/issues/1830#issuecomment-1012561166> and this discussion<https://groups.google.com/g/cp2k/c/BJ9c21ey0Ls/m/2UDxnhBRAQAJ> here on this forum for further information.

HTH

Matthias

From: "cp... at googlegroups.com<mailto:cp... at googlegroups.com>" <cp... at googlegroups.com<mailto:cp... at googlegroups.com>> on behalf of Quentin Pessemesse <q.pess... at gmail.com<mailto:q.pess... at gmail.com>>
Reply to: "cp... at googlegroups.com<mailto:cp... at googlegroups.com>" <cp... at googlegroups.com<mailto:cp... at googlegroups.com>>
Date: Wednesday, 5 October 2022 at 12:39
To: "cp... at googlegroups.com<mailto:cp... at googlegroups.com>" <cp... at googlegroups.com<mailto:cp... at googlegroups.com>>
Subject: [CP2K:17807] Memory Leak on CP2k 9.1

Dear all,
Our group is encountering a memory leak issue that makes running DFT-MD impossible with large systems (~100 atoms) on one of the clusters we have access to, even though the same calculations run correctly on other machines.
The cluster support sent me the following valgrind output and asked me to find suggestions on how to proceed. Does anyone have input on how to deal with such memory leaks ?
Best,
Quentin P.

==62== Invalid write of size 4 ==62== at 0x1EA9887: grid_ref_create_task_list (in /ccc/products2/cp2k-9.1/Rhel_8__x86_64/gcc--8.3.0__openmpi--4.0.1/plumed/bin/cp2k.psmp) ==62== by 0x1E7A772: grid_create_task_list (in /ccc/products2/cp2k-9.1/Rhel_8__x86_64/gcc--8.3.0__openmpi--4.0.1/plumed/bin/cp2k.psmp) ==62== by 0x1E790B3: __grid_api_MOD_grid_create_task_list (grid_api.F:938) ==62== by 0x104AA67: __task_list_methods_MOD_generate_qs_task_list (task_list_methods.F:623) ==62== by 0xF58353: __qs_update_s_mstruct_MOD_qs_env_update_s_mstruct (qs_update_s_mstruct.F:187) ==62== by 0xCC03AB: __qs_energy_init_MOD_qs_energies_init (qs_energy_init.F:311) ==62== by 0xCBF0A1: __qs_energy_MOD_qs_energies (qs_energy.F:84) ==62== by 0xCE087E: __qs_force_MOD_qs_forces (qs_force.F:212) ==62== by 0xCE4349: __qs_force_MOD_qs_calc_energy_force (qs_force.F:117) ==62== by 0x9AE2C0: __force_env_methods_MOD_force_env_calc_energy_force (force_env_methods.F:271) ==62== by 0x50CD0C: __md_run_MOD_qs_mol_dyn_low (md_run.F:372) ==62== by 0x50DCF2: __md_run_MOD_qs_mol_dyn (md_run.F:153) ==62== Address 0x26d18670 is 16 bytes before a block of size 10 free'd ==62== at 0x4C35FAC: free (vg_replace_malloc.c:538) ==62== by 0x2B73E68: __offload_api_MOD_offload_timeset (offload_api.F:137) ==62== by 0x2B60EDA: __timings_MOD_timeset_handler (timings.F:278) ==62== by 0x2BE2C6D: __message_passing_MOD_mp_waitany (message_passing.F:4597) ==62== by 0x2963EA5: __realspace_grid_types_MOD_rs_pw_transfer_distributed (realspace_grid_types.F:1439) ==62== by 0x2966559: __realspace_grid_types_MOD_rs_pw_transfer (realspace_grid_types.F:711) ==62== by 0xC9310B: __qs_collocate_density_MOD_calculate_rho_core (qs_collocate_density.F:966) ==62== by 0xF57698: __qs_update_s_mstruct_MOD_qs_env_update_s_mstruct (qs_update_s_mstruct.F:109) ==62== by 0xCC03AB: __qs_energy_init_MOD_qs_energies_init (qs_energy_init.F:311) ==62== by 0xCBF0A1: __qs_energy_MOD_qs_energies (qs_energy.F:84) ==62== by 0xCE087E: __qs_force_MOD_qs_forces (qs_force.F:212) ==62== by 0xCE4349: __qs_force_MOD_qs_calc_energy_force (qs_force.F:117) ==62== Block was alloc'd at ==62== at 0x4C34DFF: malloc (vg_replace_malloc.c:307) ==62== by 0x2F21116: _gfortrani_xmallocarray (memory.c:66) ==62== by 0x2F1C271: _gfortran_string_trim (string_intrinsics_inc.c:167) ==62== by 0x2B73E1C: __offload_api_MOD_offload_timeset (offload_api.F:137) ==62== by 0x2B60EDA: __timings_MOD_timeset_handler (timings.F:278) ==62== by 0x2BE2C6D: __message_passing_MOD_mp_waitany (message_passing.F:4597) ==62== by 0x2963EA5: __realspace_grid_types_MOD_rs_pw_transfer_distributed (realspace_grid_types.F:1439) ==62== by 0x2966559: __realspace_grid_types_MOD_rs_pw_transfer (realspace_grid_types.F:711) ==62== by 0xC9310B: __qs_collocate_density_MOD_calculate_rho_core (qs_collocate_density.F:966) ==62== by 0xF57698: __qs_update_s_mstruct_MOD_qs_env_update_s_mstruct (qs_update_s_mstruct.F:109) ==62== by 0xCC03AB: __qs_energy_init_MOD_qs_energies_init (qs_energy_init.F:311) ==62== by 0xCBF0A1: __qs_energy_MOD_qs_energies (qs_energy.F:84)
--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns... at googlegroups.com<mailto:cp2k+uns... at googlegroups.com>.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/4d753954-6ada-4d19-9092-931469136f8en%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/4d753954-6ada-4d19-9092-931469136f8en%40googlegroups.com?utm_medium=email&utm_source=footer>.
--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns... at googlegroups.com<mailto:cp2k+uns... at googlegroups.com>.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/c7060174-3f71-4179-8e70-ce5f179859a4n%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/c7060174-3f71-4179-8e70-ce5f179859a4n%40googlegroups.com?utm_medium=email&utm_source=footer>.
--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns... at googlegroups.com<mailto:cp2k+uns... at googlegroups.com>.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/f282de9c-e731-4232-838c-96139638c9bcn%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/f282de9c-e731-4232-838c-96139638c9bcn%40googlegroups.com?utm_medium=email&utm_source=footer>.
--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe at googlegroups.com<mailto:cp2k+unsubscribe at googlegroups.com>.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/5d62512a-817a-4566-be6c-e47a52810222n%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/5d62512a-817a-4566-be6c-e47a52810222n%40googlegroups.com?utm_medium=email&utm_source=footer>.
--
You received this message because you are subscribed to a topic in the Google Groups "cp2k" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/cp2k/uEodASnEFfQ/unsubscribe.
To unsubscribe from this group and all its topics, send an email to cp2k+unsubscribe at googlegroups.com<mailto:cp2k+unsubscribe at googlegroups.com>.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/ZRAP278MB0827795C76D23FF284B7D273F4AFA%40ZRAP278MB0827.CHEP278.PROD.OUTLOOK.COM<https://groups.google.com/d/msgid/cp2k/ZRAP278MB0827795C76D23FF284B7D273F4AFA%40ZRAP278MB0827.CHEP278.PROD.OUTLOOK.COM?utm_medium=email&utm_source=footer>.
--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe at googlegroups.com<mailto:cp2k+unsubscribe at googlegroups.com>.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/a0274c5c-8f74-4673-bf72-9347d8780773%40gmail.com<https://groups.google.com/d/msgid/cp2k/a0274c5c-8f74-4673-bf72-9347d8780773%40gmail.com?utm_medium=email&utm_source=footer>.

-- 
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe at googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/ZRAP278MB082785DD83E8980CF6B9658FF4AFA%40ZRAP278MB0827.CHEP278.PROD.OUTLOOK.COM.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20231109/84222ae9/attachment-0001.htm>


More information about the CP2K-user mailing list