# Verdi computer test, Creating and Deleting Temporary fails when work\_dir is the scratch directory in HPC **URL:** \ **Category:** New to AiiDA\ **Tags:** question\ **Created:** [March 3, 2025, 4:38pm UTC](https://aiida.discourse.group/t/verdi-computer-test-creating-and-deleting-temporary-fails-when-work-dir-is-the-scratch-directory-in-hpc/575 "2025-03-03T16:38:27Z")\ **Posts on this page:** 3\ **Page:** 1
**Author:** ![Afnan\_Shahid](https://yyz2.discourse-cdn.com/free1/user_avatar/aiida.discourse.group/afnan_shahid/32/277_2.png) [@Afnan\_Shahid](https://aiida.discourse.group/u/Afnan_Shahid)\ **Post date:** [March 3, 2025, 4:38pm UTC](https://aiida.discourse.group/t/verdi-computer-test-creating-and-deleting-temporary-fails-when-work-dir-is-the-scratch-directory-in-hpc/575/1 "2025-03-03T16:38:27Z")
Hi, I am new to aiida as well as hpc for highthroughput worklfow, so applogies in advance if I miss something obvious. So, the main issue is, verdi computer test is successful for all but the creation and deletion of temporary file when I use the scratch directory (lustre/parallel file system on hpc) as work\_dir for configuring the hpc computer : $ verdi computer show noctua2 * * * Label noctua2 PK 10 UUID 8a16267b-5570-47f3-a10e-c84764f9d953 Description Noctua 2 HPC Cluster at PC2 Hostname [fe.noctua2.pc2.uni-paderborn.de](http://fe.noctua2.pc2.uni-paderborn.de) Transport type core.ssh Scheduler type core.slurm Work directory /scratch/hpc-prf-spectr/Abdullah/temp Shebang #!/bin/bash Mpirun command srun -n {tot\_num\_mpiprocs} Default #procs/machine 64 Default memory (kB)/machine 268435456 Prepend text export PATH=/opt/software/pc2/EB-SW/software/OpenMPI/4.1.6-GCC-13.2.0/bin:$PATH export LD\_LIBRARY\_PATH=/opt/software/pc2/EB-SW/software/OpenMPI/4.1.6-GCC-13.2.0/lib:$LD\_LIBRARY\_PATH Append text echo “Job execution complete.” * * * $ verdi computer test noctua2 --print-traceback Report: Testing computer for user[aiida@localhost](mailto:aiida@localhost)… - Opening connection… [OK] - Checking for spurious output… [OK] - Getting number of jobs from scheduler… [OK]: 0 jobs found in the queue - Determining remote user name… [OK]: abshahid - Creating and deleting temporary file… [Failed]: OSError: Error during mkdir of ‘/scratch’, maybe you don’t have the permissions to do it, or the directory already exists? ([Errno 13] Permission denied) Full traceback: Traceback (most recent call last): File “/opt/conda/lib/python3.10/site-packages/aiida/cmdline/commands/cmd\_computer.py”, line 124, in \_computer\_create\_temp\_file transport.chdir(workdir) File “/opt/conda/lib/python3.10/site-packages/aiida/transports/plugins/ssh.py”, line 593, in chdir self.sftp.chdir(path) File “/opt/conda/lib/python3.10/site-packages/paramiko/sftp\_client.py”, line 659, in chdir if not stat.S\_ISDIR(self.stat(path).st\_mode): File “/opt/conda/lib/python3.10/site-packages/paramiko/sftp\_client.py”, line 493, in stat t, msg = self.\_request(CMD\_STAT, path) File “/opt/conda/lib/python3.10/site-packages/paramiko/sftp\_client.py”, line 822, in \_request return self.\_read\_response(num) File “/opt/conda/lib/python3.10/site-packages/paramiko/sftp\_client.py”, line 874, in \_read\_response self.\_convert\_status(msg) File “/opt/conda/lib/python3.10/site-packages/paramiko/sftp\_client.py”, line 903, in \_convert\_status raise IOError(errno.ENOENT, text) FileNotFoundError: [Errno 2] No such file - Checking for possible delay from using login shell… [OK] Warning: 1 out of 6 tests failed However, all the 6 tests are successful when I use the home directory (small storage). : $ verdi computer show noctua2 * * * Label noctua2 PK 11 UUID 5e018103-0bd1-457f-8f21-698dda8ca813 Description Noctua 2 HPC Cluster at PC2 Hostname [fe.noctua2.pc2.uni-paderborn.de](http://fe.noctua2.pc2.uni-paderborn.de) Transport type core.ssh Scheduler type core.slurm Work directory /pc2/users/a/abshahid/aiida\_wor\_dira Shebang #!/bin/bash Mpirun command srun -n {tot\_num\_mpiprocs} Default #procs/machine 64 Default memory (kB)/machine 268435456 Prepend text export PATH=/opt/software/pc2/EB-SW/software/OpenMPI/4.1.6-GCC-13.2.0/bin:$PATH export LD\_LIBRARY\_PATH=/opt/software/pc2/EB-SW/software/OpenMPI/4.1.6-GCC-13.2.0/lib:$LD\_LIBRARY\_PATH Append text echo “Job execution complete.” * * * $ verdi computer test noctua2 Report: Testing computer for user[aiida@localhost](mailto:aiida@localhost)… - Opening connection… [OK] - Checking for spurious output… [OK] - Getting number of jobs from scheduler… [OK]: 0 jobs found in the queue - Determining remote user name… [OK]: abshahid - Creating and deleting temporary file… [OK] - Checking for possible delay from using login shell… [OK] Success: all 6 tests succeeded I have tried to investigate, this is what I was able to understand: AiiDA relies on SSH-based SFTP operations (using Paramiko) to manage remote work directories during its computer tests. It first attempts to change to the designated work directory using sftp.chdir(), and if that fails (typically because the directory doesn’t exist or isn’t accessible), it then tries to create it using transport.makedirs(). HPC systems like Noctua 2 could have SFTP chrooted to my home directory. If the chroot restriction is true, it means that when AiiDA attempts to access directories outside the home (e.g., a scratch space), the operations fail. I have also tried using symlink but didnot work maybe due to permission issues with symbolic links. So, my question is , What do you think about the issue? I have tried to look up the discussions here, but could not find any relavant posts, does that mean it is not a typical issue? Is this my HPC specific issue or am I thinking the whole thing wrong and making a very obvious mistake? I highly appreciate any solution or advice about my problem. Thank you. 😊 [noctua2\_computer.txt](https://aiida.discourse.group/uploads/short-url/mGAM7dVAcYCIBCHrC19QZp2dqEO.txt) (649 Bytes) [noctua2\_ssh\_config.txt](https://aiida.discourse.group/uploads/short-url/dzQoqOseCI3UiJNVG4ic2mlwnsV.txt) (400 Bytes) ---
**Author:** ![giovannipizzi](https://yyz2.discourse-cdn.com/free1/user_avatar/aiida.discourse.group/giovannipizzi/32/17_2.png) [@giovannipizzi](https://aiida.discourse.group/u/giovannipizzi)\ **Post date:** [March 3, 2025, 7:37pm UTC](https://aiida.discourse.group/t/verdi-computer-test-creating-and-deleting-temporary-fails-when-work-dir-is-the-scratch-directory-in-hpc/575/2 "2025-03-03T19:37:24Z")
Hi, assuming that `/scratch` actually exists, indeed the issue is that AiiDA is trying to create `/scratch` that is clearly not possible. If you just `ssh` into the computer `fe.noctua2.pc2.uni-padeborn.de`, and do `ls /scratch`, does it work or it gives an error? In the second case, you should ask your supercomputer center. In the first case, `/scratch` does not exist, at least on the system you are trying to SSH to. From a quick google search, the instructions for your center are [these](https://upb-pc2.atlassian.net/wiki/spaces/PC2DOK/pages/1903056/Data+Transfer+File+Staging) and they say to e.g. `rsync` as ```auto scp -o 'ProxyJump @fe.noctua2.pc2.uni-paderborn.de' @n2login5:/scratch/ ``` So you actually do not want to stop in [fe.noctua2.pc2.uni-paderborn.de](http://fe.noctua2.pc2.uni-paderborn.de) but proxy through it, but then connect to one of the login nodes `n2login5`, but they say > You can use other cluster frontends (`n2login1` , `n2login2` , …) as the target. So (assuming you see also your home from the login nodes, as well as can submit/monitor jobs) I suggest you create a new computer where the hostname is `n2login5` instead (for instance) and use AiiDA’s supported ProxyJump feature to access it via [fe.noctua2.pc2.uni-paderborn.de](http://fe.noctua2.pc2.uni-paderborn.de). Hope it helps! (Otherwise I suggest to ask support also to the supercomputer center) Giovanni ---
**Author:** ![wasanovo](https://avatars.discourse-cdn.com/v4/letter/w/a88e57/32.png) [@wasanovo](https://aiida.discourse.group/u/wasanovo)\ **Post date:** [October 26, 2026, 7:50pm UTC](https://aiida.discourse.group/t/verdi-computer-test-creating-and-deleting-temporary-fails-when-work-dir-is-the-scratch-directory-in-hpc/575/3 "2026-10-26T19:50:44Z")
The traceback points to the configured endpoint/path, not a general AiiDA limitation: after `chdir` fails, AiiDA tries to create the missing path, and that attempt reaches `/scratch`, where your account cannot create directories. As Giovanni suggests, try the supported login-node hostname with `ProxyJump`. Also check that the full work directory exists and is accessible via SFTP on that endpoint; shell access alone doesn’t confirm Paramiko’s SFTP operations work there.