Skip to main content

Input/output errors on all filesystems

Operations on files across all Lustre partitions (lustrep, flash) frequently encounter the “Input/output error.” These failures happen randomly, with larger files being more susceptible to trigger errors.

It appears that the error predominantly occurs on the login nodes, particularly when they are under heavy load. To mitigate this issue, we recommend using an interactive SLURM session to perform file operations on a compute node instead of the login nodes. Please, let us know if this solution doesn’t work in your case.

A critical issue ticket has been opened at HPE, and we will provide updates as soon as we receive more information. We understand that this problem significantly impacts the usability of LUMI for various workloads, and we sincerely apologize for any inconvenience caused.

If you have any further inquiries, please don’t hesitate to reach out to the LUMI user support team.