Identified - We have a data imbalance issue within the distributed storage that underpins the FlexiHPC object storage service and "ceph-hdd" type volumes. This has caused the object store to report intermittent errors to clients over the weekend when attempting to write new data. Cloud instances using the ceph-hdd volume type may have experienced intermittent IO errors and/or high latency. We expect most FlexiHPC users to be unaffected as ceph-ssd is the default and is not impacted.
Due to the amounts of data involved, resolving this issue and rebalancing the cluster may take some time. Please stay tuned for updates. If your deployment has been impacted by these issues and you need assistance recovering then please reach out to support.
Feb 17, 2025 - 10:20 NZDT
Monitoring - Good news, everyone! The Slurm controller services are back online as of yesterday, and we're keeping a close eye on things. Unfortunately, a few hundred jobs decided to take an unexpected holiday break due to the outage. 😅
In a classic case of split-brain, the controllers couldn't agree on what was happening, so the status of some failed jobs might be a bit... unreliable. We recommend you double-check those failed jobs to see if they need restarting, some of them may have completed. Apologies for the hiccup, and may your holidays be as glitch-free as possible! 🎄✨
Dec 24, 2024 - 12:57 NZDT
Identified - We've identified an error with Slurm controller that was preventing it from starting and are now working to get it back online. Root cause is so far unknown but may be down to a filesystem issue. It is likely that many jobs have been killed or finished in a failure state. We will provide an update on follow up actions for users once the service is stable.
Dec 23, 2024 - 10:34 NZDT
Investigating - The Mahuika Slurm Controller is not functioning. This affects the ability to submit new jobs and to launch Jupyter sessions. Our apologies, due to the time of year we are unable to give a time when this will be fixed due to reduced levels of staffing over the holiday period.
Dec 22, 2024 - 10:06 NZDT
Monitoring - A fix has been implemented and we are monitoring the results.
Dec 10, 2024 - 08:37 NZDT
Investigating - We are experiencing some issues with the backend storage supporting the Maui_ancil Slurm controllers. To fix this we will be shutting down the Maui_ancil controllers for a short period, immediately. New job submissions will not be possible whilst this occurs, but existing jobs should be unaffected.
Dec 09, 2024 - 16:09 NZDT