Skip to content
AiQu Docs

v3.46.0

Administrators can add and remove directories on a connected Docker node, node services start on the right version again, release notes have their own place on the docs site, and node images run on ARM machines. The rest are fixes to impersonation, tenant registries and cloning jobs with mounts.

New

Section titled “New”
  • Mounts on a connected Docker node. A tenant administrator can add a directory on the node from the node dialog, and remove one. Each directory becomes a Storage Bucket Collection with a bucket, ready to grant to a group or a project, and the dialog shows the volume line to add to the node’s agent. A directory cannot be removed while its bucket is granted to anyone or mounted in a job that has not ended. The node’s storage service restarts so the file browser sees the change; nothing else on the node is restarted.
  • Release notes on the documentation site, in English and Swedish, with an index of every version. The start page links to them.
  • Node images for ARM. The images a compute node runs are now published for arm64 as well as amd64, so ARM machines can join as nodes.

Fixed

Section titled “Fixed”
  • Node services could fail to start, or start on an older version than the server. This affected newly connected nodes and services restarted with update. Services already running are not changed; they move to the current version the next time they are restarted with update.
  • An image that failed to download was reported as downloaded, for example when the node’s disk was full. The job or node service then failed to start without saying why. The download now fails with the reason.
  • A failed job could show a countdown bar after it had let go of its node, for example when it was canceled after failing. The bar now shows only while the job still holds its place on the node.
  • A node’s storage could be listed several times after it reconnected. The duplicates are removed on upgrade.
  • The filter field on the Storage page disappeared when nothing matched the filter, so it could not be cleared.
  • A node on NVIDIA GB10 (DGX Spark) registered without any GPUs, and said nothing about it. The GPU now shows up with its name and id; its memory size is not reported, since the system does not provide it.
  • Cloning a job dropped its mounts. This hit the clone button in the job list and in job details.
  • A tenant administrator could not make a registry available to the whole tenant. The setting was never saved, so the registry stayed private. There is now a switch for it, and a failed save shows the error.
  • Saving a registry could fail with “Database Error”, depending on which settings were changed.
  • Superadmin: impersonating a user could send you back to the login page, because one click sent the request twice. This is fixed in both interfaces.
  • The impersonate button could end up outside the user list when email addresses were long, so it could only be reached by scrolling sideways. The list now fits, and it can be filtered by email, name, username or id, or to administrators only. Users without a name no longer show as “null null”.

Changed

Section titled “Changed”
  • “Storage Bucket Collection” is now the name used throughout the interface and the documentation.
  • Documentation links follow the environment. Links from the start page used to lead to the production documentation from every environment.