A new release is available for Arkindex instances.
You can learn more about Arkindex on its official documentation.
Important changes
In the API, all the process-related endpoints and URLs have been renamed: mentions of "dataimport" were replaced by "process". For example, the endpoint to list Arkindex processes is now ListProcesses
instead of ListDataImports
, and the API URL has changed from /api/v1/imports/
to /api/v1/process/
. As a consequence, these endpoints are now under a process
section in the API documentation.
Process
- It is now possible to specify a
shm_size
parameter in a worker version's docker configuration. This allows for the increase of the available shared memory in/dev/shm
, which was required for the training of some workers. - Users with a contributor-level access to workers (or any user, for public workers) can now start training processes for these workers.
- It is now possible to create parameters in the worker's user configuration that take a list of booleans as input, using the
bool
subtype.
-
There have been some improvements to the worker activity feature:
- When a process is retried, its attached worker activities are re-initialized, allowing for the re-assignment of existing worker activities to the new process, as well as re-creating missing worker activities (for example, if new elements were created).
- This activity initialization can "steal" worker activities from other, older processes with the same worker, worker version and configuration; an older process, however, cannot "steal" a newer process' worker activities.
- A worker activity's timeout (1 hour by default) is now always calculated from when it was marked as
started
, even if the process was retried (and the activity was marked asstarted
again), not from when it was created. - When a task fails during a process, the started worker activities are all marked as
error
.
-
Our worker execution system, Ponos, also underwent a few improvements:
- A bug was fixed which prevented Ponos agents from properly freeing up space after downloading Docker images, causing hosts to run out of space.
- Empty worker runs (no tasks) are now handled when retrieving a workflow's state, instead of causing an error, which could occur when starting a workflow while the database is under heavy load.
API
- The
ListElementMetaData
endpoint (API documentation) now has aload_parents
option, which allows it to retrieve an element's parents' metadata along with its own. - HTML sanitization (which removed characters like
&
or>
) is now restricted to Markdown metadata, instead of being applied to all metadata. Read more about the rules that apply to the different metadata types in the Arkindex documentation. - The bulk creation endpoint
CreateElementTranscriptions
now supports adding a confidence score to the created elements, not just the transcriptions.
Frontend
- The Arkindex frontend now supports TypeScript, allowing developers to type variables in order to produce better code more easily and avoid mistakes.
- The element path display in the element header when browsing elements with multiple parents has ben improved: the different paths stay displayed in the same order when going from one child element to another.
- The Search form and view have been refactored, to solve a number of issues.
- The batch deletion mode is now disabled when the annotation panel is closed.
- The file import interface's error handling has been improved.
- The name editing field is now closed when navigating to another element.
- Users are now blocked from accessing the process configuration and element filtering pages for unconfigurable processes.
Transkribus import
The polygons created by a Transkribus import are now trimmed to fit inside their image if they overflow it, and ignored if they are outside the image's bounds, instead of triggering an error.