Skip to content

LUNA PLATFORM v.5.96.0#

LP changes

  • The default neural network version for extracting biometric face templates is now version 65 (the "DEFAULT_FACE_DESCRIPTOR_VERSION" setting of the Remote SDK service).

    When migrating the latest version of the Configurator service settings, the "DEFAULT_FACE_DESCRIPTOR_VERSION" setting will not be automatically updated. To continue working with existing facial biometric templates with the new 65th version of the neural network, you must manually change the value of the "DEFAULT_FACE_DESCRIPTOR_VERSION" setting, after re-extracting the existing biometric templates using the 65th model, using the task "Additional extraction. If you change the default version without running the task, the old biometric templates will become unavailable.

    See the section "Prepare to change the neural network version" for more information on updating the "DEFAULT_FACE_DESCRIPTOR_VERSION" setting.

  • A splittable streams mechanism has been added.

    A splittable stream is a stream with user-defined analytics that can be processed by different agents simultaneously. For licensing purposes, a splittable stream will be considered a single stream regardless of how many agents are involved in processing its analytics.

    All analytics will be processed (including logs, auto-restart, etc.) as a separate stream. For more information, see splittable streams.

    • The "create stream" request now supports the splittable parameter, which allows distributing stream processing to more than one agent in the event that no agent supports all specified analytics sets.

    • Updated responses to "get stream" and "get streams" requests:

    • status field: there are no changes for non-splittable streams, and the status of a splittable stream is the general status of all its parts. Added new statuses for splittable streams: unknown - a splittable stream has different statuses on each agent and handler_lost - a service status meaning that data from the agent processing the stream was not received and such a stream will be redistributed for processing (appears in the stream logs in the combination handler_lost-restart-pending).

    • the autorestart field has been moved to the analytics field: for a non-splittable stream, it will be the same for each of its parts (elements of the analytics array) and can be different for each part of the splittable stream depending on their state.

    • the status field has been added to the analytics field, describing the status of each part of the splittable stream.

    • Now for the request "stream events ws handshake" a new parameter analytic_index is available, which provides information only for analytics with the specified index, which corresponds to the analytics index from the requests "create stream" or "get stream(s)". The parameter is optional and by default the parameter takes the value 0, which corresponds to the first element of the analytics array.

  • Support for S3-compatible storages without built-in tagging functionality has been added to the Image Store service.

    Also, support for the TTL (Time To Live - automatic deletion of files after a specified period) policy has been added to these storages.

    When a bucket is created, a special file is placed in it containing the default ttl value for the objects stored in it; if the value is not specified, it is set to None, which means that the objects will not have an automatically assigned expiration date. To enable this mechanism, you must specify the TAGGING = 0 parameter in the S3 setting. When enabled, the object deletion date will be stored in the metadata for objects for which the ttl parameter was specified when created or modified. If the object does not have an explicitly specified ttl, the default ttl value of the bucket will be applied. However, if the default TTL value is not configured, the object will remain in the storage indefinitely. To explicitly prevent the deletion of an object, you should assign a ttl value of -1 to it.

  • The license feature request logic has been updated to support flexible licensing.

    The "get license" request of the Licenses service now supports any target specified by the user to get a feature from a license, if it is included in the license.

    Previously, only the following targets were available: expiration_time, faces_limit, liveness, iso, index, people_counter, body_attributes, streams_limit, lambdas. Now any other features available in the license are also available as targets.

    If the user-specified feature is not included in the license, an error with the code 33008 and the description License problem. Unknown feature: {feature_name}. will be returned.

  • Periodic log cleanup controlled by the LUNA_VIDEO_MANAGER_LOGS_CLEAR_INTERVAL setting has been added.

    This feature helps to automatically clear the Video Manager service database from outdated logs, preventing their accumulation. Periodic deletion of logs is configured using the following parameters from the LUNA_VIDEO_MANAGER_LOGS_CLEAR_INTERVAL parameter group:

    • "interval" - sets the interval for deleting logs. Logs older than this interval will be deleted. The default value is 7;
    • "interval_type" - sets the interval type (weeks, days, hours, minutes, seconds). The default value is days;
    • "check_interval" - sets the frequency of checking logs for deletion (seconds). The default value is 180;
    • "active" - ​​enables/disables automatic deletion of stream logs. By default, the parameter is disabled.

    When the "active" parameter is enabled, the default values ​​trigger automatic deletion by checking the stream logs in the database every 180 seconds and deleting logs older than 7 days.

  • The Oracle database is no longer supported in the Lambda and Video Manager services.

  • The start_platform.sh and start_logging.sh scripts now support automatic selection between docker-compose (a separate binary) and docker compose (a built-in command in docker).

    Previously, attempting to run scripts with docker compose would result in an error if docker-compose was not installed.

    The instructions for installing Docker Compose as a binary file have also been removed from the documentation.

LP fixed errors

  • Fixed a bug with Redis Sentinel authentication using login and password.

    Previously, the Tasks service did not work with a Sentinel configuration that specified a non-default user and required authentication. A MasterNotFoundError error occurred with the message Authentication required, since the system did not transmit the login/password when connecting to Sentinel.

  • A missing pattern for the stream name was added to the "create stream" and "put stream" request specification.

    The stream name pattern is ^[a-zA-Z0-9_\-]{0,128}$, which means that the stream name can only contain: - latin letters (a-z, A-Z); - digits (0-9); - underscore (_) and hyphen (-).

    The pattern allows: length from 1 to 128 characters — for meaningful names (e.g., stream_1) or an empty string ("") — as an extreme technical case, but this is not recommended for real use.

  • Fixed a bug that allowed passing an unused ttl parameter to image_retain_policy for people_count and human_tracking analytics.

    This only applies to the "/videosdk" resource, since no data is stored there, unlike streams, where data is stored.

  • The missing description of the source field in the "callback request" has been added to the People count analytics specification.

  • Fixed duplication of logs when requesting "get streams logs" to a deleted stream.

  • The logic for processing the stream when there is no response from the agent in Video Manager has been updated.

    If a stream was sent for processing, but the agent does not confirm its status (for example, due to a failure or unavailability), the stream is now successively assigned statuses: - handler_lost - a service status indicating that data from the agent processing the stream was not received. - restart - the stream is marked for restart. - pending - the stream status is changed to "waiting" for redistribution to another agent.

    Previously, in such cases, only the "restart-pending" sequence was recorded in the logs.

API service UI changes

  • "Checks" section: The ability to upload video files for Liveness, Deepfake checks, as well as for face and body attributes assessment, has been added.