> ## Documentation Index
> Fetch the complete documentation index at: https://rasa.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Rasa Pro Services Change Log

> All notable changes to Rasa Pro Services will be documented in this page.

All notable changes to Rasa Pro Services will be documented in this page.
This product adheres to [Semantic Versioning](https://semver.org/) starting with version 3.3 (initial version).

## \[3.10.1] - 2026-08-28

Rasa Pro Services 3.10.1 (2026-08-28)

### Bugfixes

* The `rasa-analytics` Kafka consumer no longer crashes with a `UniqueViolation` on `rasa_session` / `rasa_session_slot_state` when a message is redelivered after a Kafka partition rebalance.
* Fixed a race condition where concurrent inserts could create duplicate `rasa_sender` rows for the same `sender_key`, permanently breaking reads.
  Added a `UNIQUE` constraint on `sender_key` with a migration that deduplicates existing rows and made Kafka offset commits synchronous so failures surface.
* Bounded the Kafka consumer's in-memory message and DLQ queues to prevent unbounded memory growth when the analytics database is slow. `QUEUE_MAX_SIZE` limits buffered batches (not individual messages) and defaults to 10 — enough to absorb normal DB jitter without triggering backpressure every batch. The DLQ queue now uses its own constant `DLQ_QUEUE_MAX_SIZE`, sized in individual messages with a floor of `MAX_MESSAGES_TO_FETCH`, so a full batch of invalid or failed events can all reach the DLQ without silent drops under normal conditions. When the DLQ queue is full (e.g. the DLQ producer cannot keep up), messages are dropped with a warning log. Because `acknowledge()` commits per-message, dropped offsets cannot be re-consumed — this is a bounded-loss trade-off to avoid blocking the main processing loop. When the message queue is full, assigned Kafka partitions are paused until the writer catches up. All queue `put()` calls are non-blocking or use timeouts to prevent shutdown hangs and self-deadlocks.

## \[3.10.0] - 2026-07-21

Rasa Pro Services 3.10.0 (2026-07-21)

### Improvements

* Bot response payloads (which contain buttons, quick replies, custom JSON, attachment, image, elements) are now persisted in a new `response_data` column on `_rasa_raw_event` and `rasa_bot_message`. Bot responses are now fully reconstructable from the same row: `text` holds the message text and `response_data` holds the structured payload.

### Bugfixes

* The `rasa-analytics` Kafka consumer no longer fails with `PendingRollbackError` after a stale pooled database connection. Pooled connections are now validated on checkout, and a database session is rolled back after a failed operation instead of being reused in a broken state.

## \[3.9.1] - 2026-08-28

Rasa Pro Services 3.9.1 (2026-08-28)

### Bugfixes

* Upgrade:
  * `pyjwt` to version `2.13.0` to address CVE-2026-48526, CVE-2026-48522, CVE-2026-48523 & CVE-2026-48525
  * `cryptography` to version `48.0.1` to address GHSA-537c-gmf6-5ccf
* The `rasa-analytics` Kafka consumer no longer fails with `PendingRollbackError` after a stale pooled database connection. Pooled connections are now validated on checkout, and a database session is rolled back after a failed operation instead of being reused in a broken state.
* The `rasa-analytics` Kafka consumer no longer crashes with a `UniqueViolation` on `rasa_session` / `rasa_session_slot_state` when a message is redelivered after a Kafka partition rebalance.
* Fixed a race condition where concurrent inserts could create duplicate `rasa_sender` rows for the same `sender_key`, permanently breaking reads.
  Added a `UNIQUE` constraint on `sender_key` with a migration that deduplicates existing rows and made Kafka offset commits synchronous so failures surface.
* Bounded the Kafka consumer's in-memory message and DLQ queues to prevent unbounded memory growth when the analytics database is slow. `QUEUE_MAX_SIZE` limits buffered batches (not individual messages) and defaults to 10 — enough to absorb normal DB jitter without triggering backpressure every batch. The DLQ queue now uses its own constant `DLQ_QUEUE_MAX_SIZE`, sized in individual messages with a floor of `MAX_MESSAGES_TO_FETCH`, so a full batch of invalid or failed events can all reach the DLQ without silent drops under normal conditions. When the DLQ queue is full (e.g. the DLQ producer cannot keep up), messages are dropped with a warning log. Because `acknowledge()` commits per-message, dropped offsets cannot be re-consumed — this is a bounded-loss trade-off to avoid blocking the main processing loop. When the message queue is full, assigned Kafka partitions are paused until the writer catches up. All queue `put()` calls are non-blocking or use timeouts to prevent shutdown hangs and self-deadlocks.

## \[3.9.0] - 2026-06-29

Rasa Pro Services 3.9.0 (2026-06-29)

### Improvements

* Added created\_at UTC insertion timestamp column and corresponding index to all analytics tables.
  Added indexes on `timestamp` and `inserted_at` where those columns exist.

### Bugfixes

* Fixed a long-standing bug in the analytics legacy session-detection path that split a single Rasa conversation session into two analytics sessions. `action_session_start` and the `session_started_metadata` slot both unconditionally marked a new session, so every Rasa session start produced a first session holding only the events between the two markers and a second session holding the rest. The slot check is now symmetric to the action check and treats the slot as part of an in-flight session start when `action_session_start` was the most recent anchor event for the sender.
* Upgraded versions of for urllib3, idna, mako and redshift-connector and accompanying dependecies, because of reported vulnerabilites.
* Removed the unused `cargo` build toolchain, including its bundled rustc source tree, from the production image.

## \[3.8.2] - 2026-08-28

Rasa Pro Services 3.8.2 (2026-08-28)

### Bugfixes

* Updated `mako`, `idna`, `redshift-connector`, `python-dotenv `, `pip` and `urllib3` to address security vulnerabilities.
* Removed the unused `cargo` build toolchain, including its bundled rustc source tree, from the production image.
* Upgrade:
  * `pyjwt` to version `2.13.0` to address CVE-2026-48526, CVE-2026-48522, CVE-2026-48523 & CVE-2026-48525
  * `cryptography` to version `48.0.1` to address GHSA-537c-gmf6-5ccf
* The `rasa-analytics` Kafka consumer no longer fails with `PendingRollbackError` after a stale pooled database connection. Pooled connections are now validated on checkout, and a database session is rolled back after a failed operation instead of being reused in a broken state.
* The `rasa-analytics` Kafka consumer no longer crashes with a `UniqueViolation` on `rasa_session` / `rasa_session_slot_state` when a message is redelivered after a Kafka partition rebalance.
* Fixed a race condition where concurrent inserts could create duplicate `rasa_sender` rows for the same `sender_key`, permanently breaking reads.
  Added a `UNIQUE` constraint on `sender_key` with a migration that deduplicates existing rows and made Kafka offset commits synchronous so failures surface.
* Bounded the Kafka consumer's in-memory message and DLQ queues to prevent unbounded memory growth when the analytics database is slow. `QUEUE_MAX_SIZE` limits buffered batches (not individual messages) and defaults to 10 — enough to absorb normal DB jitter without triggering backpressure every batch. The DLQ queue now uses its own constant `DLQ_QUEUE_MAX_SIZE`, sized in individual messages with a floor of `MAX_MESSAGES_TO_FETCH`, so a full batch of invalid or failed events can all reach the DLQ without silent drops under normal conditions. When the DLQ queue is full (e.g. the DLQ producer cannot keep up), messages are dropped with a warning log. Because `acknowledge()` commits per-message, dropped offsets cannot be re-consumed — this is a bounded-loss trade-off to avoid blocking the main processing loop. When the message queue is full, assigned Kafka partitions are paused until the writer catches up. All queue `put()` calls are non-blocking or use timeouts to prevent shutdown hangs and self-deadlocks.

## \[3.8.1] - 2026-04-17

Rasa Pro Services 3.8.1 (2026-04-17)

### Bugfixes

* Updated the following Poetry dependencies with vulnerability patches:

  * `requests` to `2.33.1` fixing CVE-2026-25645.
  * `cryptography` to `46.0.7` fixing CVE-2026-39892.
  * `setuptools` to `82.0.1`

  Additionally, updated pip installed in Dockerfile to `==26.x` fixing CVE-2026-24049 vulnerability in `wheel` and updated `setuptools`
  to the version locked in `poetry.lock` which updated `jaraco.context` dependency to `6.1.0` fixing CVE-2026-23949.

## \[3.8.0] - 2026-03-26

Rasa Pro Services 3.8.0 (2026-03-26)

### Improvements

* Add the ability to disable stack event processing by the analytics service through an environment variable.
  Set the `DISABLE_STACK_EVENT_PROCESSING` variable to `true` to disable analytics stack event processing.
  By default, stack event processing is enabled.
* Add new composite and covering indexes to existing tables in Rasa Analytics to improve the performance of high-load
  analytics queries and reduce CPU utilization. These indexes will complement the current indexes and
  help optimize queries that frequently filter by sender\_id, timestamp or session\_id.
* Add a new stack event query for the `event_was_processed` check to handle edge cases when stack events are stored in
  the `rasa_event` table but not in the `rasa_dialogue_stack_frame` table because of being skipped during processing and
  published to the DLQ.  This allows the DLQ stack events to be reprocessed correctly when the source Kafka topic offset
  is reset to the minimum offset among the events published to the DLQ.
* Update Analytics session processing to use the `session_id` provided by Rasa Pro 3.16 in `event.metadata.session_id` directly as the session identifier, replacing the Analytics-generated UUID.
  A new `rasa_user` table is introduced to track end users across multiple conversations, populated from the `user_id` field on Kafka event payloads.
  Events from older Rasa versions without `metadata.session_id` continue to be processed correctly via the existing heuristic session detection.

### Bugfixes

* Upgrade `urllib3`, `werkzeug` and `brotli` to fix security vulnerabilities.
* Update urllib3 to 2.6.3 to address CVE-2026-21441.
  Update werkzeug to 3.1.5 to address CVE-2026-21860.
* Update flask to 3.1.3 to address CVE-2026-27205
  Update werkzeug to 3.1.6 to address CVE-2026-27199
  Update cryptography to 46.0.5 to address CVE-2026-26007
  Update sentry-sdk to 2.42.1 to address CVE-2024-40647
* Pin Certifi to 2026.2.25 to address CVE-2024-39689
  Updated PyJWT to 2.12.0 to address CVE-2026-32597

## \[3.7.0] - 2025-11-26

Rasa Pro Services 3.7.0 (2025-11-26)

### Improvements

* Add the original Rasa event message offset to the header of the message being sent to the Kafka dead-letter-queue.
  This allows easier tracing of the original message in case of errors.
  The offset header is named `original_offset`.
* Allow skipping of db migration run when starting the Analytics service by setting the environment variable RUN\_ANALYTICS\_DB\_MIGRATIONS to "false".
  This can be useful in scenarios where migrations have already been applied or when managing migrations separately.
* Add database indexes to analytics tables to improve query performance.
  The following indexes were added to speed up analytics throughput and reduce query execution time during event processing
  and transformation:
  * `idx_rasa_dialogue_stack_frame_sender_session_seq` on `rasa_dialogue_stack_frame` table
  * `idx_rasa_event_pattern_query` on `rasa_event` table
  * `idx_rasa_event_stack_query` on `rasa_event` table
  * `idx_rasa_event_was_processed` on `rasa_event` table
  * `idx_rasa_sender_sender_key` on `rasa_sender` table
  * `idx_rasa_session_sender_seq` on `rasa_session` table
  * `idx_rasa_turn_sender_session_seq` on `rasa_turn` table

## \[3.6.2] - 2025-11-25

Rasa Pro Services 3.6.2 (2025-11-25)

### Bugfixes

* Implement custom timestamp type inheriting from `sqlalchemy.types.TypeDecorator`.
  This custom type normalizes and denormalizes timestamp columns across all db tables to ensure consistent timezone handling
  during inserting and querying.

## \[3.6.1] - 2025-10-29

Rasa Pro Services 3.6.1 (2025-10-29)

### Bugfixes

* Catch all exceptions raised during DialogueStackUpdated event processing and skip the event if an exception occurs.
  Add the skipped events to the DLQ topic, while allowing other events in the batch to be processed normally.
* Update `urllib3` version to `2.5.0` to address security vulnerabilities CVE-2024-37891 & CVE-2025-50181.

### Miscellaneous internal changes

*Miscellaneous internal changes.*

## \[3.6.0] - 2025-10-10

Rasa Pro Services 3.6.0 (2025-10-10)

### Features

* Add support for IAM authentication for AWS RDS database in the analytics service.
  To enable this feature, set the following environment variables:

  * `IAM_CLOUD_PROVIDER`: set to `aws` to enable IAM authentication for AWS RDS
  * `RASA_ANALYTICS_DB_HOST_NAME`: `<your-db-hostname>` the hostname of the RDS instance
  * `RASA_ANALYTICS_DB_PORT`: `<your-db-port>` the port of the RDS instance
  * `RASA_ANALYTICS_DB_NAME`: `<your-db-name>` the name of the database
  * `RASA_ANALYTICS_DB_USERNAME`: `<your-db-username>` the username to connect to the database
  * `AWS_DEFAULT_REGION`: `<your-aws-region>` the AWS region where the RDS instance is hosted

  Additionally, you can also set the following optional environment variables:

  * `RASA_ANALYTICS_DB_SSL_MODE`: the SSL mode to use when connecting to the database, e.g. `verify-full` or `verify-ca`
  * `RASA_ANALYTICS_DB_SSL_CA_LOCATION`: the path to the SSL root certificate to use when connecting to the database

  When these environment variables are set, the analytics service will use IAM authentication to connect to the AWS RDS database.
* Add support for IAM authentication for AWS Managed Streaming for Apache Kafka in the analytics service.
  To enable this feature, set the following environment variables:

  * `IAM_CLOUD_PROVIDER`: set to `aws` to enable IAM authentication for AWS MSK
  * `AWS_DEFAULT_REGION`: `<your-aws-region>` the AWS region where the MSK instance is hosted
  * `KAFKA_SECURITY_PROTOCOL`: set to `SASL_SSL` to use SASL over SSL
  * `KAFKA_SASL_MECHANISM`: set to `OAUTHBEARER` to use OAuth Bearer token authentication
  * `KAFKA_SSL_CA_LOCATION`: the path to the SSL root certificate to use when connecting to the MSK cluster,
    this can be downloaded from Amazon Trust Services.

  When these environment variables are set, the analytics service will use IAM authentication to generate temporary credentials
  to connect to AWS MSK.
* Configure Analytics service to connect to Kafka broker using mTLS.
  To enable this feature, set the following environment variables:
  * `KAFKA_SSL_CERTFILE_LOCATION`: Path to the client certificate file.
  * `KAFKA_SSL_KEYFILE_LOCATION`: Path to the client private key file.

### Improvements

* Add new environment variables for each AWS service integration (RDS, MSK) that indicates whether
  to use IAM authentication when connecting to the service:
  * `KAFKA_MSK_AWS_IAM_ENABLED` - set to `true` to enable IAM authentication for MSK connections.
  * `RDS_SQL_DB_AWS_IAM_ENABLED` - set to `true` to enable IAM authentication for RDS connections.

### Bugfixes

* Refactor the logic for creating new sessions. It now gets the previous event context directly from the database which makes the logic more robust against cases where the in-memory event stream might not have the previous event.

## \[3.5.10] - 2025-11-25

Rasa Pro Services 3.5.10 (2025-11-25)

### Bugfixes

* Implement custom timestamp type inheriting from `sqlalchemy.types.TypeDecorator`.
  This custom type normalizes and denormalizes timestamp columns across all db tables to ensure consistent timezone handling
  during inserting and querying.

## \[3.5.9] - 2025-10-29

Rasa Pro Services 3.5.9 (2025-10-29)

### Bugfixes

* Catch all exceptions raised during DialogueStackUpdated event processing and skip the event if an exception occurs.
  Add the skipped events to the DLQ topic, while allowing other events in the batch to be processed normally.
* Update `urllib3` version to `2.5.0` to address security vulnerabilities CVE-2024-37891 & CVE-2025-50181.

### Miscellaneous internal changes

*Miscellaneous internal changes.*

## \[3.5.8] - 2025-10-10

Rasa Pro Services 3.5.8 (2025-10-10)

### Bugfixes

* Refactor the logic for creating new sessions. It now gets the previous event context directly from the database which makes the logic more robust against cases where the in-memory event stream might not have the previous event.

## \[3.5.7] - 2025-10-06

Rasa Pro Services 3.5.7 (2025-10-06)

### Bugfixes

* Add the installation of `zlib` OS dependency to Analytics Dockerfile to fix build issues when running the service.

## \[3.5.6] - 2025-09-11

Rasa Pro Services 3.5.6 (2025-09-11)

### Bugfixes

* Upgrade alpine base image to 3.19 in order to a security vulnerability in alpine:3.17

## \[3.5.5] - 2025-09-03

Rasa Pro Services 3.5.5 (2025-09-03)

### Bugfixes

* Upgrade vulnerabilities:
  * `requests` to version `2.32.5`
  * `cryptography` to version `43.0.3`
* Fix vulnerabilities in outdated `setuptools` and `pip` installed via the `python:3.9-slim` original base image used by
  the Analytics Docker image build.
  Replace `python:3.9-slim` base image with `alpine:3.17` which installs python 3.10.
  Drop python 3.9 support which has reached its EOL.

### Miscellaneous internal changes

*Miscellaneous internal changes.*

## \[3.5.4] - 2025-07-29

Rasa Pro Services 3.5.4 (2025-07-29)

### Bugfixes

* Upgrade the following dependencies to fix security vulnerabilities in the analytics service:
  * `setuptools` to version 78.1.1
  * `gunicorn` to version 23.0.0
  * `redshift-connector` to version 2.1.8
* Check if the message has already been processed before processing it.
  Commit message offset only after it was processed successfully.
  If message processing fails retry the transaction.
  If the message offset commit fails, retry the command.

  Added environment variables:

  * `RETRY_CONNECTION_COUNT` - how many times the connection to the Kafka broker is retried when it fails, Confluent Kafka producer and consumer are recreated on each retry
  * `RETRY_DB_TRANSACTION_COUNT` - how many times the DB transaction is retried when it fails
  * `KAFKA_SOCKET_KEEP_ALIVE_ENABLED` - whether the socket keep alive is enabled for the Kafka connection. Corresponds to `socket.keepalive.enable` in librdkafka.
  * `KAFKA_METADATA_MAX_AGE_MS` - the maximum age of the Kafka metadata in milliseconds. Corresponds to `metadata.max.age.ms` in librdkafka.
  * `KAFKA_PRODUCER_RETRIES` - how many times the librdkafka producer retries sending a message when it fails. Corresponds to `retries` in librdkafka.
  * `KAFKA_PRODUCER_TIMEOUT_MS` - the timeout for the librdkafka Kafka producer in milliseconds. Corresponds to `request.timeout.ms` in librdkafka.
  * `KAFKA_PRODUCER_PARTITIONER` - the partitioner used by the librdkafka Kafka producer. Corresponds to `partitioner` in librdkafka.
  * `KAFKA_COMPRESSION_CODEC` - the compression codec used by the librdkafka Kafka producer. Corresponds to `compression.codec` in librdkafka.
  * `KAFKA_CONSUMER_HEARTBEAT_INTERVAL_MS` - the heartbeat interval for the librdkafka Kafka consumer in milliseconds. Corresponds to `heartbeat.interval.ms` in librdkafka.
  * `KAFKA_CONSUMER_SESSION_TIMEOUT_MS` - the session timeout for the librdkafka Kafka consumer in milliseconds. Corresponds to `session.timeout.ms` in librdkafka.
  * `KAFKA_CONSUMER_MAX_POLL_INTERVAL_MS` - the maximum poll interval for the librdkafka Kafka consumer in milliseconds. Corresponds to `max.poll.interval.ms` in librdkafka.
  * `MAX_MESSAGES_TO_FETCH` - how many messages are fetched from the queue at once, default is 10

  For more info about the librdkafka configuration options see [https://github.com/confluentinc/librdkafka/blob/master/CONFIGURATION.md](https://github.com/confluentinc/librdkafka/blob/master/CONFIGURATION.md).

## \[3.5.3] - 2025-06-27

Rasa Pro Services 3.5.3 (2025-06-27)

### Bugfixes

* Handle `JsonPatchConflict` exceptions gracefully when loading pre-existing dialogue stack events or when applying current event patches.
  This prevents the Analytics service from crashing due to conflicts in the JSON patch operations.

## \[3.5.2] - 2025-05-21

Rasa Pro Services 3.5.2 (2025-05-21)

### Bugfixes

* Fixed a bug in Rasa Analytics for Session Creation Logic. A session can be created by events `slot` event with the slot `session_started_metadata` followed by `action` event for `action_session_start` OR just the event `action` for `action_session_start` alone.

## \[3.5.1] - 2025-05-12

Rasa Pro Services 3.5.1 (2025-05-12)

### Bugfixes

* Fix the processing of stack events by the analytics service in the case of multiple parallel conversations whose events
  are split across several batches consumed by the service.
* Replace sqlalchemy's `DateTime` type with postgresql dialect specific `TIMESTAMP` type in columns that record date and time.
  This is required to keep fractional seconds precision which is essential when retrieving `stack` events from the database
  to reconstruct the dialogue stack for each different conversation id.

## \[3.5.0] - 2025-03-20

Rasa Pro Services 3.5.0 (2025-03-20)

### Improvements

* MTS: Update MTS to train without the need for NFS.
* MRS: Update MRS to run with bot config and model trained from remote storage instead of using NFS.

### Bugfixes

* Updated `jinja2`, `werkzeug`, `idna`, `requests`, `zipp` and `urllib3` to address security vulnerabilities.

## \[3.4.1] - 2025-05-12

Rasa Pro Services 3.4.1 (2025-05-12)

### Bugfixes

* Updated `jinja2`, `werkzeug`, `idna`, `requests`, `zipp` and `urllib3` to address security vulnerabilities.
* Fix the processing of stack events by the analytics service in the case of multiple parallel conversations whose events
  are split across several batches consumed by the service.
* Replace sqlalchemy's `DateTime` type with postgresql dialect specific `TIMESTAMP` type in columns that record date and time.
  This is required to keep fractional seconds precision which is essential when retrieving `stack` events from the database
  to reconstruct the dialogue stack for each different conversation id.

## \[3.4.0] - 2024-12-12

Rasa Pro Services 3.4.0 (2024-12-12)

### Improvements

* Added Python 3.10 and Python 3.11 support to Rasa Analytics

### Bugfixes

* MTS and MRS: Add retry capability to Kubernetes API calls on `5xx` errors.

### Miscellaneous internal changes

*Miscellaneous internal changes.*

## \[3.3.5] - 2024-10-29

Rasa Pro Services 3.3.5 (2024-10-29)

### Bugfixes

* MTS and MRS: Add retry capability to Kubernetes API calls on `5xx` errors.

## \[3.3.4] - 2024-10-02

Rasa Pro Services 3.3.4 (2024-10-02)

### Bugfixes

* Return status code 503 when the Analytics service is unavailable during a healthcheck request.
  This allows the user to implement liveness probes that could automatically restart the service upon failure.

## \[3.3.3] - 2024-09-25

Rasa Pro Services 3.3.3 (2024-09-25)

### Bugfixes

* MTS: Fix bug (ATO-2257) when training pod's status is not caught when MTS consumer job restarts.
* \[MTS] Update certifi to 2023.7.22 to resolve vulnerability CVE-2023-37920.

## \[3.3.2] - 2024-05-28

Rasa Pro Services 3.3.2 (2024-05-28)

### Improvements

* MRS: Upload rasa pod logs to remote storage for user-friendly access to logs in case the running of the assistant fails.

## \[3.3.1] - 2024-05-27

Rasa Pro Services 3.3.1 (2024-05-27)

### Bugfixes

* Allow users to specify image pull secrets for MTS / MRS

## \[3.3.0] - 2024-04-03

Rasa Pro Services 3.3.0 (2024-04-03)

### Improvements

* Align column names representing the `flow_id` as defined in the yaml file across `rasa_flow_status`, `rasa_llm_command` and `rasa_dialogue_stack_frame` tables:
  * `rasa_llm_command` table: `flow_name` column has been renamed to `flow_identifier`.
  * `rasa_dialogue_stack_frame` table: `active_flow` column has been renamed to `active_flow_identifier`.
* MTS: Handle MTS Job Consumer restarts when rasa training pod continues to run.
  If the `TrainingManager` finds a running training job when it restarts, it will check the status of the pod:
  * If the pod is pending or running, it will watch the pod until training is complete.
  * If the pod has completed, it will upload the logs and trained model (only if it is present).
* MTS: Accept `nlu` in the config data of CALM assistants in order to use `nlu_triggers`.
* MTS and MRS: Support debug logs in rasa pods.

### Miscellaneous internal changes

*Miscellaneous internal changes.*

## \[3.2.3] - 2023-12-20

Rasa Pro Services 3.2.3 (2023-12-20)

### Improvements

* MTS: Setup alembic schema migration mechanism for the database of the model training orchestrator. Add initial table creation migration file.
* \[MTS] Add capability to configure log level for MTS orchestrator.
  Log level can be configured through `LOG_LEVEL` environment variable.
  Default log level is `INFO`.

### Bugfixes

* Fix telemetry reporting in shipped Docker images.

## \[3.2.2] - 2023-12-05

Rasa Pro Services 3.2.2 (2023-12-05)

### Bugfixes

* Remove obsolete component RemoteGCSFetcher from MTS orchestrator.

## \[3.2.1] - 2023-12-01

Rasa Pro Services 3.2.1 (2023-12-01)

### Improvements

* Align column names representing the `flow_id` as defined in the yaml file across `rasa_flow_status`, `rasa_llm_command` and `rasa_dialogue_stack_frame` tables:
  * `rasa_llm_command` table: `flow_name` column has been renamed to `flow_identifier`.
  * `rasa_dialogue_stack_frame` table: `active_flow` column has been renamed to `active_flow_identifier`.
* Add environment variables to control resource requirements and limits for Rasa pod.
  MTS and MRS job consumers can now be configured to use specify resource requirements and limits for the Rasa pod.
  This can be done by setting the following environment variables in the Rasa pod:
  * RASA\_REQUESTS\_CPU
  * RASA\_REQUESTS\_MEMORY
  * RASA\_LIMITS\_CPU
  * RASA\_LIMITS\_MEMORY

## \[3.2.0] - 2023-11-22

Rasa Pro Services 3.2.0 (2023-11-22)

### Features

* Added new table `rasa_dialogue_stack_frame` to store active flow names and steps for each event sequence in the conversation.
* Add new table `rasa_llm_command` to store LLM generated commands for each user message.
  Add new column in the `_rasa_raw_event` table to store the serialized LLM generated commands.
* Add new table `rasa_flow_status` to store the transformations of rasa flow events.
  Add new columns in the `_rasa_raw_event` table to store the flow\_id and step\_id of these events where applicable.

## \[3.1.1] - 2023-07-17

Rasa Pro Services 3.1.1 (2023-07-17)

### Miscellaneous internal changes

*Miscellaneous internal changes.*

## \[3.1.0] - 2023-07-03

Rasa Pro Services 3.1.0 (2023-07-03)

### Features

* Added Real Time Processing of Markers.
  Markers can now be evaluated real time by the Analytics Data Pipeline. We've added event handlers for evaluation all events from Kafka to extract markers. The extracted markers are saved into `rasa_marker` database table. These markers are evaluated with the patterns stored in `rasa_pattern` table.

  Added API endpoints to create patterns in `rasa_pattern` table. This endpoint is used by Rasa Plus for `rasa markers upload` command.

### Miscellaneous internal changes

*Miscellaneous internal changes.*

## \[3.0.2] - 2023-06-13

Rasa Pro Services 3.0.2 (2023-06-13)

### Improvements

* Adds an environment variable to control logging level of the application.

### Miscellaneous internal changes

*Miscellaneous internal changes.*

## \[3.0.1] - 2022-10-26

Rasa Pro Services 3.0.1 (2023-10-26)

### Miscellaneous internal changes

*Miscellaneous internal changes.*

## \[3.0.0] - 2022-10-24

Rasa Pro Services 3.0.0 (2023-10-24)

### Features

* Analytics Data Pipeline helps visualize and process Rasa assistant metrics in the tooling (BI tools, data warehouses) of your choice. Visualizations and analysis of the production assistant and its conversations allow you to assess ROI and improve the performance of the assistant over time.


## Related topics

- [Rasa Pro Change Log](/docs/reference/changelogs/rasa-pro-changelog.md)
- [Studio Change Log](/docs/reference/changelogs/studio-changelog.md)