Tuesday, July 14, 2026
2 changes · 19.0
Enhancements to existing features
This update strengthens marketing automation behavior around campaign synchronization, message failures, and user-triggered activities such as opens and clicks. It fixes scheduling and trace handling issues across email, SMS, and WhatsApp journeys, helping campaigns behave more consistently as they scale.
Original PR description
RATIONALE In order to prepare upcoming improvements for marketing automation application, as well as performance improvements, some tests are added to improve coverage and cover some synchronization…
RATIONALE
In order to prepare upcoming improvements for marketing automation
application, as well as performance improvements, some tests are added
to improve coverage and cover some synchronization use cases.
SPECIFICATIONS
Add some tests improve coverage of synchronization, as we recently
discovered limitations
* "opposite" triggers when checking brother traces to skip in
'action_update_participants' (which synchronizes traces): sub
addons (sms, whatsapp) is not taken into account;
* 'schedule_date' is not correct for user-based activities (e.g.
mail_open, mail_click, ...) when new activities are added to
a campaign. They should not have scheduled dates, as it depends
on user action. Date is correct for activities when participants
enter child activities but not when doing the synchronize;
* add some checks on participant state;
* globally try to improve some corner cases coverage;
Add some tests to improve coverage of bounce / fail behavior with
various activities, as we want to make it clearer how MA should
behave when dealing with issue. First step is to assert current
behavior and fix some odd bits.
Notably in some cases trace update is missing, notably with SMS
sending with does not call trace update method, which means some
triggers are not processed.
Various fixes are included in this branch, spotted by newly added tests.
See commits for more details.
Task-4224152: [marketing_automation] Performance / ScalabilityThe AI module now moves away from an embedding model that will soon be retired. It also refreshes existing AI data created with deprecated models, helping keep AI-powered features reliable without manual intervention.
Original PR description
The model gemini-embedding-001 will be deprecated soon, so this commit is: - Deprecating that model. - Introducing an autovacuum to recompute embeddings that were done using embedding models that have been deprecated. task-6004597