1. Phone recognizer

In Saga 1.3.1 the Phone Recognizer now has a dictionary of Area codes that the user can fill and then use to filter phones by area code.

Due to this addition the mapping in the index for this recognizer changed in Saga 1.3.1.

In order to fix the problem just remove the Phone Recognizer and add it again. When adding it again it will be saved with the correct mapping in the index.

In Saga 1.3.3 the Phone Recognizer now supports area codes for US and UK.  This means the area code dictionary now contains an additional column for country code and consequently the corresponding index and index mapping in Elasticsearch was modified to include that field.  To update from a previous version the following needs to be done:

1. Shut down Saga
2. In Elasticsearch/OpenSearch backup the saga_area_codes index
3. Remove the saga_area_codes index
4. Bring Saga up.  Saga will recreate the area codes index with the new mapping.
5. Create a new Phone Number Recognizer
6. Copy the contents of the old area codes index into the new one and also add the value for country code in the country column.  Right now it's either "US" or "UK".

2. Geonames Recognizer

Geonames index mapping changed in Saga 1.3.1. The field "id" changed from Integer to Keyword. This makes resources not to load for geonames, rest handler not to load in SagaServer and therefore the recognizer won't be displayed in the UI.

In order to fix this issue you can remove the geonames index by using DELETE  http request using Kibana or something like Postman. SO if using "saga" indices prefix you'll need to do a DELETE saga_geonames. Then in the UI add the recognizer back to the tag and reload the dictionary.

3. New field in Processing Units Index

There is a new field in the Processing Units index. This index is used to stored persistent Processing Units. Therefore if you have a persistent processing unit in Saga 1.3 and then migrate to Saga 1.3.1+ you'll get an error in the log when starting Saga.

Due to this error, Processing Unit endpoints to create, list or delete units won't be loaded therefore not available and will return a not found error code when trying calling those endpoints.

4. New Bert Model handling on the Python-Bridge

To tackle vulnerabilities regarding old python libraries, we did an update on all the requirements for executing the SAGA Python Bridge, one of this changes affects directly how the BERT models work, we had to change the way the wrappers for these BERT models calculate and extract vectors from text. This change is a breaking change because this makes ALL the vectors from the old models USELESS, the reason is that the vectors are now calculated different from the old versions and they will not behave the same when using the new Python Bridge.

The solution could be as easy as just re-calculating all the vectors using SAGA (all the recognizers that use vectors have the button "recalculate" in order to send the entries to the Python-Bridge and store new calculated vectors) or very difficult and time-consuming like if you have the vectors stored in an outside source from SAGA, you would need to re-calculate all the vectors "manually" and then replace all the old vectors in order to use them with the new BERT models.