Affects Version/s: 2.1.10 (Lovelace SR10)
I discussed with Mark Paluch and I created a minimal sample to reproduce the issue. Mark already stated they use a LinkedHashSet and thus the order will preserve in CassandraPersistentEntitySchemaCreator. However, the ** specifications variable returns different orders as shown in the screenshot.
I have to upload the example since JIRA tells me "JIRA could not attach the file as there was a missing token. Please try attaching the file again."
link to java example - https://share.infinkon.de/index.php/s/jKiSjbokw9LQTnZ
- correct UDT order https://share.infinkon.de/index.php/s/smd5LHfF2wQ9fsT
- wrong UDT order https://share.infinkon.de/index.php/s/M6EjQCNKXyoRYJ2
You can simply extract it and start it in Eclipse "ProductApplication" with a local Cassandra instance.
Sometimes, the application will start and the keyspaces will be created and in cqlsh "describe types" will show me in keyspace inproduct all expected UDTs:
benefitudt documentmetadataudt spproductudt inproductudt
It will take usually less than 10 tried executions to reproduce the issue and show the error.
It is mandatory that documentmetadataudt is created as first UDT. If this is not the case then no UDT is created.