[ 
https://issues.apache.org/jira/browse/CASSANALYTICS-217?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Jyothsna Konisa reassigned CASSANALYTICS-217:
---------------------------------------------

    Assignee: Jyothsna Konisa

> Non-frozen UDTs with unset fields or partial updates fail to deserialize in 
> bulk reader and CDC
> -----------------------------------------------------------------------------------------------
>
>                 Key: CASSANALYTICS-217
>                 URL: https://issues.apache.org/jira/browse/CASSANALYTICS-217
>             Project: Apache Cassandra Analytics
>          Issue Type: Bug
>            Reporter: Jyothsna Konisa
>            Assignee: Jyothsna Konisa
>            Priority: Normal
>
> A non-frozen UDT stores one cell per field, so a field that wasn't written 
> has no cell. This happens after a partial update (UPDATE t SET udt.field = 
> value) or a full write that leaves out a field (INSERT ... {status: 'X'}). 
> UdtBuffer packs the cells in order without their field positions,so 
> CqlUdt.deserializeUdt fails:                                                  
>                                                                               
>                                                                               
>                                                      
>                                                                               
>                                                                               
>                                                                               
>                                                           
> IllegalArgumentException: Unexpected number of fields deserializing UDT 
> 'single_result', expected 2 fields but 1 found                                
>                                                                               
>                                                               
>                                                                               
>                                                                               
>                                                                               
>                                                           
>   Impact:                                                                     
>                                                                               
>                                                                               
>                                                           
>   - Bulk reader: the Spark job aborts when it reads such a row.               
>                                                                               
>                                                                               
>                                                           
>   - CDC: the event is retried forever and blocks the partition.               
>                                                                               
>                                                                               
>                                                           
>   - CDC: SET udt.field = null publishes the whole column as null, so 
> consumers can't tell which field was cleared.                                 
>                                                                               
>                                                                    
>                                                                               
>                                                                               
>                                                                               
>                                                           
>   Fix:                                                                        
>                                                                               
>                                                                               
>                                                           
>   - UdtBuffer places each cell at the field position from its cell path and 
> packs missing fields as null. This fixes both the bulk reader and CDC.        
>                                                                               
>                                                             
>   - CDC updateFields lists the fields written by a partial update as 
> udt.field (e.g. result.code); a full write still lists the column.            
>                                                                               
>                                                                    
>   - UDT fields in CDC Avro payload schemas are now nullable, and the envelope 
> version is bumped to "3".



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to