deepakpanda93 commented on issue #10309:
URL: https://github.com/apache/hudi/issues/10309#issuecomment-4845921621

   This is addressed in Hudi 1.x.
   
   There were two separate problems in this thread:
   
   - The write-side error (cannot update origin type double to long) reported 
by @lei-su-awx was fixed in 0.14.1 (confirmed above).
   - The OP's read-side failure — reading old parquet files physically stored 
as INT64 after evolving the column to double 
(SchemaColumnConvertNotSupportedException: Expected double, Found INT64) — 
happened because on 0.14/0.15 the read went through Spark's native vectorized 
parquet reader, which doesn't promote INT64→double.
   
   In 1.x this is handled by the engine-agnostic FileGroup reader (default 
since 1.0.0, hoodie.file.group.reader.enabled=true), which adds 
schema-evolution support and disables the vectorized reader for unsupported 
type cases, with numeric type-promotion fixes:
   
   - HUDI-7567 [#10957](https://github.com/apache/hudi/pull/10957) (schema 
evolution in the filegroup reader, 1.0.0)
   - HUDI-7068 [#10043](https://github.com/apache/hudi/pull/10043) (disable 
vectorized read when schema not supported, 1.0.0)
   - HUDI-8902 [#13188](https://github.com/apache/hudi/pull/13188) 
(float→double read correctness, 1.0.2)
   - HUDI-9705 [#13714](https://github.com/apache/hudi/pull/13714) 
(type-promotion/field-rename reader fixes, 1.1.0)
   
   long → double is a valid promotion, so reading the old long-typed files back 
as double works on 1.x. Your 0.15.0 test predates the FileGroup reader becoming 
the default, which is why it still failed there.
   
   Please retest on a recent 1.1.x release (with 
hoodie.schema.on.read.enable=true and the file group reader default). Closing 
as fixed in 1.x and  if you still hit Expected double, Found INT64 on 1.1.x, 
please reopen with a minimal repro. Thanks!


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to