[HUDI-5839] Insert in non-strict mode deduplices dataset in "append" mode - spark

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 0.13.0
• Component/s: spark, writer-core
• Assignee: Unassigned
• Reporter: kazdy
• Created: 23/Feb/23 15:38
• Updated: 25/Feb/23 10:10

Description

There seem to be a bug with non-strict insert mode when precombine is not defined (but I have not checked for when it is).
When using spark datasource it can insert duplicates only in overwrite mode or append mode when data is inserted to the table for the first time, but if I want to insert in append mode for the second time it deduplicates the dataset as if it was working in upsert mode. Found in master (0.13.0).

I happens to be a regression, because I'm using this functionality in Hudi 0.12.1.

from pyspark.sql.functions import expr

path = "/tmp/huditbl"

opt_insert = {
    'hoodie.table.name': 'huditbl',
    'hoodie.datasource.write.recordkey.field': 'keyid',
    'hoodie.datasource.write.table.name': 'huditbl',
    'hoodie.datasource.write.operation': 'insert',
    'hoodie.sql.insert.mode': 'non-strict',
    'hoodie.upsert.shuffle.parallelism': 2,
    'hoodie.insert.shuffle.parallelism': 2,
    'hoodie.combine.before.upsert': 'false',
    'hoodie.combine.before.insert': 'false',
    'hoodie.datasource.write.insert.drop.duplicates': 'false'
}

df = spark.range(0, 10).toDF("keyid") \
  .withColumn("age", expr("keyid + 1000"))

df.write.format("hudi"). \
options(**opt_insert). \
mode("overwrite"). \
save(path)

spark.read.format("hudi").load(path).count() # returns 10

# df = df.union(df) # creates duplicates
df.write.format("hudi"). \
options(**opt_insert). \
mode("append"). \
save(path)

spark.read.format("hudi").load(path).count() # returns 10 but should return 20 

# note
# this works:
df = df.union(df) # creates duplicates 
df.write.format("hudi"). \ 
options(**opt_insert). \ 
mode("overwrite"). \ 
save(path)

spark.read.format("hudi").load(path).count() # returns 20 as it should