nsivabalan commented on code in PR #10915:
URL: https://github.com/apache/hudi/pull/10915#discussion_r1600668418
##########
hudi-spark-datasource/hudi-spark/src/test/scala/org/apache/hudi/functional/cdc/TestCDCDataFrameSuite.scala:
##########
@@ -407,27 +409,32 @@ class TestCDCDataFrameSuite extends HoodieCDCTestBase {
val inputDF7 = spark.read.json(spark.sparkContext.parallelize(records7, 2))
inputDF7.write.format("org.apache.hudi")
.options(options)
+ .option("hoodie.compact.inline", "false")
.mode(SaveMode.Append)
.save(basePath)
+ totalInsertedCnt += 7
val records8 = recordsToStrings(dataGen.generateInserts("007",
3)).asScala.toList
val inputDF8 = spark.read.json(spark.sparkContext.parallelize(records8, 2))
inputDF8.write.format("org.apache.hudi")
.options(options)
+ .option("hoodie.compact.inline", "false")
.mode(SaveMode.Append)
.save(basePath)
val instant8 = metaClient.reloadActiveTimeline.lastInstant().get()
val commitTime8 = instant8.getTimestamp
+ totalInsertedCnt += 3
// 8. Upsert Operation With Clean Operation
- val records9 = recordsToStrings(dataGen.generateUniqueUpdates("008",
30)).asScala.toList
- val inputDF9 = spark.read.json(spark.sparkContext.parallelize(records9, 2))
+ val inputDF9 = inputDF6.limit(30) // 30 updates to inserts added after
insert overwrite table. if not for this, updates generated from datagne,
+ // could split as inserts and updates from hudi standpoint due to insert
overwrite table operation.
inputDF9.write.format("org.apache.hudi")
.options(options)
.option("hoodie.clean.automatic", "true")
- .option("hoodie.keep.min.commits", "4")
- .option("hoodie.keep.max.commits", "5")
- .option("hoodie.clean.commits.retained", "3")
+ .option("hoodie.keep.min.commits", "16")
Review Comment:
hey @bvaradar : I had a long discussion w/ @danny0405 on this.
here is the issue:
our data table archival polls for compaction commit and does something based
on that. Before this patch, we had a bug no canSchedule call wrt compaction
scheduling. This patched fixed the issue as you might know.
But this specific test started to fail w/ FileNotFound. Apparently in
HoodieCDCExtractor, we parse all entries in commit metadata and [poll file
status/listStatus](https://github.com/apache/hudi/blob/c8dec0ef523e998b80838af9d52323e3ae95cebf/hudi-common/src/main/java/org/apache/hudi/common/table/cdc/HoodieCDCExtractor.java#L331)
for them.
Prior to this patch, due to archival behavior, every commit in active
timeline is uncleaned and we were good. after this patch, apparently archival
is trailing, and so we do have some commits in active timeline which are
cleaned up and hence the fileStatus/listStatus polling results in FileNotFound
Issue.
Danny agreed that this is a long pending/known limitation and we could
definitely improve user exp by throwing some legible exception/error msg. But
it increases the scope of this patch. so, we agreed to tweak the test so that
it will not hit FileNotFound issue.
--
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]
