Incorrect replace operation in compaction strategy filter

There are some incorrect replace operation to sort all partition paths.
{code:java}
return allPartitionPaths.stream().map(partition -> partition.replace("/", "-"))
.sorted(Comparator.reverseOrder()).map(partitionPath -> partitionPath.replace("-", "/")) {code}
the hive partition before replace is dllr_date=2023-10-10, then after will convert to dllr_date=2023/10/10, this is an incorrect partition.
# org.apache.hudi.table.action.compact.strategy.DayBasedCompactionStrategy
# org.apache.hudi.table.action.compact.strategy.BoundedPartitionAwareCompactionStrategy
# org.apache.hudi.table.action.compact.strategy.UnBoundedPartitionAwareCompactionStrategy

!image-2023-11-08-16-02-39-291.png!

## JIRA info

- Link: https://issues.apache.org/jira/browse/HUDI-7051
- Type: Bug
- Attachment(s):
- 08/Nov/23 08:01;vmaster;image-2023-11-08-16-01-46-166.png;https://issues.apache.org/jira/secure/attachment/13064245/image-2023-11-08-16-01-46-166.png
- 08/Nov/23 08:02;vmaster;image-2023-11-08-16-02-39-291.png;https://issues.apache.org/jira/secure/attachment/13064244/image-2023-11-08-16-02-39-291.png

---

## Comments

06/Dec/23 03:39;shivnarayan;hey [~vmaster] : 
sorry I am bit confused. 

as per master, filterPartitionPaths in DayBasedCompactionStrategy is as below 

 
{code:java}
@Override
public List filterPartitionPaths(HoodieWriteConfig writeConfig, List allPartitionPaths) {
return allPartitionPaths.stream().sorted(comparator)
.collect(Collectors.toList()).subList(0, Math.min(allPartitionPaths.size(),
writeConfig.getTargetPartitionsPerDayBasedCompaction()));
} {code}
 

 

Only in 

BoundedPartitionAwareCompactionStrategy.filterPartitionPaths I see the replace operations. 

But can you help me understand whats the issue in there. I understand "dllr_date=2023/10/10" may not be an actual partition present physcially, but thats interim state used for comparison and later we switch it back. 

 

in other words. 

if original partition is hypehnated. 

 

dllr_date=2023-10-10 -> gets converted to "dllr_date=2023/10/10", and then comparisons are performed to sort them and then converted back to dllr_date=2023-10-10. So, not sure where is the bug here. can you throw some light please

 ;;;

---

02/Jan/24 01:53;vmaster;[~shivnarayan] thanks for your reply, as you say, the class of DayBasedCompactionStrategy has been fixed by issue HUDI-6975, but problem still exists in follow class:
# org.apache.hudi.table.action.compact.strategy.BoundedPartitionAwareCompactionStrategy
# org.apache.hudi.table.action.compact.strategy.UnBoundedPartitionAwareCompactionStrategy

if we have a hive partition like this: dllr_date=2023-10-10 ,the replace operation will result in an error, final get 'dllr_date=2023/10/10'

 
{code:java}
List allPartitionPaths =
partitionPaths.stream().map(partition -> partition.replace("/", "-")).sorted(Comparator.reverseOrder())
.map(partitionPath -> partitionPath.replace("-", "/")).collect(Collectors.toList()); {code}
in other words, there has second replace operation, but the first will has no effect, only the second replace works, as this point, there is no converted back.

 

 ;;;

Start at filterPartitionPaths in BoundedPartitionAwareCompactionStrategy and UnBoundedPartitionAwareCompactionStrategy, then compare their path handling with DayBasedCompactionStrategy after HUDI-6975. Trace the example partition path dllr_date=2023-10-10 through sorting and verify that filtering preserves the original valid partition path rather than producing dllr_date=2023/10/10.

Written by the indexing model from the issue text.