In case of Positional Deletes "file_path" in lowerbound & upperbound do not have full file_path Buffer

[agrawalreetika]

Apache Iceberg version

None

Query engine

PrestoDB

Please describe the bug 🐞

I am trying to introduce Delete File pruning based on the file_path stored as part of lowerbound & upperbound in case of Positional Delete File for PrestoDB Engine.
While experimenting with this, I observed "file_path" in lowerbound & upperbound do not have full path Buffer.

Refernce - https://github.com/prestodb/presto/blob/master/presto-iceberg/src/main/java/com/facebook/presto/iceberg/delete/DeleteFile.java#L48

While the actual file path looks like this -

[singhpk234]

which iceberg version and what is the writer used ?

seems like the file_path metrics were truncated :

This is because these metrics were truncated, Iceberg's default metrics mode for column metric is truncate(16).

This should be fixed by #6313.

more on this via : #6694

[agrawalreetika]

Thank you for your response @singhpk234
So we are using iceberg version 1.4.3 in Presto.
And I checked #6313 looks like it's fixed for Flink, so it wouldn't work for others?

[singhpk234]

are you using presto as a writer as well ? if you check it's still 16 characters so the writer which is generating the delete files is generating wrong metrics i.e still using truncate(16).

Essentially the fix here is to use this API :

MetricsConfig.forPositionDelete(table)

rather than the table default so if presto is the writer here as well then we should use this in presto, as per my understanding.

[singhpk234]

seems like we always get default config setting in presto

https://github.com/prestodb/presto/blob/c5674d6e7c1008eb7935542a836f8f90d6bf09c6/presto-iceberg/src/main/java/com/facebook/presto/iceberg/IcebergParquetFileWriter.java#L74C1-L74C194

even for deletes we need to have special handling here i think :), if i am not mistaken it would be interesting to see if it causes regression for already released version of presto which are being used as a writer it would just end up reading more data i guess , not sure though

[agrawalreetika]

Thanks for your input @singhpk234 I will explore around the write from Presto side to see where truncate is happening as part of write.