row_number is not unique for duplicate records-CodePudding

I am trying to find the latest update of a particular row from a bunch of rows. For that we use row_number() over a partition as shown below,

"row_number"() OVER (
    PARTITION BY "uuid"
    ORDER BY "uuid" ASC,
        "status"."sequence" DESC,
        "modifiedon" DESC
) "row_id_ranked",

After this, anything other than 'row_id_ranked = 1' are discarded. But when a retry/re-transmission happens, "status"."sequence" and "modifiedon" will be exactly same. This creates mutiple rows with row_id_ranked=1. Making it difficult to remove them. As per my understanding per the documentation, the row_number should be a unique value. But in case of these duplicates, it clearly is not.

row_number() → bigint#

Returns a unique, sequential number for each row, starting with one, according to the ordering of rows within the window partition.

How do I de-duplicate in this case?

CodePudding user response：

Assuming that there does not exist some third column beyond the status sequence and modified on date to break a potential tie, and also assuming you don't care which single record gets retained, you may use RANDOM() here:

ROW_NUMBER() OVER (PARTITION BY uuid
                   ORDER BY "status"."sequence" DESC, modifiedon DESC, RANDOM())
    AS row_id_ranked

CodePudding user response：

If you use "Partition by uuid" in the "over" clause of row number you will get a row 1 for each distinct uuid. That is you have partitioned your data by uuid and the numbering will restart for each partition, as in the definition you have posted.
I think that you just want to use order by, which will return 1 row number 1.

"row_number"() OVER (
    ORDER BY "uuid" ASC,
        "status"."sequence" DESC,
        "modifiedon" DESC
) "row_id_ranked",