Purging Vertica tables/ partitions at regular intervals

First thing first, especially, when there is potential to misunderstand due to definitions of specific words – purging vs truncating vs deleting data. In this article, my interest is to purge the tables of deleted rows.

Truncate => Removes all storage associated with a table, while preserving the table definitions.
Purge => Permanently removes deleted data from physical storage so that the disk space can be reused.  You can purge historical data up to and including the epoch in which the Ancient History Mark is contained.
Delete => It marks tuples as no longer valid in the current epoch. It does not delete data from disk storage for base tables.

Vertica purge_table or purge_partitions statement purges all projections of the specified table and can temporarily take significant disk space while performing the purge.

Following query returns list of queries that when executed purges data for each partitioned tables.

   — Look into only large tables (TABLE_SIZE_OF_INTEREST), say 1Billion row,
   —  that have good chunk deletes (>20%)
   SELECT ‘SELECT purge_partition(”’|| P.projection_schema ||’.’|| P.anchor_table_name || ”’,”’||PRT.partition_key||”’);’
    FROM projections P inner join
    (   SELECT projection_id, partition_key
        ,SUM(deleted_row_count)
        ,SUM(ros_row_count) as total_rows
        FROM partitions
        GROUP BY 1,2
        HAVING SUM(deleted_row_count)/SUM(ros_row_count) > 0.2
    ) PRT
    ON PRT.projection_id = P.projection_id
    GROUP BY PRT.partition_key, P.projection_schema, P.anchor_table_name
    HAVING max(total_rows) <= {TABLE_SIZE_OF_INTEREST}
    ORDER BY max(total_rows) DESC
    LIMIT 25 ;

Similarly for those tables that are non-partitioned use following query.

    SELECT ‘select purge_table(”’|| projection_schema ||’.’|| anchor_table_name || ”’);’
    FROM projections P inner join
    (   SELECT schema_name, projection_name ,
            sum(deleted_row_count) delete_rows ,
            sum(delete_vector_count) delete_vector_count ,
            sum(total_row_count) total_rows,
            sum(deleted_row_count)/sum(total_row_count) as delete_percent
        FROM storage_containers
        GROUP BY 1,2
        HAVING sum(deleted_row_count)/sum(total_row_count) > 0.2
    ) A
    ON A.schema_name = P.projection_schema
      AND A.projection_name = P.projection_name
    INNER JOIN tables T on P.anchor_table_id = T.table_id
    WHERE length(T.partition_expression) = 0
    GROUP BY 1
    HAVING max(total_rows) <= {TABLE_SIZE_OF_INTEREST}
    ORDER BY max(total_rows) DESC
    LIMIT 25 ;

With a wrapper around these queries in Python/ Perl/ Bash, etc one can easily go through the list executing each statement and cleaning up old deleted data for improved performance.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s