Copy data from Amazon S3 to Redshift and avoid duplicate rows - amazon-web-services

Question

I am copying data from Amazon S3 to Redshift. During this process, I need to avoid the same files being loaded again. I don't have any unique constraints on my Redshift table. Is there a way to implement this using the copy command?
http://docs.aws.amazon.com/redshift/latest/dg/r_COPY_command_examples.html
I tried adding unique constraint and setting column as primary key with no luck. Redshift does not seem to support unique/primary key constraints.

Answer 1

As user1045047 mentioned, Amazon Redshift doesn't support unique constraints, so I had been looking for the way to delete duplicate records from a table with a delete statement.
Finally, I found out a reasonable way.
Amazon Redshift supports creating an IDENTITY column that is stored an auto-generated unique number.
http://docs.aws.amazon.com/redshift/latest/dg/r_CREATE_TABLE_NEW.html
The following sql is for PostgreSQL to delete duplicated records with OID that is unique column, and you can use this sql by replacing OID with the identity column.
DELETE FROM duplicated_table WHERE OID > (
　SELECT MIN(OID) FROM duplicated_table d2
　　WHERE column1 = d2.dupl_column1
　　AND column2 = d2.column2
);
Here is an example that I tested on my Amazon Redshift cluster.
create table auto_id_table (auto_id int IDENTITY, name varchar, age int);
insert into auto_id_table (name, age) values('John', 18);
insert into auto_id_table (name, age) values('John', 18);
insert into auto_id_table (name, age) values('John', 18);
insert into auto_id_table (name, age) values('John', 18);
insert into auto_id_table (name, age) values('John', 18);
insert into auto_id_table (name, age) values('Bob', 20);
insert into auto_id_table (name, age) values('Bob', 20);
insert into auto_id_table (name, age) values('Matt', 24);
select * from auto_id_table order by auto_id;
auto_id | name | age
---------+------+-----
1 | John | 18
2 | John | 18
3 | John | 18
4 | John | 18
5 | John | 18
6 | Bob | 20
7 | Bob | 20
8 | Matt | 24
(8 rows)
delete from auto_id_table where auto_id > (
select min(auto_id) from auto_id_table d
where auto_id_table.name = d.name
and auto_id_table.age = d.age
);
select * from auto_id_table order by auto_id;
auto_id | name | age
---------+------+-----
1 | John | 18
6 | Bob | 20
8 | Matt | 24
(3 rows)
Also it works with COPY command like this.
auto_id_table.csv
John,18
Bob,20
Matt,24
copy sql
copy auto_id_table (name, age) from '[s3-path]/auto_id_table.csv' CREDENTIALS 'aws_access_key_id=[your-aws-key-id] ;aws_secret_access_key=[your-aws-secret-key]' delimiter ',';
The advantage of this way is that you don't need to run DDL statements. However it doesn't work with existing tables that do not have an identity column because an identity column cannot be added to an existing table. The only way to delete duplicated records with existing tables is migrating all records like this. (same as user1045047's answer)
insert into temp_table (select distinct from original_table);
drop table original_table;
alter table temp_table rename to original_table;

Answer 2

Mmm..
What about just never loading data into your master table directly.
Steps to avoid duplication:
begin transaction
bulk load into a temp staging table
delete from master table where rows = staging table rows
insert into master table from staging table (merge)
drop staging table
end transaction.
this is also super somewhat fast, and recommended by redshift docs.

Answer 3

My solution is to run a 'delete' command before 'copy' on the table. In my use case, each time I need to copy the records of a daily snapshot to redshift table, thus I can use the following 'delete' command to ensure duplicated records are deleted, then run the 'copy' command.
DELETE from t_data where snapshot_day = 'xxxx-xx-xx';

Answer 4

Currently there is no way to remove duplicates from redshift. Redshift doesn't support primary key/unique key constraints, and also removing duplicates using row number is not an option (deleting rows with row number greater than 1) as the delete operation on redshift doesn't allow complex statements (Also the concept of row number is not present in redshift).
The best way to remove duplicates is to write a cron/quartz job that would select all the distinct rows, put them in a separate table and then rename the table to your original table.
Insert into temp_originalTable (Select Distinct from originalTable)
Drop table originalTable
Alter table temp_originalTable rename to originalTable

Answer 5

There's another solution to really avoid data duplication although it's not as straightforward as removing duplicated data once inserted.
The copy command has the manifest option to specify which files you want to copy
copy customer
from 's3://mybucket/cust.manifest'
iam_role 'arn:aws:iam::0123456789012:role/MyRedshiftRole'
manifest;
you can build a lambda that generates a new manifest file every time before you run the copy command. That lambda will compare the files already copied with the new files arrived and will create a new manifest with only the new files so that you will never ingest the same file twice

Answer 6

We remove duplicates weekly, but you could also do this during the load transaction as mentioned by #Kyle. Also, this does require the existence of an autogenerated ID column as an eventual target of the delete :
DELETE FROM <your table> WHERE ID NOT IN (
SELECT ID FROM (
SELECT *, ROW_NUMBER() OVER
( PARTITION BY <your constraint columns> ORDER BY ID ASC ) DUPLICATES
FROM REQUESTS
) WHERE DUPLICATES=1
); COMMIT;
for example:
CREATE TABLE IF NOT EXISTS public.requests
(
id BIGINT NOT NULL DEFAULT "identity"(1, 0, '1,1'::text) ENCODE delta
kaid VARCHAR(50) NOT NULL
,eid VARCHAR(50) NOT NULL ENCODE text32k
,aid VARCHAR(100) NOT NULL ENCODE text32k
,sid VARCHAR(100) NOT NULL ENCODE zstd
,rid VARCHAR(100) NOT NULL ENCODE zstd
,"ts" TIMESTAMP WITHOUT TIME ZONE NOT NULL ENCODE delta32k
,rtype VARCHAR(50) NOT NULL ENCODE bytedict
,stype VARCHAR(25) ENCODE bytedict
,sver VARCHAR(50) NOT NULL ENCODE text255
,dmacd INTEGER ENCODE delta32k
,reqnum INTEGER NOT NULL ENCODE delta32k
,did VARCHAR(255) ENCODE zstd
,"region" VARCHAR(10) ENCODE lzo
)
DISTSTYLE EVEN
SORTKEY (kaid, eid, aid, "ts")
;
. . .
DELETE FROM REQUESTS WHERE ID NOT IN (
SELECT ID FROM (
SELECT *, ROW_NUMBER() OVER
( PARTITION BY DID,RID,RTYPE,TS ORDER BY ID ASC ) DUPLICATES
FROM REQUESTS
) WHERE DUPLICATES=1
); COMMIT;