views:

66

answers:

2

I have a table user_interactions with 4 columns:

 user_1
 user_2
 type
 timestamp

The primary key is (user_1,user_2,type)
and I want to change to (user_2,user_1,type)

So what I did was :

drop primary key ...  
add primary key (user_2,user_1,type)...

and voila...

The problem is that database is live on a server.

So before I could update the primary key, many duplicates already crept in, and they are continuously creeping in.

What to do?

What I want to do now is to remove duplicates and keep the ones with the latest timestamp (which is a column in the table).

And then somehow update the primary key again.

A: 

You can use the IGNORE keyword too, example:

 update IGNORE table set primary_field = 'value'...............
Sarfraz
+1  A: 

Next time, use a single "alter table" statement to update the primary key.

alter table xx drop primary key, add primary key(k1, k2, k3);

To fix things:

create table fixit (user_2, user_1, type, timestamp, n, primary key( user_2, user_1, type) );
lock table fixit write, user_interactions u write, user_interactions write;

insert into fixit 
select user_2, user_1, type, max(timestamp), count(*) n from user_interactions u 
group by user_2, user_1, type
having n > 1;

delete u from user_interactions u, fixit 
where fixit.user_2 = u.user_2 
  and fixit.user_1 = u.user_1 
  and fixit.type = u.type 
  and fixit.timestamp != u.timestamp;

alter table user_interactions add primary key (user_2, user_1, type );

unlock tables;

The lock should stop further updates coming in while your are doing this. How long this takes obviously depends on the size of your table.

The main problem is if you have some duplicates with the same timestamp.

Martin
You are great !I wish I could thank you enough.
dta