Re: upsert functionality

On May 15, 2013, at 11:52 PM, Thomas Kellerer  wrote:

> Sajeev Mayandi, 16.05.2013 07:01:
>> Hi,
>> 
>> Our company is planning to move to postreSQL. We were initially using
>> sybase where upsert functionality was available using "insert on
>> existing update" clause.  I know there multiple ways to fix this
>> using RULE or separate function in postgresql.  But I would like to
>> know which version of postgresql has support for upsert planned using
>> an official syntax. I have postgresql 9.2 which does not have this
>> feature, if its planned in near future, I would rather wait to
>> migrate to PostgreSQL.
>> 
> 
> You can use writeable CTEs for this purpose.
> 
> There are several examples out there:
> 
> http://www.xzilla.net/blog/2011/Mar/Upserting-via-Writeable-CTE.html
> http://www.depesz.com/2011/03/16/waiting-for-9-1-writable-cte/
> http://www.depesz.com/2012/06/10/why-is-upsert-so-complicated/
> 
> http://stackoverflow.com/a/8702291/330315

One thing I didn't see mentioned in two of the links -- they mention race
conditions, where multiple writers can still cause the faked UPSERT to fail.

This can be avoided using SERIALIZABLE transactions, now that Postgres has
SSI.  http://wiki.postgresql.org/wiki/SSI

I can vouch that we use writable CTEs and SERIALIZABLE to implement UPSERT
in production with no issues thus far.