Re: managing primary key conflicts while restoring data to table with existing data - Mailing list pgsql-general

On 26/09/19 12:03 AM, Adrian Klaver wrote:

On 9/25/19 8:04 AM, Rob Sargent wrote:

On Sep 25, 2019, at 8:24 AM, Krishnakant Mane <kkmane@riseup.net <mailto:kkmane@riseup.net>> wrote:

On 25/09/19 7:50 PM, Adrian Klaver wrote:

On 9/25/19 12:15 AM, Krishnakant Mane wrote:

Hello all,

I have been using postgresql for an enterprise quality account's automation and inventory management software called GNUKhata <https://gnukhata.in>

Our team is planning to add backup and restore function in the software.

But we don't want to dump the entire database and then restore the same.

What we are trying to do is to copy data specific to an organization.

The challenge here is that I might copy all data (account heads, bills, vouchers etc ) for one organization from an instance on one machine.

I take the archive in what ever format to another machine and now attempt to restore.

The risk here is for example if the primary key value for orgcode in the organization table is 5, it might conflict with the data where I am attempting it to be restored.

Same holds true for bills, invoices etc.

A certain account head with accountcode 1 might be already present on the second machine.

I am not expecting the users to empty all data from the destination machine before restoring a backup.

The reason is that an auditor may have many client's data and one can't predict what primary key values are going to come from a backup.

Basically I can even say this is a copy paste instead of a pure backup and restore.

Can any one suggest how to handle such conflicts?

Hard to say. If the data is held in common tables(bills, vouchers, etc)then the only thing I see happening is changing the PK values to an unused value. That could turn into a nightmare though. Not only that you lose the connection to the original data source. If the data can be broken out into separate tables then I could see placing them in their own schema.

-- 
Regards,
Krishnakant Mane,
Project Founder and Leader,
GNUKhata <https://gnukhata.in/>
//(Opensource Accounting, Billing and Inventory Management Software)//

Hi Adrian,

Even I am thinnking to do some kind of upsert with this situation.

So to be clear the tables you are working can have records from multiple organizations in a single table?

And I would have to set the pkey to an unassigned value when there is conflict.

I am seeing nextval() in your future:)

I may also choose to revamp the serial by timestamps but don't know if the target customers would like it.

I would avoid that. In my opinion timestamps are to too volatile to serve as a PK. If you are going to change I would go with the previous suggestion of UUID:
https://www.postgresql.org/docs/11/datatype-uuid.html

Not sure your customers would like that either.

Hi Adrian,

I think I would make them like the uuid idea.

So now what I am thinking is to first revamp the database by first removing all the primary key constraints and then deleting all the values.

Then loop through the existing data and get uuid in that colum for every row.

I might also require to update all the references to this value as foreign key in related tables.

But I guess some kind of on update cascade might do well.

I know this would slow down the system, but given that this will be a one time process for an individual user (that too if he has existing data ), I would take that trade-off.

What do you say?

-- 
Regards,
Krishnakant Mane,
Project Founder and Leader,
GNUKhata <https://gnukhata.in/>
//(Opensource Accounting, Billing and Inventory Management Software)//

It would likely be easier to rethink your backup and restore plan. Putting each restore into its own space would be one tack.

--
Regards,
Krishnakant Mane,
Project Founder and Leader,
GNUKhata
(Opensource Accounting, Billing and Inventory Management Software)

You might think about adding the new UUID column and use the existing primary key to inform the updates in dependent tables. Then remove the old PK  column and constraint followed by promoting the UUID to primary key. This could be safely scripted and applied to all instances of your data.