Thread: Non blocking type change for primary column
Hi, I wonder if there is a feature planned (or already available in 11+) that allows one to change the column type of a table without making it block until finished. As seemingly many others, we used serial instead of bigserial in our tables and some will at a not too distant future hit their limit. I am currently evaluating our options. One, that seems possible would be 1. adding a new nullable column of type bigint. 2. copying the primary key into this field in small batches 3. Once done. rename the primary column and new column, to swap them 4. attach the sequence to the new column and deattach it from the old column 5. Make new column NON NULL (Not sure if this will block) 6. Remove primary key from old column, make new column to be the primary key (I fear this will block to create the index, but I guess not nearly as long as ALTER COLUMN type) 7. Delete old column. This could be prevented, if this would be done in a non blocking way by the database system itself, so I hope this is already implemented or at least on the roadmap. Any thoughts? Greetings, Torge.
Hi,
Le jeu. 14 janv. 2021 à 17:27, Torge Kummerow <tk@panaccess.com> a écrit :
Hi,
I wonder if there is a feature planned (or already available in 11+)
that allows one to change the column type of a table without making it
block until finished.
As seemingly many others, we used serial instead of bigserial in our
tables and some will at a not too distant future hit their limit.
I am currently evaluating our options. One, that seems possible would be
1. adding a new nullable column of type bigint.
2. copying the primary key into this field in small batches
3. Once done. rename the primary column and new column, to swap them
4. attach the sequence to the new column and deattach it from the old column
5. Make new column NON NULL (Not sure if this will block)
6. Remove primary key from old column, make new column to be the primary
key (I fear this will block to create the index, but I guess not nearly
as long as ALTER COLUMN type)
7. Delete old column.
This could be prevented, if this would be done in a non blocking way by
the database system itself, so I hope this is already implemented or at
least on the roadmap.
Any thoughts?
Nope, the table needs to be re-written to change the 4-byte column into an 8-byte column. And I don't expect to see this anytime soon.
Guillaume.
1. adding a new nullable column of type bigint.
2. copying the primary key into this field in small batches
3. Once done. rename the primary column and new column, to swap them
4. attach the sequence to the new column and deattach it from the old column
5. Make new column NON NULL (Not sure if this will block)
6. Remove primary key from old column, make new column to be the primary
key (I fear this will block to create the index, but I guess not nearly
as long as ALTER COLUMN type)
7. Delete old column.
This could be prevented, if this would be done in a non blocking way by
the database system itself, so I hope this is already implemented or at
least on the roadmap.
Any thoughts?
Hye Torge,
I was recently involved with such a change. See this thread[1] for discussion. In particular, it touches on point 5, where you ask what's blocking. Hope it helps.
[1] https://www.postgresql.org/message-id/flat/CABZeWdybDVTgCQFZf0jDtj0p4yLxNrYd8aw61iFxY-Xt%2BfqPkw%40mail.gmail.com