Thanks for the article. I thought through most of these issues prior to choosing schemas-based. I have already designed in most of the mitigations mentioned in the article. I'm also using UUIDs to identify everything, so moving to column-based is on the table if scale gets large enough.
For me there is an additional wrinkle with column-based, however. Migrations may actually be worse. We store most of the data as deltas in one main table and other tables are built from those. Migrating is literally dropping table(s), recreating, and replaying deltas. We can do that in parallel per node. But as column-based, loading the deltas in order of occurrence will be essentially random-access across nodes. Seems like that will be far slower.
Edit Nvm, reviewing my notes, I had planned to move deltas to another type of DB at that point. It wouldn't be a concern for Postgres.
For further actions, you may consider blocking this person and/or reporting abuse
We're a place where coders share, stay up-to-date and grow their careers.
Thanks for the article. I thought through most of these issues prior to choosing schemas-based. I have already designed in most of the mitigations mentioned in the article. I'm also using UUIDs to identify everything, so moving to column-based is on the table if scale gets large enough.
For me there is an additional wrinkle with column-based, however. Migrations may actually be worse. We store most of the data as deltas in one main table and other tables are built from those. Migrating is literally dropping table(s), recreating, and replaying deltas. We can do that in parallel per node. But as column-based, loading the deltas in order of occurrence will be essentially random-access across nodes. Seems like that will be far slower.
Edit Nvm, reviewing my notes, I had planned to move deltas to another type of DB at that point. It wouldn't be a concern for Postgres.