Re: PGDump question/issue - Mailing list pgsql-novice
| From | Ayden Gera |
|---|---|
| Subject | Re: PGDump question/issue |
| Date | |
| Msg-id | CANYJdWKsznOaaka97FRwHgDFJBTJ2YU9g_S-S5xC_rh+T1Vb0Q@mail.gmail.com Whole thread Raw |
| In response to | Re: PGDump question/issue (Laurenz Albe <laurenz.albe@cybertec.at>) |
| Responses |
Re: PGDump question/issue
Re: PGDump question/issue |
| List | pgsql-novice |
Hi Laurenz,
Thank you very much for your reply.
Your assumption was close but not quite correct.
Database A is SaaS provided PGDump. it comes with Drop tables commands in it. It has no row level security on it.
We want to daily populate DB A into a new DB B (supabase), but we want to add row level security to DB B. Issue is the drop table commands in the PGDump A we believe we will lose any RLS we put in place in DB B?
We could put data into a DB C (our own PG instance), then extract data only from DB C with PGDump-a, and then insert that into DB B (supabase) but we need a script to drop all data in DB B before uploading PGDump... What is command we can use either separately or part of the restore PGDump commands to first delete all data before restoring data into DB B with PGDump DB A?
Or we could put PGDump A into a separate Supabase Db B2, and then somehow stream only that data into DB B.. but unsure if we can script dropping all data automatically in DB B before DB2 streams data into it as data doesn't always have keys/unique identifiers.. so don't want to risk data double-ups.. dropping data from all table in DB B seems best.. but unclear if streaming can accommodate some automated scripting to drop DB B before it starts..?
Im open to any other ideas on how to get around the drop table that comes in PG Dump from DB A.
Many Thanks.
Ayden
On Fri, Apr 25, 2025 at 6:15 PM Laurenz Albe <laurenz.albe@cybertec.at> wrote:
On Fri, 2025-04-25 at 09:50 +1200, Ayden Gera wrote:
> Hoping someone may have a solution to this problem.
> We get a daily PGDump file (@3Gb) from our SaaS provider (for BI purpose).
> In it, it has a Drop Table IF Exists command..
> This file has no row level security etc.
> We want to use the same file to populate Supabase with and add row level
> security.. but I believe the drop table will destroy the rls each day and
> manually adding it back *unless mabe scripted) isn't an option.
>
> We have an inhouse Postgresql we can also use to potentially load and then
> do its own PGDump with data only..
>
> But the other issue we have is the source tables don't always have any
> unique keys that we can tell.. so to be safe and avoid data duplicate risk..
> we prefer to delete the entire tables data before inserting..
>
> Does anyone have any suggestions on how to best automate the daily updating
> of data into the supabase tables without losing any RLS we might configure
> on those tables?
> Or what commands should we run on our own PG to get our own data only/insert
> + commands to drop all data in all tables before running it.
>
> I was also wondering if we could send PGDump from SaaS to Supabase Db1 and
> then stream data to DB2 (Prod) but unclear if we can and/or risk data
> duplication risk if we cannot somehow delete the tables in Prod just before
> streaming..
I am not sure I understand correctly: your problem is that you want to copy
data from a database (let's call it database A) to another database B.
In the process, you want to wipe out all the data in B, but not the table
definitions, because there are different row-level security policies on the
tables in A and B.
Correct?
Then perhaps this will help:
1. export the object definitions from B:
pg_dump -F c --schema-only -f dumpb B
2. export the data from A:
pg_dump -F c --data-only -f dumpa A
3. drop database B, create it again and create all the objects:
pg_restore -d postgres --clean --create --section=pre-data dumpb
4. restore the data from A into the new database:
pg_restore -d B --section=data dumpa
5. restore the original index definitions and constraints:
pg_restore -d B --section=post-data dumpb
Yours,
Laurenz Albe
pgsql-novice by date: