💄 Surface better error when a publication is in an invalid state #1844
Conversation
dosubot
bot
added
the
size:M
acco
force-pushed
the
acco/06-18-_surface_better_error_when_a_publication_is_in_an_invalid_state
branch
from
addb683 to
57f6055
Compare
dosubot
bot
added
documentation
acco
force-pushed
the
acco/06-18-_surface_better_error_when_a_publication_is_in_an_invalid_state
branch
from
57f6055 to
8d5f6c0
Compare
dosubot
bot
added
size:L
|
|
||
| This is a [known Postgres issue](https://www.postgresql.org/message-id/18683-a98f79c0673be358%40postgresql.org). | ||
|
|
||
| **Fix**: drop and re-create the replication slot *after* the publication exists: |
There was a problem hiding this comment.
If you want to prevent missing any changes, you should create a new slot, move Sequin to the new slot, then drop the old slot.
There was a problem hiding this comment.
@RTLS The old slot will contain changes that the new slot will not when you create it :( so data loss feels certain. It's a big bummer, as dropping a publication feels non-deleterious, but it can be problematic.
Interestingly, by happenstance, it might work as when we connect to the new slot we might send back an old LSN first (cached) and possibly traverse backward - but this would be basically unsupported behavior. We might consider ways to make this operation "first class"
| def to_existing_atom_safe(string) do | ||
| String.to_existing_atom(string) | ||
| rescue | ||
| _ -> string | ||
| end |
There was a problem hiding this comment.
should this return a status tuple? I feel like this will lead to raises later due to an expected string
There was a problem hiding this comment.
Good point, will change!
No description provided.