You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Stepping through the postgres source AFAICT the binary protocol expects that TIMESTAMP WITH TIME ZONE records are sent as int64 values, but that int64 value is supposed to represent a local not and not UTC (as it stored natively in arrow):
I didn't see anything in libpq-fe.h to help with this, but there may be something in another header. Alternately we could try setting the connection timezone to UTC as documented at the end of this section; not sure if there are other reasons for the client to be aware of local time when transmitting data like this:
closes#867
The lifecycle of setting the session time zone in this isn't great. I
think should be handled during connect somehow? But not sure if postgres
even supports that reading things like:
https://stackoverflow.com/a/11779621/621736
I think there is also something awry with the release callback for
timezone schemas; needs further investigation
---------
Co-authored-by: David Li <[email protected]>
Stepping through the postgres source AFAICT the binary protocol expects that
TIMESTAMP WITH TIME ZONE
records are sent as int64 values, but that int64 value is supposed to represent a local not and not UTC (as it stored natively in arrow):https://github.com/postgres/postgres/blob/e251e780bfe62168d0edfec459ebdc66b991753f/src/backend/utils/adt/timestamp.c#L1885
Guessing we need to build out some way to localize those values to the server time before passing via the protocol.
The text was updated successfully, but these errors were encountered: