Technically, this is a change in the public interface. But it seems
extremely unlikely that it would cause any issues (and any that do
appear would be trivial to fix).
fixes#516
Previously, a failed connection could be put back in a pool and when the
next query was attempted it would fail immediately trying to prepare the
query or reset the deadline. It wasn't clear if the Query or Exec call
could safely be retried since there was no way to know where it failed.
You can now call LastQuerySent and if it returns false then you're
guaranteed that the last call to Query(Ex)/Exec(Ex) didn't get far enough
to attempt to send the query. The call can be retried with a new
connection.
This is used in the stdlib to return a ErrBadConn if a network error
occurred and the statement was not attempted.
Fixes#427
CockroachDB doesn't support 80877102 and doesn't plan to, so instead allow
the user to customize the cancellation with their own function. In our
function, we call CANCEL QUERY with the query_id based on the LocalAddr().
The WaitForReady method can be used by a pool to not put a connection back
in the pool until it is finished cancelled and ready for a new query.
This addresses https://github.com/jackc/pgx/issues/321 with the
fix @jackc proposed there. Redshift users that need to connect
w/ SSL currently fork the library to delete this parameter, e.g.
8e0028d742
And, that's annoying to keep up-to-date :)
It's possible to define a type (e.g., an enum) with the same name in two
different schemas. When initializing data types after connecting, types
defined within schemas other than pg_catalog or public should be
qualified with their schema name to disambiguate them and ensure all
types with the same base name get added to the map of OID to type.
Prior to this commit, the last type scanned would "win", and all others
with the same name would be missing from the ConnInfo type maps, which
would subsequently cause any PREPARE involving columns of those missing
types to return the error "unknown oid".
This more appropriately aligns the behaviour of the library with
that advertised by the postgres documentation.
According to the table on the official documentation page
https://www.postgresql.org/docs/current/static/libpq-ssl.html,
the "require" mode should be used when:
"I want my data to be encrypted, and I accept the overhead. I trust that the network will make sure I always connect to the server I want."
This maps reasonably well to a TLS config that skips certificate verification.
Prior to this commit, execEx() would write the one round trip exec to
the connection before first calling ensureConnectionReadyForQuery, which
ultimately caused any errors to be suppressed if the exec followed a
valid query, because the receive message processing would finish
successfully as soon as it received the ReadyForQuery that actually
belonged to the preceding query. So, the exec would never actually
receive the error message that it caused, leaving it to be incorrectly
received by the first subsequent query sent.
If the Write call in sendPreparedQuery encountered a non-fatal error - which means it sent
no bytes. It still was marking the connection as not ready for query. That caused the next
call to hang.
