Bump postgresql driver to 42.3.3

[hashhar]

Description

It addresses CVE-2022-21724.

It also includes some fixes around calendars being used in thread-safe
way, result of PgObject#isNull, possible connection leak and loss of
microseconds when reading TIME(6) via getTimestamp.

Documentation

(x) No documentation is needed.

Release notes

(x) Release notes entries required with the following suggested text:

# PostgreSQL
* Fix possible connection leak when connecting to PostgreSQL failed. ({issue}`11449`)

[findepi]

It also includes some fixes around calendars being used in thread-safe
way

Does this affect the connector?

possible connection leak

Does this affect the connector?

and loss of
microseconds when reading TIME(6) via getTimestamp.

it doesn't affect the connector, right?

[hashhar]

The connection leak does affect us, I was able to reproduce it by setting low per-user connection limit in PostgreSQL. I saw open sockets even if connection attempt failed. The connection does get cleaned up eventually but under load it seems very much possible to have idle sockets.

The calendar thread-safety issue is possible to hit by sharing connections across threads - I couldn't find existing code that does this though.

The loss of microseconds doesn't affect us since we don't use getTimestamp - sorry for keeping it in the commit message.

(All of this context could've gone to the commit message 😞 )

[findepi]

The calendar thread-safety issue is possible to hit by sharing connections across threads - I couldn't find existing code that does this though.

the connector doesn't do that (and shouldn't)

The connection leak does affect us, I was able to reproduce it by setting low per-user connection limit in PostgreSQL.

Let's have this in RNs.