Postgres Connector inconsistent handling of timestamp precision

We have noticed inconsistent behaviour in how timestamp precision is preserved by Debezium during snapshotting vs. normal CDC operations on the Postgres connector.

For example on a DB field of type:

timestamp without time zone

containing value:

2018-03-28 12:03:35.406753

during a snapshot we get:

1522238615406000000 (notice loss of nanosecond precision)

However, if we update the same record we get:

1522238615406753000 (no loss of nanosecond precision)

We are not setting time.precision.mode in our config (so it should default to adaptive). We should see full nanosecond precision during both the snapshot and normal CDC.