Loading data into PG 10 partitions about pg_bulkload

[bwtakacy]

Nice feature :)

Let's discuss about the specification of this.
Maybe it enables a input file load into a single partition, isn't it?
Is it useful to detect the target partition automatically from the input file or control file?

[amitlan]

Handling loading into partitioned tables would be quite some work as it requires some initial setup for tuple routing. For each input tuple, an executor function called ExecFindPartition() is called to determine the target partition whose RelationData pointer is passed to heap_insert(), along with the HeapTuple corresponding to the tuple. In other words, the answer to your last question is perhaps no - routing individual tuples to target partitions needs to be handled explicitly.

On the other hand, loading into a partition works just like any regular table. If the file contains a record that doen't satisfy the partition constraint, it has to be rejected. pg_bulkload's writers need to learn how to do that. We cannot skip checking the partition constraint unlike regular user-defined CHECK constraints, which is a pg_bulkload feature. Since checking partition constraint is embedded inside ExecConstraints(), that means we cannot skip calling it, which in turn means we will end up checking the regular CHECK constraints even if don't want. Perhaps the fact that partition constraint checking is embedded inside ExecConstraints() is undesirable, so I need to talk to pgsql-hackers about that - we may need publicly accessible ExecPartitionCheck().

Anyway, it might be too early to consider partitioned tables at this point. So closing this issue.

[bwtakacy]

Hmm.

Many hardles.
But, it may be a big feature for new major version up.
I hope we will back to this again:)