This change is 

Heh, I hit a duplicate key violation when generating a tpcc-10k dataset: Error: creating backup for table history: pq: SST creation error at /Table/51/1/724811875183904935/0; this can happen when a primary or unique index has duplicate keys: Invalid argument: Keys must be added in order.

I see two options here: I can tweak the generation of these random rowids, or I can switch to using a UUID. The latter is what we'd recommend for production use. Let me see how bad it is.

Ok, this is ready for another look.

Review status: 0 of 4 files reviewed at latest revision, 1 unresolved discussion.

pkg/workload/tpcc/ddls.go, line 72 at r1 (raw file):

	// No PK necessary for this table.
	tpccHistorySchema = `(
		rowid    uuid PRIMARY KEY DEFAULT uuid_v4()::uuid,

@jordanlewis uuid_v4() returns BYTES. I can't recall whether it is feasible to have a specialization that also returns UUID. Do you know?

Comments from Reviewable

Review status: 0 of 4 files reviewed at latest revision, 1 unresolved discussion.

pkg/workload/tpcc/ddls.go, line 72 at r1 (raw file):

Use gen_random_uuid(), which produces a uuid type.

Comments from Reviewable

Review status: 0 of 4 files reviewed at latest revision, 1 unresolved discussion.

pkg/workload/tpcc/ddls.go, line 72 at r1 (raw file):

Done. Unfortunate that we have 2 functions that or so similar except in their return type (and names).

Comments from Reviewable

Review status: 0 of 4 files reviewed at latest revision, 1 unresolved discussion, some commit checks pending.

pkg/workload/tpcc/ddls.go, line 72 at r1 (raw file):

Yeah - gen_random_uuid is what pg calls it.

Comments from Reviewable