Review gist of CrateDB adapter

[amotl]

Hi there,

we just talked about TSPERF again, because both TSBS and RockBench will no longer be maintained going forward, so TSPERF could be considered as a good alternative.

Can I humbly ask you to evaluate and/or confirm this is the optimal / most-performant way to insert data into CrateDB?

tsperf/tsperf/adapter/cratedb.py

Lines 66 to 79 in 666ddfe

	# Create table.
	stmt = f"""CREATE TABLE {self.table_name} ("ts" TIMESTAMP WITH TIME ZONE,
	"g_ts_{self.partition}" TIMESTAMP WITH TIME ZONE GENERATED ALWAYS AS date_trunc('{self.partition}', "ts"),
	"payload" OBJECT(DYNAMIC))
	CLUSTERED INTO {self.shards} SHARDS
	PARTITIONED BY ("g_ts_{self.partition}")
	WITH (number_of_replicas = {self.replicas})"""
	logger.info(f"Preparing database with statement:\n{stmt}")
	self.cursor.execute(stmt)
	@timed_function()
	def insert_stmt(self, timestamps: list, batch: list):
	stmt = f"""INSERT INTO {self.table_name} (ts, payload) (SELECT col1, col2 FROM UNNEST(?,?))""" # noqa: S608
	self.cursor.execute(stmt, (timestamps, batch))

With kind regards,
Andreas.

/cc @WalBeh, @proddata, @hammerhead, @hlcianfagna

[hammerhead]

According to the documentation, UNNEST is still faster than multi-value statements. But it also says that the HTTP bulk endpoint is comparable to UNNEST:

The advantages are the same as using the UNNEST method:

From that perspective, I think we can use the more common self.cursor.executemany over UNNEST. Which is also what we typically use in production (for Python environments).