How use multiprocessing with function arcpy.da.InsertCursor

Question

I'm working with a table in the database that has more than 1 million records and a total load of this table takes about 8 hours because it has geographic data, the table is a featureclass. Basically, the solution I see is to apply threads, I already tried using concurrente.futures, but for some reason, this solution cannot open the table folder, the next solution is to use the multiprocessing library, but I am not able to understand how to add it to this project. Or how can I use the threads in this? The important thing would be to run in parallel.

def divide_data(query, batch_size):
 con = Connection.connect()
 cur = con.cursor()
 cur.execute(query)
 while True:
 batch = cur.fetchmany(batch_size)
 if not batch:
 break
 yield batch
 cur.close()
 con.close()
def insert_table(input, query, camp, batch_size=10000):
 with arcpy.da.InsertCursor(input, camp) as cursor:
 for batch in divide_data(query, batch_size):
 for r in tqdm(batch, desc='Insert lines'):
 id = r[0]
 WKB = r[1].read()
 row = [id, arcpy.FromWKB(WKB)]
 for i in range(2, len(camp)):
 row.append(r[i])
 try:
 cursor.insertRow(row)
 except Exception as e:
 print(e)
 pass
 del cursor

• Michael Stimson: ArcGIS Desktop objects are not thread safe, you will not be able declare more than one InsertCursor on a feature class/table as the first will write lock the data. The only exception is an enterprise geodatabase.
• Vince: Eight hours to load a million records? That's probably seven and three quarters hours too long (in a single thread solution). There's a lot missing here, including an indication of how long the processing takes with a pass in lieu of cursor.insertRow(row), which database is being used, whether a spatial index is present on the table, the number of vertices in each feature,... I just created a million row point table in PostgreSQL 11 via SQL in 22 seconds (including adding the spatial index), then queried it into a WKT list in 22 sec, then inserted it back into a second table in 121 sec
• Vince: The same load code with a GIST index on the geometry column increased runtime to 132 sec.