Benchmarking MySQL INSERTs into InnoDB and MyISAM

bench_mysql_innodb_myisam.py

import csv
import json
import os
import pymysql
import statistics
import time


def prepare_values(row):
    new_row = []
    for cell in row:
        if isinstance(cell, list):
            new_row.append(json.dumps(cell))
        else:
            new_row.append(cell)
    return new_row


def batch_insert(tbl_name, rows, cursor, is_full):
    query = f"INSERT INTO `{tbl_name}` (user_id,user_email,created_at,tags,shared_with,is_completed,is_overdue,description,title,user_name) VALUES "
    values_str_list = []

    for row in rows:
        row = prepare_values(row)
        value_str = "(" + ",".join(["%s" for _ in row]) + ")"
        values_str_list.append(cursor.mogrify(value_str, row))

    query += ",".join(values_str_list) + ";"
    start_time = time.time()
    cursor.execute(query)
    end_time = time.time()
    if is_full:
        timings[tbl_name]["full"].append(end_time - start_time)
    else:
        timings[tbl_name]["empty"].append(end_time - start_time)


def process_csv_file(
    cursor, csv_file_path, tbl_name, chunk_size, num_rows, is_full: bool
):
    values = []

    with open(csv_file_path, "r") as csvfile:
        csvreader = csv.reader(csvfile, quotechar="'")

        for i, row in enumerate(csvreader, 1):
            values.append(row)

            if i % chunk_size == 0:
                print(
                    f"inserting batch {i // chunk_size} / {num_rows // chunk_size} into {tbl_name} ({'full' if is_full else 'empty'})"
                )
                batch_insert(tbl_name, values, cursor, is_full)
                values = []

        if values:
            batch_insert(tbl_name, values, cursor, is_full)


db_params = {
    "host": "YOUR_HOST",
    "password": "YOUR_PASS,
    "user": "YOUR_USER",
    "db": "YOUR_DB",
    "autocommit": True,
    "ssl": {"fake_flag_to_enable_tls": True},
}

csv_file_path1 = "/mnt/ramdisk/xaa.csv"
csv_file_path2 = "/mnt/ramdisk/xab.csv"
csv_file_size = os.path.getsize(csv_file_path1)
table_name1 = "test_innodb"
table_name2 = "test_myisam"
num_rows = 12_500_000
chunk_size = 10_000
timings = {
    table_name1: {"empty": [], "full": []},
    table_name2: {"empty": [], "full": []},
}

connection = pymysql.connect(**db_params)
cursor = connection.cursor()

for i, tbl in enumerate([table_name1, table_name2]):
    for j, f in enumerate([csv_file_path1, csv_file_path2]):
        print(f"\nloading file {j} / 2 into {tbl} in batches of {chunk_size} rows\n")
        process_csv_file(cursor, f, tbl, chunk_size, num_rows, bool(j))

cursor.close()
connection.close()

for table, table_timings in timings.items():
    for category, category_timings in table_timings.items():
        min_time = min(category_timings)
        max_time = max(category_timings)
        avg_time = statistics.mean(category_timings)
        stdev_time = statistics.stdev(category_timings)
        total_time = sum(category_timings)

        print(f"\ntimings for {table} ({category}):")
        print(f"\tMin: {min_time:.3f} seconds")
        print(f"\tMax: {max_time:.3f} seconds")
        print(f"\tAvg: {avg_time:.3f} seconds")
        print(f"\tStd Dev: {stdev_time:.3f} seconds")
        print(
            f"\tLoaded {num_rows} rows in {total_time:.2f} seconds at {((csv_file_size / total_time) / 2**20):.2f} MiB/s"
        )

timings for test_innodb (empty):
        Min: 0.437 seconds
        Max: 1.631 seconds
        Avg: 1.034 seconds
        Std Dev: 0.180 seconds
        Loaded 12500000 rows in 1292.94 seconds at 2.03 MiB/s

timings for test_innodb (full):
        Min: 0.458 seconds
        Max: 1.688 seconds
        Avg: 1.069 seconds
        Std Dev: 0.169 seconds
        Loaded 12500000 rows in 1336.56 seconds at 1.96 MiB/s

timings for test_myisam (empty):
        Min: 0.486 seconds
        Max: 2.424 seconds
        Avg: 1.017 seconds
        Std Dev: 0.230 seconds
        Loaded 12500000 rows in 1271.32 seconds at 2.06 MiB/s

timings for test_myisam (full):
        Min: 0.537 seconds
        Max: 2.427 seconds
        Avg: 1.033 seconds
        Std Dev: 0.219 seconds
        Loaded 12500000 rows in 1291.43 seconds at 2.03 MiB/s

Data sample that was used in this and the bulk load. The PK (the only key) for the tables is a CHAR(32), and is a UUIDv7 as shown here. While not as fast as an AUTO_INCREMENT INT, it is K-sortable so it won't cause nearly the page split issue as a UUIDv4 would.

'user_id','user_email','created_at','tags','shared_with','is_completed','is_overdue','description','title','user_name'
'018aae66103a7e7aa19e35f1b4330c22','killie.kironde@moonstone.com','2014-12-26 12:30:05','[]','[]','1','0','refute skittle unshipped thinly stark icing bribe subsystem ecard prudishly','polar','Killie.Kironde'
'018aae66103b752a81967307d5898b67','augusto.schluter@critter.com','2023-03-20 16:38:05','[]','[]','1','0','video badass surprise gander bloating jalapeno overpower conjure underpay oboe','starboard','Augusto.Schluter'
'018aae66103c7f4399a5aedd96521b89','orton.wong@widen.com','2017-08-07 21:17:02','[]','[]','0','0','basin proximity bonfire fantasy camcorder broadness waged willed strike moaner','chevy','Orton.Wong'
'018aae66103d7db488774840e2c2387d','bucky.shaffert@roulette.com','2014-06-26 22:37:13','[]','[]','1','0','shock happier tiara galleria sharply expire coeditor baritone zen dangling','cardinal','Bucky.Shaffert'
'018aae66103e7aa1a5a0b5717a876eb5','vasily.colbye@stifling.com','1997-07-23 11:14:30','[]','[]','1','0','tinderbox darkening dangle expand rash marine caliber hardness foyer wound','resort','Vasily.Colbye'
'018aae66103f7df78a68aaaaa6668501','maddy.childs@conducive.com','2016-10-13 00:52:01','[]','[]','1','0','circus saffron draw supreme audience stipulate acrobat context womankind impure','armless','Maddy.Childs'
'018aae6610407e4dbdba2d01093860b4','lucine.kunz@ambiguity.com','2001-07-03 16:29:45','[]','[]','1','0','omen skillful purposely drove pond uptown clumsily hardly womanlike obsession','backspin','Lucine.Kunz'
'018aae6610417848b92c5b8d19efe100','roderick.linson@stress.com','2000-04-14 23:49:16','[]','[]','1','0','passing puzzling until condiment getting spud coconut quiver pronounce casually','flick','Roderick.Linson'
'018aae6610427200b0e03f984f352ed4','jorrie.atkinson@pronounce.com','2010-09-25 08:27:01','[]','[]','1','0','smartness banknote thursday kindly pesticide selection dude snub shrink drone','geek','Jorrie.Atkinson'
'018aae66104372ad91c0fbcd9aae46b9','artur.sikata@frying.com','2014-12-12 16:44:49','[]','["018aae050842771f9eaa5b9ab581cfe1","018aae4c6c967046a0548438d1b7b9dc","018aadaf5172779f84b5e508557dbd9b"]','0','0','approval squeamish cornstalk spectator perfectly residue crushed resubmit swipe student','capillary','Artur.Sikata'