peewee.OperationalError: too many SQL variables on upsert of only 150 rows * 8 columns

Question

With the below example, on my machine, setting range(150) leads to the error, while range(100) does not:

from peewee import *

database = SqliteDatabase(None)

class Base(Model):
    class Meta:
        database = database


colnames = ["A", "B", "C", "D", "E", "F", "G", "H"]
cols = {x: TextField() for x in colnames}

table = type('mytable', (Base,), cols)
database.init('test.db')
database.create_tables([table])

data = []
for x in range(150):
    data.append({x: 1 for x in colnames})


with database.atomic() as txn:
    table.insert_many(data).upsert().execute()

Leads to:

Traceback (most recent call last):
  File "<stdin>", line 2, in <module>
  File "/cluster/home/ifiddes/python2.7/lib/python2.7/site-packages/peewee.py", line 3213, in execute
    cursor = self._execute()
  File "/cluster/home/ifiddes/python2.7/lib/python2.7/site-packages/peewee.py", line 2628, in _execute
    return self.database.execute_sql(sql, params, self.require_commit)
  File "/cluster/home/ifiddes/python2.7/lib/python2.7/site-packages/peewee.py", line 3461, in execute_sql
    self.commit()
  File "/cluster/home/ifiddes/python2.7/lib/python2.7/site-packages/peewee.py", line 3285, in __exit__
    reraise(new_type, new_type(*exc_args), traceback)
  File "/cluster/home/ifiddes/python2.7/lib/python2.7/site-packages/peewee.py", line 3454, in execute_sql
    cursor.execute(sql, params or ())
peewee.OperationalError: too many SQL variables

This seems very low to me. I am trying to use peewee to replace existing pandas based SQL construction, because pandas lacks support for a primary key. Only being able to insert ~100 records per loop is very low, and fragile if the number of columns goes up some day.

How can I make this work better? Is it possible?

Answer 1

After some investigation, the problem appears to be related with the maximum number of parameters that a sql query can have: SQLITE_MAX_VARIABLE_NUMBER.

To be able to do big bulk inserts I first estimate SQLITE_MAX_VARIABLE_NUMBER and then use it to create chunks in the list of dictionaries I want to insert.

To estimate the value I use this function inspired by this answer:

def max_sql_variables():
    """Get the maximum number of arguments allowed in a query by the current
    sqlite3 implementation. Based on `this question
    `_

    Returns
    -------
    int
        inferred SQLITE_MAX_VARIABLE_NUMBER
    """
    import sqlite3
    db = sqlite3.connect(':memory:')
    cur = db.cursor()
    cur.execute('CREATE TABLE t (test)')
    low, high = 0, 100000
    while (high - 1) > low: 
        guess = (high + low) // 2
        query = 'INSERT INTO t VALUES ' + ','.join(['(?)' for _ in
                                                    range(guess)])
        args = [str(i) for i in range(guess)]
        try:
            cur.execute(query, args)
        except sqlite3.OperationalError as e:
            if "too many SQL variables" in str(e):
                high = guess
            else:
                raise
        else:
            low = guess
    cur.close()
    db.close()
    return low

SQLITE_MAX_VARIABLE_NUMBER = max_sql_variables()

Then I use the above variable to slice the data

with database.atomic() as txn:
    size = (SQLITE_MAX_VARIABLE_NUMBER // len(data[0])) -1
    # remove one to avoid issue if peewee adds some variable
    for i in range(0, len(data), size):
        table.insert_many(data[i:i+size]).upsert().execute()

---

An update about execution speed of max_sql_variables.

On a 3 years old Intel machine with 4 cores and 4 Gb of RAM, running OpenSUSE tumbleweed, with SQLITE_MAX_VARIABLE_NUMBER set to 999, the function runs in less that 100ms. If I set high = 1000000, the execution time becomes of the order of 300ms.

On a younger Intel machine with 8 cores and 8Gb of RAM, running Kubuntu, with SQLITE_MAX_VARIABLE_NUMBER set to 250000, the function runs in about 2.6 seconds and returns 99999. If I set high = 1000000, the execution time becomes of the order of 4.5 seconds.