Most performant method to update many records?

[StayFoolisj]

I use Postgres 12 and a few times a day I need to update 30-80k rows and ideally in less than a minute.

Using SQLAlchemy / psycopg2 bulk_update_mappings is not really performant, its taking +10minutes even with chunking the data and using multithreading. So I started looking for options, which led me to find asyncpg.

Account model:

class Account(Base):
    __tablename__ = 'accounts'

    address                  = Column(String, primary_key=True)
    protocols_used           = Column(ARRAY(String))

A snippet of what I'm doing now (minus chunking & multithreading for readability). My desired outcome is a significant speed improvement on this. 1k row updates per 1s would be ideal. I'm not sure what is realistic.

users_to_update_data = [acc for acc in all_accounts_data if acc['address'] in users_to_update] 
updates = []

for account_data in users_to_update_data:
    updates.append(
        {
            "address": account_data['address'],
          "protocols_used": account_data['protocols_used'] + [protocol]
      }
  )

session.bulk_update_mappings(Account, entries_to_update)

From what I understand, this could be a job for executemany() ? I can prepare a list of either dicts or tuples (I believe tuples are more efficiently iterated due to their immutability?). I failed to find any code examples in the docs or elsewhere. Would be very grateful if someone could point me in the right direction with an example or suggestions. Thanks!

[elprans]

Indeed, use executemany:

updates = [(account_id, new_address, additional_protocol) from <data_source>]

await connection.executemany(
    """
    UPDATE accounts
    SET 
      address = $2,
      protocols_used = array_append(protocols_used, $3)
    WHERE
      id = $1
    """,
    updates,
)

[StayFoolisj]

Thank you so much!

I refactored my script to using your suggestion + using sets instead of lists and total duration for my script now takes 21s instead of 20minutes! That involves fetching 500k records and updating 50k of them.

[elprans]

Did you use asyncpg to fetch records as well?

[StayFoolisj]

Yes but its a bit slower than the SQLAlchemy equivalent. Maybe because of the connection management?

async def get_all_accounts_asyncpg():
    """
    Returns all accounts in the database
    """
    connection = await asyncpg.connect(**db_params)
    query = await connection.fetch('SELECT * FROM accounts')
    await connection.close()
    return [dict(r) for r in query]

def get_all_accounts_sqlalchemy():
    """
    Returns all accounts in the database
    """
    query = session.query(Account).all()
    return [r.address for r in query]

[elprans]

You don't need to cast asyncpg.Record to dict, it's already dict-like. Remove the cast and see what happens.

[bobir01]

You don't need to cast asyncpg.Record to dict, it's already dict-like. Remove the cast and see what happens.

Good part of the day
I would like to construct data structure similar to asyncpg.Record for my project, is it possible with only pure paython ?
thanks