How to upsert pandas DataFrame to PostgreSQL table?

Question

I've scraped some data from web sources and stored it all in a pandas DataFrame. Now, in order harness the powerful db tools afforded by SQLAlchemy, I want to convert said DataFrame into a Table() object and eventually upsert all data into a PostgreSQL table. If this is practical, what is a workable method of going about accomplishing this task?

Answer 1

Update: You can save yourself some typing by using this method.

---

If you are using PostgreSQL 9.5 or later you can perform the UPSERT using a temporary table and an INSERT ... ON CONFLICT statement:

import sqlalchemy as sa

# …

with engine.begin() as conn:
    # step 0.0 - create test environment
    conn.exec_driver_sql("DROP TABLE IF EXISTS main_table")
    conn.exec_driver_sql(
        "CREATE TABLE main_table (id int primary key, txt varchar(50))"
    )
    conn.exec_driver_sql(
        "INSERT INTO main_table (id, txt) VALUES (1, 'row 1 old text')"
    )
    # step 0.1 - create DataFrame to UPSERT
    df = pd.DataFrame(
        [(2, "new row 2 text"), (1, "row 1 new text")], columns=["id", "txt"]
    )
    
    # step 1 - create temporary table and upload DataFrame
    conn.exec_driver_sql(
        "CREATE TEMPORARY TABLE temp_table AS SELECT * FROM main_table WHERE false"
    )
    df.to_sql("temp_table", conn, index=False, if_exists="append")

    # step 2 - merge temp_table into main_table
    conn.exec_driver_sql(
        """\
        INSERT INTO main_table (id, txt) 
        SELECT id, txt FROM temp_table
        ON CONFLICT (id) DO
            UPDATE SET txt = EXCLUDED.txt
        """
    )

    # step 3 - confirm results
    result = conn.exec_driver_sql("SELECT * FROM main_table ORDER BY id").all()
    print(result)  # [(1, 'row 1 new text'), (2, 'new row 2 text')]

Answer 2

I have needed this so many times, I ended up creating a gist for it.

The function is below, it will create the table if it is the first time persisting the dataframe and will update the table if it already exists:

import pandas as pd
import sqlalchemy
import uuid
import os

def upsert_df(df: pd.DataFrame, table_name: str, engine: sqlalchemy.engine.Engine):
    """Implements the equivalent of pd.DataFrame.to_sql(..., if_exists='update')
    (which does not exist). Creates or updates the db records based on the
    dataframe records.
    Conflicts to determine update are based on the dataframes index.
    This will set unique keys constraint on the table equal to the index names
    1. Create a temp table from the dataframe
    2. Insert/update from temp table into table_name
    Returns: True if successful
    """

    # If the table does not exist, we should just use to_sql to create it
    if not engine.execute(
        f"""SELECT EXISTS (
            SELECT FROM information_schema.tables 
            WHERE  table_schema = 'public'
            AND    table_name   = '{table_name}');
            """
    ).first()[0]:
        df.to_sql(table_name, engine)
        return True

    # If it already exists...
    temp_table_name = f"temp_{uuid.uuid4().hex[:6]}"
    df.to_sql(temp_table_name, engine, index=True)

    index = list(df.index.names)
    index_sql_txt = ", ".join([f'"{i}"' for i in index])
    columns = list(df.columns)
    headers = index + columns
    headers_sql_txt = ", ".join(
        [f'"{i}"' for i in headers]
    )  # index1, index2, ..., column 1, col2, ...

    # col1 = exluded.col1, col2=excluded.col2
    update_column_stmt = ", ".join([f'"{col}" = EXCLUDED."{col}"' for col in columns])

    # For the ON CONFLICT clause, postgres requires that the columns have unique constraint
    query_pk = f"""
    ALTER TABLE "{table_name}" DROP CONSTRAINT IF EXISTS unique_constraint_for_upsert;
    ALTER TABLE "{table_name}" ADD CONSTRAINT unique_constraint_for_upsert UNIQUE ({index_sql_txt});
    """
    engine.execute(query_pk)

    # Compose and execute upsert query
    query_upsert = f"""
    INSERT INTO "{table_name}" ({headers_sql_txt}) 
    SELECT {headers_sql_txt} FROM "{temp_table_name}"
    ON CONFLICT ({index_sql_txt}) DO UPDATE 
    SET {update_column_stmt};
    """
    engine.execute(query_upsert)
    engine.execute(f"DROP TABLE {temp_table_name}")

    return True

Answer 3

Here is my code for bulk insert & insert on conflict update query for postgresql from pandas dataframe:

Lets say id is unique key for both postgresql table and pandas df and you want to insert and update based on this id.

import pandas as pd
from sqlalchemy import create_engine, text

engine = create_engine(postgresql://username:pass@host:port/dbname)
query = text(f""" 
                INSERT INTO schema.table(name, title, id)
                VALUES {','.join([str(i) for i in list(df.to_records(index=False))])}
                ON CONFLICT (id)
                DO  UPDATE SET name= excluded.name,
                               title= excluded.title
         """)
engine.execute(query)

Make sure that your df columns must be same order with your table.

EDIT 1:

Thanks to Gord Thompson's comment, I realized that this query won't work if there is single quote in columns. Therefore here is a fix if there is single quote in columns:

import pandas as pd
from sqlalchemy import create_engine, text

df.name = df.name.str.replace("'", "''")
df.title = df.title.str.replace("'", "''")
engine = create_engine(postgresql://username:pass@host:port/dbname)
query = text(""" 
            INSERT INTO author(name, title, id)
            VALUES %s
            ON CONFLICT (id)
            DO  UPDATE SET name= excluded.name,
                           title= excluded.title
     """ % ','.join([str(i) for i in list(df.to_records(index=False))]).replace('"', "'"))
engine.execute(query)

Answer 4

Consider this function if your DataFrame and SQL Table contain the same column names and types already. Advantages:

• Good if you have a long dataframe to insert. (Batching)
• Avoid writing long sql statement in your code.
• Fast

.

from sqlalchemy import Table
from sqlalchemy.engine.base import Engine as sql_engine
from sqlalchemy.dialects.postgresql import insert
from sqlalchemy.ext.automap import automap_base
import pandas as pd


def upsert_database(list_input: pd.DataFrame, engine: sql_engine, table: str, schema: str) -> None:
    if len(list_input) == 0:
        return None
    flattened_input = list_input.to_dict('records')
    with engine.connect() as conn:
        base = automap_base()
        base.prepare(engine, reflect=True, schema=schema)
        target_table = Table(table, base.metadata,
                             autoload=True, autoload_with=engine, schema=schema)
        chunks = [flattened_input[i:i + 1000] for i in range(0, len(flattened_input), 1000)]
        for chunk in chunks:
            stmt = insert(target_table).values(chunk)
            update_dict = {c.name: c for c in stmt.excluded if not c.primary_key}
            conn.execute(stmt.on_conflict_do_update(
                constraint=f'{table}_pkey',
                set_=update_dict)
            )

Answer 5

If you already have a pandas dataframe you could use df.to_sql to push the data directly through SQLAlchemy

from sqlalchemy import create_engine
#create a connection from Postgre URI
cnxn = create_engine("postgresql+psycopg2://username:password@host:port/database")
#write dataframe to database
df.to_sql("my_table", con=cnxn, schema="myschema")