What is the proper formatting for a jsonb array in Python for Postgres?

问题

I have a schema that looks like

Column                  |            Type             |
-------------------------------------------------------
message_id              | integer                     | 
 user_id                | integer                     |
 body                   | text                        |
 created_at             | timestamp without time zone |
 source                 | jsonb                       |
 symbols                | jsonb[]                     |

I am trying to use psycopg2 to insert data via psycopg2.Cursor.copy_from() but I am getting numerous issues trying to figure out how a jsonb[] object should be formatted. When I do a straight list of JSON objects, I get an error that looks like

psycopg2.errors.InvalidTextRepresentation: malformed array literal: "[{'id': 13016, 'symbol':
.... 
DETAIL:  "[" must introduce explicitly-specified array dimensions.

I've tried numerous different escapes on the double quotes and curly braces. If I do a json.dumps() on my data, I get the below error.

psycopg2.errors.InvalidTextRepresentation: invalid input syntax for type json
DETAIL:  Token "'" is invalid.

This error is received from this code snippet

messageData = []
symbols = messageObject["symbols"]
newSymbols = []
for symbol in symbols:
    toAppend = symbol
    toAppend = refineJSON(json.dumps(symbol))
    toAppend = re.sub("{", "\{", toAppend)
    toAppend = re.sub("}", "\}", toAppend)
    toAppend = re.sub('"', '\\"', toAppend)
    newSymbols.append(toAppend)
messageData.append(set(newSymbols))

I'm also open to defining the column as a different type (e.g., text) and then attempting a conversion but I haven't been able to do that either.

messageData is the input to a helper function that calls psycopg2.Cursor.copy_from()

def copy_string_iterator_messages(connection, messages, size: int = 8192) -> None:
    with connection.cursor() as cursor:
        messages_string_iterator = StringIteratorIO((
            '|'.join(map(clean_csv_value, (messageData[0], messageData[1], messageData[2], messageData[3], messageData[4], messageData[5], messageData[6], messageData[7], messageData[8], messageData[9], messageData[10], 
                messageData[11],
            ))) + '\n'
            for messageData in messages
        ))
        # pp.pprint(messages_string_iterator.read())
        cursor.copy_from(messages_string_iterator, 'test', sep='|', size=size)
        connection.commit()

EDIT: Based on the input from Mike, I updated the code to use execute_batch() where messages is a list containing messageData for each message.

def insert_execute_batch_iterator_messages(connection, messages, page_size: int = 1000) -> None:
    with connection.cursor() as cursor:
        iter_messages = ({**message, } for message in messages)

        print("inside")

        psycopg2.extras.execute_batch(cursor, """
            INSERT INTO test VALUES(
                %(message_id)s,
                %(user_id)s,
                %(body)s,
                %(created_at)s,
                %(source)s::jsonb,
                %(symbols)s::jsonb[]
            );
        """, iter_messages, page_size=page_size)
        connection.commit()

回答1:

Your question made me curious. This below works for me. I have doubts whether the escaping going to/from CSV can be resolved.

My table:

=# \d jbarray
                             Table "public.jbarray"
 Column  |  Type   | Collation | Nullable |               Default
---------+---------+-----------+----------+-------------------------------------
 id      | integer |           | not null | nextval('jbarray_id_seq'::regclass)
 symbols | jsonb[] |           |          |
Indexes:
    "jbarray_pkey" PRIMARY KEY, btree (id)

Completely self-contained Python code:

mport json
import psycopg2

con = psycopg2.connect('dbname=<my database>')

some_objects = [{'id': x, 'array': [x, x+1, x+2, {'inside': x+3}]} for x in range(5)]

insert_array = [json.dumps(x) for x in some_objects]
print(insert_array)

c = con.cursor()

c.execute("insert into jbarray (symbols) values (%s::jsonb[])", (insert_array,))

con.commit()

Result:

=# select * from jbarray;
-[ RECORD 1 ]-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
id      | 1
symbols | {"{\"id\": 0, \"array\": [0, 1, 2, {\"inside\": 3}]}","{\"id\": 1, \"array\": [1, 2, 3, {\"inside\": 4}]}","{\"id\": 2, \"array\": [2, 3, 4, {\"inside\": 5}]}","{\"id\": 3, \"array\": [3, 4, 5, {\"inside\": 6}]}","{\"id\": 4, \"array\": [4, 5, 6, {\"inside\": 7}]}"}

=# select id, unnest(symbols) from jbarray;
-[ RECORD 1 ]----------------------------------------
id     | 1
unnest | {"id": 0, "array": [0, 1, 2, {"inside": 3}]}
-[ RECORD 2 ]----------------------------------------
id     | 1
unnest | {"id": 1, "array": [1, 2, 3, {"inside": 4}]}
-[ RECORD 3 ]----------------------------------------
id     | 1
unnest | {"id": 2, "array": [2, 3, 4, {"inside": 5}]}
-[ RECORD 4 ]----------------------------------------
id     | 1
unnest | {"id": 3, "array": [3, 4, 5, {"inside": 6}]}
-[ RECORD 5 ]----------------------------------------
id     | 1
unnest | {"id": 4, "array": [4, 5, 6, {"inside": 7}]}

If the insert performance is too slow for you, then you can use a prepared statement with execute_batch() as documented here. I have used that combination, and it was very fast.

来源：https://stackoverflow.com/questions/62685857/what-is-the-proper-formatting-for-a-jsonb-array-in-python-for-postgres

标签

python

arrays

postgresql

psycopg2

jsonb