17

I'd like to merge SQLite databases, and some may be in memory. I create the in-memory databases by specifying the database path as :memory:. Following this post, using the attach feature of SQLite seems like both a simple and efficient approach. But how can I specify my in-memory database as the source to attach?

For example, I'd want to do something like:

c1 = sqlite3.connect(":memory:")
c1.execute(...create table, insert a bunch, commit...)

c2 = sqlite3.connect(":memory:")
c2.execute("""
  ATTACH ? AS ToMerge;
  BEGIN; 
    INSERT INTO Records SELECT * FROM ToMerge.Records; 
  COMMIT;
""", (c1.get_attach_id(), ))

but, of course, c1.get_attach_id() is a method I made up for demonstration purposes, since using the string :memory: would be ambiguous. How can I specify the existing c1 database?

1 Answer 1

42

A plain :memory: string connecting to an in-memory database cannot be shared or attached to from other connections.

You need to use a file: URI filename connection string with a ?cache=shared parameter to be able to share the in-memory database between connections; then you can also attach to it:

# first connection
c1 = sqlite3.connect("file::memory:?cache=shared", uri=True)

# second connection, to the *same database*
c2 = sqlite3.connect("file::memory:?cache=shared", uri=True)

# third connection, to a different database altogether
c3 = sqlite3.connect('/tmp/sqlite3.db', uri=True)
# can attach to the shared in-memory database, but only if you used
# uri=True on the original connection
c3.execute("ATTACH DATABASE 'file::memory:?cache=shared' AS inmem")

See the In-Memory Databases documentation.

Note that there can only be one such shared in-memory database; all other in-memory databases must remain private to their connection. Use databases with an actual filesystem storage if you need more complex setups; these are easy enough to clean up afterwards anyway if you create these in a tempfile.mkdtemp() temporary directory each.

Demo:

>>> import sqlite3
>>> c1 = sqlite3.connect("file::memory:?cache=shared", uri=True)
>>> c1.execute('CREATE TABLE foo (bar, baz)')
<sqlite3.Cursor object at 0x106839490>
>>> c1.execute("INSERT INTO foo VALUES ('spam', 'ham')")
<sqlite3.Cursor object at 0x106839500>
>>> c1.commit()
>>> c2 = sqlite3.connect("file::memory:?cache=shared", uri=True)
>>> list(c2.execute('SELECT * FROM foo'))
[(u'spam', u'ham')]
>>> c3 = sqlite3.connect('/tmp/sqlite3.db', uri=True)
>>> c3.execute("ATTACH DATABASE 'file::memory:?cache=shared' AS inmem")
<sqlite3.Cursor object at 0x1068395e0>
>>> list(c3.execute('SELECT * FROM inmem.foo'))
[(u'spam', u'ham')]

Support for in-memory shared-cache connections was added to SQLite version 3.7.13; for Python you can check the version of the underlying library with sqlite3.sqlite_version (string) or sqlite3.sqlite_version_info (tuple with integers):

>>> sqlite3.sqlite_version_info
(3, 8, 10, 2)
Sign up to request clarification or add additional context in comments.

1 Comment

@MartijnPieters Should the ‘In-Memory Databases’ link go to sqlite.org/inmemorydb.html instead of being a second link to url.html?

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.