CREATE GRAPH feature

[adsharma]

CREATE GRAPH feature implemented in #123 melds two separate concepts:

• CREATE DATABASE concept in SQL databases such as postgres
• CREATE GRAPH concept in GQL

This combination is unusual in that most single file database implementations (SQLite, DuckDB) implement one instance = one file concept. Dealing with multiple databases is handled via ATTACH DATABASE like syntax.

The simplest implementation strategy was to have one storage manager per graph, which implicitly creates multiple files per database.

So if you create:

lbug foo.lbdb
lbug> CREATE GRAPH mygraph1;
lbug> CREATE GRAPH mygraph2;
..

you'll see:

foo.lbdb
foo.mygraph1.lbdb
foo.mygraph2.lbdb

all sharing one WAL file foo.lbdb.wal (similar to postgres)

This design needs to be validated for:

• crash recovery - at any point the WAL should be able to reconstruct the previous state
• isolation - some properties (e.g. disk size, memory limits, strict schema) enforceable on a per graph basis

Bug fix:

lbug> CREATE GRAPH mygraph1;
lbug> exit;
# restart
lbug foo.lbdb
lbug> use graph mygraph1;
Error: Binder exception: No graph named mygraph1.

[adsharma]

Some of this addressed via: #164