This repo contains my implementation for CS186 RookieDB projects. There is a gitbook for CS186 projects, but it may be updated each semester, so I cloned the 2021 spring version. You can find the projects handout that I used here.
The master branch contains a bare-bones database implementation, which supports executing simple transactions in series. This is the skeleton code you will use throughout the projects. In the assignments of this class, you will be adding support for B+ tree indices, efficient join algorithms, query optimization, multigranularity locking to support concurrent execution of transactions, and database recovery.
My implementation for each project is in the corresponding branch as follows (you can use git branch -a to see them), you can git checkout branch_name to see my implementation for each project.
| Assignment | Branch name |
|---|---|
| Skeleton code | Master |
| Project 2: B+ Trees | b+-tree |
| Project 3: Joins and Query Optimization | join_query_opt |
| Project 4: Concurrency | concurrency |
| Project 5: Recovery | recovery |
To start your Rookiedb projects journey, first clone the skeleton code (i.e. master branch) , then follow the instruction below to set up your local development environment. When you start one specific project, you can create a new branch from master then implement it to keep your code tree clean.
You are free to use any text editor or IDE to complete the assignments, but we will build and test your code in a docker container with Maven.
We recommend setting up a local development environment by installing Java 8 locally (the version our Docker container runs) and using an IDE such as IntelliJ.
If you have another version of Java installed, it's probably fine to use it, as long as you do not use any features not in Java 8. You should run tests somewhat frequently inside the container to make sure that your code works with our setup.
To import the project into IntelliJ, make sure that you import as a Maven project (select the pom.xml file when importing). Make sure that you can compile your code and run tests (it's ok if there are a lot of failed tests - you haven't begun implementing anything yet!). You should also make sure that you can run the debugger and step through code.
If you are using IntelliJ, and wish to run the tests for a given assignment follow the instructions in the following document:
As you will be working with this codebase for the rest of the semester, it is a good idea to get familiar with it. The code is located in the src/main/java/edu/berkeley/cs186/database directory, while the tests are located in the src/test/java/edu/berkeley/cs186/database directory. The following is a brief overview of each of the major sections of the codebase.
The cli directory contains all the logic for the database's command line interface. Running the main method of CommandLineInterface.java will create an instance of the database and create a simple text interface that you can send and review the results of queries in. The inner workings of this section are beyond the scope of the class (although you're free to look around), you'll just need to know how to run the Command Line Interface.
The subdirectory cli/parser contains a lot of scary looking code! Don't be intimidated, this is all generated automatically from the file RookieParser.jjt in the root directory of the repo. The code here handles the logic to convert from user inputted queries (strings) into a tree of nodes representing the query (parse tree).
The subdirectory cli/visitor contains classes that help traverse the trees created from the parser and create objects that the database can work with directly.
The common directory contains bits of useful code and general interfaces that
are not limited to any one part of the codebase.
The concurrency directory contains a skeleton for adding multigranularity
locking to the database. You will be implementing this in Project 4.
Our database has, like most DBMS's, a type system distinct from that of the programming language used to implement the DBMS. (Our DBMS doesn't quite provide SQL types either, but it's modeled on a simplified version of SQL types).
The databox directory contains classes which represents values stored in
a database, as well as their types. The various DataBox classes represent
values of certain types, whereas the Type class represents types used in the
database.
An example:
DataBox x = new IntDataBox(42); // The integer value '42'.
Type t = Type.intType(); // The type 'int'.
Type xsType = x.type(); // Get x's type, which is Type.intType().
int y = x.getInt(); // Get x's value: 42.
String s = x.getString(); // An exception is thrown, since x is not a string.The index directory contains a skeleton for implementing B+ tree indices. You
will be implementing this in Project 2.
The memory directory contains classes for managing the loading of data
into and out of memory (in other words, buffer management).
The BufferFrame class represents a single buffer frame (page in the buffer
pool) and supports pinning/unpinning and reading/writing to the buffer frame.
All reads and writes require the frame be pinned (which is often done via the
requireValidFrame method, which reloads data from disk if necessary, and then
returns a pinned frame for the page).
The BufferManager interface is the public interface for the buffer manager of
our DBMS.
The BufferManagerImpl class implements a buffer manager using
a write-back buffer cache with configurable eviction policy. It is responsible
for fetching pages (via the disk space manager) into buffer frames, and returns
Page objects to allow for manipulation of data in memory.
The Page class represents a single page. When data in the page is accessed or
modified, it delegates reads/writes to the underlying buffer frame containing
the page.
The EvictionPolicy interface defines a few methods that determine how the
buffer manager evicts pages from memory when necessary. Implementations of these
include the LRUEvictionPolicy (for LRU) and ClockEvictionPolicy (for clock).
The io directory contains classes for managing data on-disk (in other words,
disk space management).
The DiskSpaceManager interface is the public interface for the disk space
manager of our DBMS.
The DiskSpaceMangerImpl class is the implementation of the disk space
manager, which maps groups of pages (partitions) to OS-level files, assigns
each page a virtual page number, and loads/writes these pages from/to disk.
The query directory contains classes for managing and manipulating queries.
The various operator classes are query operators (pieces of a query), some of which you will be implementing in Project 3.
The QueryPlan class represents a plan for executing a query (which we will be
covering in more detail later in the semester). It currently executes the query
as given (runs things in logical order, and performs joins in the order given),
but you will be implementing
a query optimizer in Project 3 to run the query in a more efficient manner.
The recovery directory contains a skeleton for implementing database recovery
a la ARIES. You will be implementing this in Project 5.
The table directory contains classes representing entire tables and records.
The Table class is, as the name suggests, a table in our database. See the
comments at the top of this class for information on how table data is layed out
on pages.
The Schema class represents the schema of a table (a list of column names
and their types).
The Record class represents a record of a table (a single row). Records are
made up of multiple DataBoxes (one for each column of the table it belongs to).
The RecordId class identifies a single record in a table.
The PageDirectory class is an implementation of a heap file that uses a page directory.
The table/stats directory contains classes for keeping track of statistics of
a table. These are used to compare the costs of different query plans, when you
implement query optimization in Project 4.
The Transaction interface is the public interface of a transaction - it
contains methods that users of the database use to query and manipulate data.
This interface is partially implemented by the AbstractTransaction abstract
class, and fully implemented in the Database.Transaction inner class.
The TransactionContext interface is the internal interface of a transaction -
it contains methods tied to the current transaction that internal methods
(such as a table record fetch) may utilize.
The current running transaction's transaction context is set at the beginning
of a Database.Transaction call (and available through the static
getCurrentTransaction method) and unset at the end of the call.
This interface is partially implemented by the AbstractTransactionContext abstract
class, and fully implemented in the Database.TransactionContext inner class.
The Database class represents the entire database. It is the public interface
of our database - users of our database can use it like a Java library.
All work is done in transactions, so to use the database, a user would start
a transaction with Database#beginTransaction, then call some of
Transaction's numerous methods to perform selects, inserts, and updates.
For example:
Database db = new Database("database-dir");
try (Transaction t1 = db.beginTransaction()) {
Schema s = new Schema()
.add("id", Type.intType())
.add("firstName", Type.stringType(10))
.add("lastName", Type.stringType(10));
t1.createTable(s, "table1");
t1.insert("table1", 1, "Jane", "Doe");
t1.insert("table1", 2, "John", "Doe");
t1.commit();
}
try (Transaction t2 = db.beginTransaction()) {
// .query("table1") is how you run "SELECT * FROM table1"
Iterator<Record> iter = t2.query("table1").execute();
System.out.println(iter.next()); // prints [1, John, Doe]
System.out.println(iter.next()); // prints [2, Jane, Doe]
t2.commit();
}
db.close();More complex queries can be found in
src/test/java/edu/berkeley/cs186/database/TestDatabase.java.