Assignment 3 - Record Manager
In this assignment you are creating a record manager. The record manager handles tables with a fixed schema. Clients can insert records, delete records, update records, and scan through the records in a table. A scan is associated with a search condition and only returns records that match the search condition. Each table should be stored in a separate page file and your record manager should access the pages of the file through the buffer manager implemented in the last assignment.
Hints: This assignment is much more complex than the previous assignments and it is easy to get stuck if you are unclear about how to structure your solution and what data structures to use. Sit down with a piece of paper first and design the data structures and architecture for your implementation.
-
Record Representation: The data types we consider for this assignment are all fixed length. Thus, for a given schema, the size of a record is fixed too.
-
Page Layout: You will have to define how to layout records on pages. Also you need to reserve some space on each page for managing the entries on the page. Refresh your memory on the page layouts discussed in class! For example, how would you represent slots for records on pages and manage free space.
-
Table information pages: You probably will have to reserve one or more pages of a page file to store, e.g., the schema of the table.
-
Record IDs: The assignment requires you to use record IDs that are a combination of page and slot number.
-
Free Space Management: Since your record manager has to support deleting records you need to track available free space on pages. An easy solution is to link pages with free space by reserving space for a pointer to the next free space on each page. One of the table information pages can then have a pointer to the first page with free space. One alternative is to use several pages to store a directory recording how much free space you have for each page.
tables.h
This header defines basic data structures for schemas, tables, records, record ids (RIDs), and values. Furthermore, this header defines functions for serializing these data structures as strings. The serialization functions are provided ( rm_serializer.c
). There are four datatypes that can be used for records of a table: integer ( DT_INT
), float ( DT_FLOAT
), strings of a fixed length ( DT_STRING
), and boolean ( DT_BOOL
). All records in a table conform to a common schema defined for this table. A record is simply a record id ( rid
consisting of a page number and slot number) and the concatenation of the binary representation of its attributes according to the schema ( data
).
Schema
A schema consists of a number of attributes ( numAttr
). For each attribute we record the name ( attrNames
) and data type ( dataTypes
). For attributes of type DT_STRING
we record the size of the strings in typeLength
. Furthermore, a schema can have a key defined. The key is represented as an array of integers that are the positions of the attributes of the key ( keyAttrs
). For example, consider a relation R(a,b,c)
where a
then keyAttrs would be [0]
.
Data Types and Binary Representation
Values of a data type are represented using the Value
struct. The value struct represents the values of a data type using standard C data types. For example, a string is a char *
and an integer using a C int
. Note that values are only used for expressions and for returning data to the client of the record manager. Attribute values in records are stored slightly different if the data type is string. Recall that in C a string is an array of characters ended by a 0
byte. In a record, strings are stored without the additional 0
byte in the end. For example, for strings of length 4 should occupy 4 bytes in the data
field of the record.
Interface
|
|
expr.h
This header defines data structures and functions to deal with expressions for scans. These functions are implemented in expr.c
. Expressions can either be constants (stored as a Value
struct), references to attribute values (represented as the position of an attribute in the schema), and operator invocations. Operators are either comparison operators (equals and smaller) that are defined for all data types and boolean operators AND
, OR
, and NOT
. Operators have one or more expressions as input. The expression framework allows for arbitrary nesting of operators as long as their input types are correct. For example, you cannot use an integer constant as an input to a boolean AND operator. As explained below, one of the parameters of the scan operation of the record manager is an expression representing the scan condition.
Interface
|
|
record_mgr.h
We now discuss the interface of the record manager as defined in record_mgr.h
. There are five types of functions in the record manager: functions for table and record manager management, functions for handling the records in a table, functions related to scans, functions for dealing with schemas, and function for dealing with attribute values and creating records. We now discuss each of these function types.
Table and Record Manager Functions
Similar to previous assignments, there are functions to initialize and shutdown a record manager. Furthermore, there are functions to create, open, and close a table. Creating a table should create the underlying page file and store information about the schema, free-space, … and so on in the Table Information pages. All operations on a table such as scanning or inserting records require the table to be opened first. Afterwards, clients can use the RM_TableData struct to interact with the table. Closing a table should cause all outstanding changes to the table to be written to the page file. The getNumTuples
function returns the number of tuples in the table.
Record Functions
These functions are used to retrieve a record with a certain RID
, to delete a record with a certain RID
, to insert a new record, and to update an existing record with new values. When a new record is inserted the record manager should assign an RID
to this record and update the record
parameter passed to insertRecord
.
Scan Functions
A client can initiate a scan to retrieve all tuples from a table that fulfill a certain condition (represented as an Expr
). Starting a scan initializes the RM_ScanHandle
data structure passed as an argument to startScan
. Afterwards, calls to the next
method should return the next tuple that fulfills the scan condition. If NULL
is passed as a scan condition, then all tuples of the table should be returned. next
should return RC_RM_NO_MORE_TUPLES
once the scan is completed and RC_OK
otherwise (unless an error occurs of course). Below is an example of how a client can use a scan.
|
|
Closing a scan indicates to the record manager that all associated resources can be cleaned up.
Schema Functions
These helper functions are used to return the size in bytes of records for a given schema and create a new schema.
Attribute Functions
These functions are used to get or set the attribute values of a record and create a new record for a given schema. Creating a new record should allocate enough memory to the data
field to hold the binary representations for all attributes of this record as determined by the schema.
Interface
|
|
Optional Extensions
You can earn up to 20 bonus points for implementing optional extensions. A good implementation of one or two extensions will give you the maximum of 20 points. So rather then implementing 5 incomplete extensions, I suggest you to focus on one extension first and if there is enough time, then add additional ones.
-
TIDs and tombstones: Implement the TID and Tombstone concepts introduced in class. Even though your implementation does not need to move around records, because they are fixed size, TIDs and Tombstones are important for real systems.
-
Null values: Add support for SQL style
NULL
values to the data types and expressions. This requires changes to the expression code, values, and binary record representation (e.g., you can use the NULL bitmaps introduced in class). -
Check primary key constraints: On inserting and updating tupes, check that the primary key constraint for the table holds. That is you need to check that no record with the same key attribute values as the new record already exists in the table.
-
Ordered scans: Add an parameter to the scan that determines a sort order of results, i.e., you should pass a list of attributes to sort on. For dare-devils: Implement this using external sorting, so you can sort arbitrarily large data.
-
Interactive interface: Implement a simple user interface. You should be able to define new tables, insert, update, and delete tuples, and execute scans. This can either be a shell or menu-based interface.
-
Conditional updates using scans: Extend the scan code to support updates. Add a new method
updateScan
that takes a condition (expression) which is used to determine which tuples to update and a pointer to a function which takes a record as input and returns the updated version of the record. That is the user of theupdateScan
method should implement a method that updates the record values and then pass this function toupdateScan
. Alternatively, extend the expression model with new expression types (e.g., adding two integers) and letupdateScan
take a list of expressions as a parameter. In this case the new values of an updated tuple are produced by applying the expressions to the old values of the tuple. This would closer to real SQL updates.
Source Code Structure
You source code directories should be structured as follows. You should reuse your existing storage manager and buffer manager implementations. So before you start to develop, please copy your storage manager and buffer manager implementations.
Put all source files in a folder assign3
in your git repository
This folder should contain at least …
-
the provided header and C files
-
a make file for building your code
Makefile
-
a bunch of
*.c
and*.h
files implementing the record manager -
README.txt
/README.md
: A markdown or text file with a brief description of your solution
E.g., the structure may look like that:
git assign3 README.md Makefile buffer_mgr.h buffer_mgr_stat.c buffer_mgr_stat.h dberror.c dberror.h expr.c expr.h record_mgr.h rm_serializer.c storage_mgr.h tables.h test_assign3_1.c test_expr.c test_helper.h
Test Cases
test_helper.h
Defines several helper methods for implementing test cases such as ASSERT_TRUE
.
test_expr.c
This file implements several test cases using the expr.h
interface. Please let your make file generate a test_expr
binary for this code. You are encouraged to extend it with new test cases or use it as a template to develop your own test files.
test_assign3_1.c
This file implements several test cases using the record_mgr.h
interface. Please let your make file generate a test_assign3
binary for this code. You are encouraged to extend it with new test cases or use it as a template to develop your own test files.