Scratch Pad

From OPeNDAP Documentation
Revision as of 21:23, 24 April 2009 by Ndp (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
⧼opendap2-jumptonavigation⧽

The DAP Relational Database Server II (DRDS-II) is a (as yet to funded) project that will provide a DAP service layer for relational databases systems. Previously this was (partially) accomplished by a Java servlet known as the DODS Relational Database Server (DRDS).

Overview

Automated catalog generation vs. external configuration

Should the DRDS-II interrogate the database to learn the table and key structures so that it can automatically build a catalog of DAP data objects that represent the database holdings? Or, should the DRDS-II rely (as it's predecessor did) on having the operator develop a DDS/DDX (or some other configuration item) for each data holding?

Data model representation: Tables/Queries as Sequences

In the world of the RDBMS everything is structurally a row set: A collection (tuple) of columns (variables) whose length my not be known until the content is delivered in total. Tables are row sets. A database view is a row set. SQL queries return row sets. SQL JOIN operations take row sets' as input and produce row sets.

Row sets map naturally to the DAP Sequence data type, thus a reasonable representation of an SQL query is a Sequence: A repeating tuple of unknown length.

When a database contains multiple tables or views, each one could be represented as a DAP Sequence. Thus a DDS might contain multiple Sequences each representing a single table or view in the RDBMS:

Dataset {
    Sequence {
        String FirstName;
        String LastName;
        String PhoneNumber;
    } Authors;

    Sequence {
        String Title;
        String Publisher;
        Int16  Pages;
        String   CopyrightDate;
        String   ISBN;
    } Books;

} Inventory;

Databases are typically much more complex. In our example above it would more likely that the Books table contains one or more foreign keys that references the authors of the book. So how to capture this? If we simply add the key field to the Sequence how do users of the system know how to construct queries that can return the data they want? Would this representation allow them to make requests that could return data with the key relationships expressed in the DAP structure (I don't think so) How do we prevent them from making requests the create problems for the RDBMS by creating, via the DAP constraint expression, unreasonable JOIN requests. Additionally, this very flat representation doesn't clearly establish the relationships between the tables.

On possible solution is to represent the databse tables as a series of nested sequences:

Dataset {

    Sequence {
        String Title;
        String Publisher;
        Int16  Pages;
        String   CopyrightDate;
        String   ISBN;
        Sequence {
            String FirstName;
            String LastName;
            String PhoneNumber;
        } Authors;
    } Books;

} Inventory;

Does this actually represent the case of foreign keys?

Tasks

A server that adds a DAP interface to a RDBMS needs:

  1. A component that can translate a DAP query into an SQL query string. This might happen at the string level by simply converting one to the other. The DRDS accomplished this by processing the DAP query string (marking variables in a DDS instance as projected and parsing the selection part of the constraint into Clause objects) and then working with the in memory objects (DDS and Clause objects) to build the SQL query.
  2. A mechanism through which the RDBMS holdings (a collection of one or more tables and views) can be represented as a collection of DAP data sets. For this document we will refer to this as catalog generation. This is inexorably connected to the type mapping activities explained next.
  3. A mapping between the data types found in the RDBMS to the appropriate DAP data types. For many simple/atomic type this is simple: An SQL FLOAT maps naturally to a DAP Float64. For other types, such as an database table with foreign keys the mapping is less straightforward. This activity s further complicated by the fact that nearly every RDBMS implementation contains a number of useful, yet proprietary data types. Thus a proper data model might have to be generated for each RDBMS encountered.

Previous Design/Implementation

The first DRDS (DODS Relational Database Server) was implemented as a Java Servlet. It relied on the Java-DODS implementation of the DAP to provide the DAP type classes with constraint expression evaluation and data value serialization. Some documentation can be found here:


Important features/activities:

  • The server used filesystem directories containing DDS and DAS documents to represent the holdings in the RDBMS. These files were used to generate the DDS and DAS responses.
  • The opendap.servers.sql.sqlCEEval.getSQLQueries() method that takes a constrained DDS object and interrogates it to form an SQL query.
  • Each table or view in the database could only be represented as a data set containing a singe DAP Sequence.
  • The individual DAP type classes implement the read() and serialize() methods that extracts values from the SQL query result and write the values to the client. The DAP Sequence implementation manages which "row" is being read from the SQL response object.

Despite its limitations and lack of ongoing support the DRDS was useful for a number of data centers and continues to be used today. However, the DRDS had many drawbacks:

  • It required the operator to engage in a significant configuration steps:
    • Locating a JDBC driver for their database implementation.
    • Configuring the DRDS to correctly use the JDBC driver.
    • Hand writing the DDS and DAS documents for each table or view in the database for which DAP 2 services were to be provided. This in effect placed the operator of the server in charge of mapping the different types of data held in the database tables into the different types in the DAP data model. The DRDS implementation only allowed certain mappings (as many of the possible combinations were non-sensical), thus if the operator made a poor mapping (For example creating a floating point variable in the DDS and then trying to read a SQL CHAR type into it) the server would not function correctly.
  • It could not automatically build a catalog of the RDBMS holdings. (This was done by the operator as described above)
  • The representation of the RDBMS holds was limited to a single table per DAP data set. Thus each DDS in the catalog contained a single Sequence representing a single table or view from within the RDBMS. Since most RDBMS holds are a collection of tables linked together through keys, this single Sequence representation was a very "flat" view of the RDBMS which many users found particularly restrictive. Making different views of the database alleviated the effects of this limitation to a certain extent, but typically caused some (many?) data values to be sent in a repetitive manner.

Data model representation

Mapping the a the SQL data model to the DAP data model is problematic due to the diversity of the implementations of an SQL database.


Atomic (Simple) Types

Tables as Sequences

Single Tables

Multiple Tables

Multiple Tables as Nested Sequences

Other Complex Types

Relational Database Data Types

Accessing data in an RDBMS is often a many layered affair. Many RDBMS implementations provide a native API library that can be used with a common programming language such as C, C++, C#, Visual Basic, or even Java. These API's interact directly with the RDBMS and provide the maximum level of flexibility, control and access. If one was using a native API to work with a RDBMS to get data then there would be a mapping from the native RDBMS data types to the language data types:


Native RDBMS Data Types ---> C++ Data Types


However, most software that works with RDMSs does so through a database connectivity API. The two most common are the Open Database Connectivity (ODBC) API and the Java Database Connectivity (JDBC) API. Using one of these database connectivity APIs adds a layer of data type mapping to the process:


Native RDBMS Data Types ---> Database Connectivity API Data Types ---> C++ Data Types


To complete the picture, generating DAP data objects requires the finally mapping from the programming language data types in to the DAP data model:


Native RDBMS Data Types ---> Database Connectivity API Data Types ---> C++ Data Types ---> DAP Data Types


For simple/atomic types such as integers an booleans this doesn't require much thought. For more complex types it may require both thought ad careful development to preserve the original semantics of the RDBMS data type across the various mappings and into the appropriate DAP data type. Even then the DAP data model may not have a data type that directly captures the semantics of the original type. For example PostgreSQL has a number of geometric types (such as point, box, and circle) whose content can be held in DAP types, but the semantic meaning (these two floating point values represent a point on a cartesian plain) is not explicit in the DAP object holding the values. That information can only be included in the metadata associated with the instance of the DAP variable.



Database Connectivity APIs





ODBC Data Types





JDBC Data Types

The DRDS used a JDBC connection to retrieve data from a relational database. It is important to note that their are several layers of type translation happening in this:

Database -> JDBC -> Java -> DODS

The Database types are the native types for the particular database that is being read from. The translation from Database->JDBC was handled before the data arrived at the DRDS (most likely by the JDBC Drivers). The mapping of JDBC type to DODS types (the intermediate Java types happen in the process) looked like this:

Mapping from JDBC Types to DODS Types
JDBC Type DAP Type
TINYINT DByte
SMALLINT DInt16
INTEGER DInt32
BIGINT DInt32 **NO SENSIBLE MAPPING (Need DInt64)
REAL DFloat32
FLOAT DFloat64
DOUBLE DFloat64
DECIMAL DFloat64 **NO SENSIBLE MAPPING (Need Some Kind Monsterous Floating point value)
NUMERIC DFloat64 **NO SENSIBLE MAPPING (ibid)
BIT DBoolean
CHAR DString
VARCHAR DString
LONGVARCHAR Implemented to be read into a DString, although it is a "BLOB" type and might be better represented as a DArray(of bytes).
BINARY DArray(of bytes)
VARBINARY DArray(of bytes)
LONGVARBINARY DArray(of bytes)
DATE DString
TIME DString
TIMESTAMP DString

Native SQL Data Types by Implementation





PostgreSQL Data Types

Todo:
 !! This section needs to be updated and otherwise corrected. !!
Boolean and Binary Types
Data Type Description Standardization Logical DAP data type association.
boolean, bool A single true or false value. SQL99 Boolean
bit(n) An n -length bit string (exactly n binary bits). SQL92 None
bit varying(n), varbit(n) A variable n -length bit string (up to n binary bits) SQL92 None
Character Types
Data Type Description Storage Standardization Logical DAP data type association.
character (n ), char(n ) A fixed n -length character string. (4+n) bytes SQL89 String
character varying(n), varchar(n) A variable length character string of up to n characters. Up to (4+n) bytes SQL92 String
text A variable length character string, of unlimited length. Variable PostgreSQL-specific String
Numeric Types
Data Type Description Storage Standardization Logical DAP data type association.
smallint, int2 A signed 2-byte integer 2 bytes SQL89 Int16
integer, int, int4 A signed, fixed-precision 4-byte number. 4 bytes SQL92 Int32
bigint, int8 A signed 8-byte integer, up to 18 digits in length. 8 bytes PostgreSQL-specific None (Need Int64)
real, float4 A 4-byte floating point number. 4 bytes SQL89 Float32
double precision, float8, float An 8-byte floating point number 8 bytes SQL89 Float64
numeric(p,s), decimal(p,s) An exact numeric type with arbitrary precision p, and scale s. Variable SQL99 None
money A fixed precision, U.S.-style currency. 4 bytes PostgreSQL-specific, deprecated. None
serial An auto-incrementing 4-byte integer. 4 bytes PostgreSQL-specific. None
Date and Time Types
Data Type Description Storage Standardization Logical DAP data type association.
date A calendar date (day, month, year). 4 bytes SQL92 String
time The time of day. 4 bytes SQL92 String
time with time zone The time of day, including time zone information. 4 bytes SQL92 String
timestamp (includes time zone) Both date and time. 8 bytes SQL92 String
interval An arbitrary specified length of time 12 bytes SQL92 String
Geometric Types
Data Type Description Storage Standardization Logical DAP data type association.
box A rectangular box in a 2D plane. 32 bytes PostgreSQL-specific None ([2][2] Array of Float32)
line An infinite line in a 2D plane. ?? bytes PostgreSQL-specific None ([2][2] Array of Float32)
lineseg A finite line segment in a 2D plane. 32 bytes PostgreSQL-specific None ([2][2] Array of Float32)
circle A circle with center and radius. 24 bytes PostgreSQL-specific None ([3]Array of Float32)
path Open and closed geometric paths in a two-dimensional plane . 4+32*n bytes PostgreSQL-specific None (Array of Float32)
point geometric point in a 2D plane 16 bytes PostgreSQL-specific None ([2] Array of Float32)
polygon A closed geometric path in a 2D plane 4+32*n bytes PostgreSQL-specific None (Array of Float32)
Network Types
Data Type Description Standardization Logical DAP data type association.
cdir An IP network specification PostgreSQL-specific None
inet A network IP address, with optional subnet bits. PostgreSQL-specific None
macaddr A MAC address (e.g., an Ethernet card's hardware address). PostgreSQL-specific None
System Types
Data Type Description Standardization Logical DAP data type association.
oid An object (row) identifier. PostgreSQL-specific None
xid A transaction identifier PostgreSQL-specific None





Transact-SQL (Microsoft)

Todo:
 !! This section needs to be updated and otherwise corrected. !!
Type Description Storage Bytes DAP equiv.
bigint Integer (whole number) data from -2^63 (-9,223,372,036,854,775,808) through 2^63-1 (9,223,372,036,854,775,807). 8 bytes none
int Integer (whole number) data from -2^31 (-2,147,483,648) through 2^31 - 1 (2,147,483,647). 4 bytes
numeric Functionally equivalent to decimal.
decimal Fixed precision and scale numeric data from -10^38 +1 through 10^38 –1.
bit Integer data with either a 1 or 0 value
smallint Integer data from -2^15 (-32,768) through 2^15 - 1 (32,767). 2 bytes
tinyint Integer data from 0 through 255 1 bytes
smallmoney Monetary data values from -214,748.3648 through +214,748.3647, with accuracy to a ten-thousandth of a monetary unit.
money Monetary data values from -2^63 (-922,337,203,685,477.5808) through 2^63 - 1 (+922,337,203,685,477.5807), with accuracy to a ten-thousandth of a monetary unit.
float Floating precision number data with the following valid values: -1.79E + 308 through -2.23E - 308, 0 and 2.23E + 308 through 1.79E + 308. 4 or 8 bytes
real Floating precision number data with the following valid values: -3.40E + 38 through -1.18E - 38, 0 and 1.18E - 38 through 3.40E + 38. 4 bytes
date
datetimeoffset
datetime2
smalldatetime Date and time data from January 1, 1900, through June 6, 2079, with an accuracy of one minute.
datetime Date and time data from January 1, 1753, through December 31, 9999, with an accuracy of three-hundredths of a second, or 3.33 milliseconds.
time
char Fixed-length non-Unicode character data with a maximum length of 8,000 characters.
varchar Variable-length non-Unicode data with a maximum of 8,000 characters.
next Variable-length non-Unicode data with a maximum length of 2^31 - 1 (2,147,483,647) characters
nchar Fixed-length Unicode data with a maximum length of 4,000 characters
nvarchar Variable-length Unicode data with a maximum length of 4,000 characters. sysname is a system-supplied user-defined data type that is functionally equivalent to nvarchar(128) and is used to reference database object names.
ntext Variable-length Unicode data with a maximum length of 2^30 - 1 (1,073,741,823) characters.
binary Fixed-length binary data with a maximum length of 8,000 bytes.
varbinary Variable-length binary data with a maximum length of 8,000 bytes.
image Variable-length binary data with a maximum length of 2^31 - 1 (2,147,483,647) bytes.
cursor A reference to a cursor.
timestamp A database-wide unique number that gets updated every time a row gets updated.
hierarchyid
uniquieidentifier A globally unique identifier (GUID).
sql_variant
xml
table

Template

Types

Data Type Description Standardization Logical DAP data type association.

Desired Features

Implementation Target