Copyright (C) 2003-2006 Jeffrey I Cohen. All rights reserved, worldwide. NAME Genezzo - an extensible database system DESCRIPTION The Genezzo modules implement a hierarchy of persistent hashes using a fixed amount of memory and disk. This system is designed to be easily configured and extended with custom functions, persistent storage representations, and novel data access methods. In its current incarnation it supports a limited subset of SQL with a command line tool [gendba] and a partial DBI-style interface. REQUIREMENTS Perl 5.6 INSTALLATION If you can read this, then you figured out how to uncompress and untar the archive. perl Makefile.PL # constructs the makefile make make test make install # (as superuser [optional]) If you install Genezzo, you should be able to run the line-mode tool directly by typing "gendba.pl" at the command line. You also should be able to run scripts from the directory containing this file, something like: perl -Iblib/lib lib/Genezzo/gendba.pl USAGE line mode available via (/usr/bin) gendba.pl. Try: gendba.pl -help gendba.pl -man or: perl -Iblib/lib lib/Genezzo/gendba.pl -help perl -Iblib/lib lib/Genezzo/gendba.pl -man for more information. The GenDBI module supports a subset of DBI-style calls. See dbi_example.pl for some sample usage. Try: perl -Iblib/lib lib/Genezzo/dbi_example.pl # runs an example perldoc Genezzo::GenDBI for more information. The simplest way to create an instance of a Genezzo database is to use: gendba.pl -init This command will create a new database and login to the command-line. SQL SUPPORT -- see LIMITATIONS EXPRESSIONS: The Genezzo parser can handle expressions composed of column names, numeric or quoted string constants, logical constants like TRUE, FALSE and NULL, concatenation, scalar functions, and basic arithmetic like '+', '-', '*', '/'. However, logical expressions using AND, OR, and NOT are only legal in the WHERE clause. INSERT: INSERT INTO tablename VALUES (expr [, expr ...]); INSERT INTO tablename (colname [, colname ...]) VALUES (expr [, expr ...]); INSERT INTO tablename SELECT ... FROM ... UPDATE: UPDATE tablename SET colname = expr [, colname = expr ...] [WHERE ...] DELETE: DELETE FROM tablename [WHERE ...] SELECT: The Genezzo parser can parse complex SELECT statements, but the execution engine can only handle simple SELECTS and basic joins. SELECT * FROM tablename [WHERE ...] SELECT COUNT(*) FROM tablename SELECT ECOUNT(*) FROM tablename SELECT * FROM t1, t2 WHERE t1.col1 = t2.col1 SELECT expr [[AS] column_alias] [, expr [[AS] column_alias]] FROM tablename [WHERE ...] CREATE TABLE: CREATE TABLE tablename (colname coltype [, colname coltype ...]); CREATE INDEX: CREATE INDEX iname on tablename (colname [, colname ...]); ALTER TABLE ADD CONSTRAINT: supports primary and unique key and check constraints. NOT NULL and foreign keys are not currently supported, though you can implement NOT NULL as CHECK (colname IS NOT NULL). The CHECK expression is equivalent to a WHERE clause. A table can have multiple CHECK constraints, but currently only one key constraint. ALTER TABLE tablename ADD [CONSTRAINT constraint_name] PRIMARY KEY (colname [, colname ...]); ALTER TABLE tablename ADD [CONSTRAINT constraint_name] UNIQUE (colname [, colname ...]); ALTER TABLE tablename ADD [CONSTRAINT constraint_name] CHECK (conditional_expression); ADDFILE: Not really a SQL command, but an extension to grow the datafiles. See "addfile help". DROP: drop a table or index. COMMIT: write changes to disk. SYNC: write changes to disk like "commit", but don't end a transaction. ROLLBACK: discard all uncommitted changes. DESCRIBE: DESCRIBE tablename EXPLAIN PLAN: dump parsing and plan information for any SQL command, including CREATE and ALTER TABLE. Does not require a "plan table". Can perform partial explanations for invalid statements, such as missing tables/columns, etc. Example - "explain plan for select * from t1;" or even complex sql like: Explain plan for select * from t1 where c1 in (select avg(c2) from t2) having avg(c3) > ((select min(c4) from (select c5 from t3 union select c6 from t4) ) + 5) * 7 / min(c7) ; STARTUP/SHUTDOWN: LIMITATIONS General: No locking/multiuser support, bind variables, incomplete DBI support. SELECT: Simple queries only -- no ORDER BY, GROUP BY, HAVING, subqueries, set functions (UNION, INTERSECT). Only basic FROM clause joins: FROM t1, t2, ... are supported. No aggregation functions other than COUNT(*), ECOUNT(*). COUNT and ECOUNT do not accept a WHERE clause. ALTER TABLE: no foreign key, NULL/NOT NULL constraints. CREATE TABLE: Only numeric and character datatypes are accepted. Datatype length and precision are not enforced. Numeric columns will accept arbitrary character input, but will attempt to use numeric (versus character) comparisons. No table or column constraints. No CREATE TABLE AS SELECT. EXAMPLES Some very basic SQL create/drop/describe table, select, insert, update and delete syntax is supported, and like standard SQL, table and column names are case-insensitive unless quoted. More complex SQL parses, but is ignored. The only supported functions are count(*) and ecount(*), a non-blocking count estimation function. The database also supports commit to force changes to disk, and a crude rollback. NOTE: Data definition (such as create table or ct) must be manually committed to keep the database in a consistent state. Uncommitted inserts and updates will only be block-consistent -- there is no guarantee that the data will get flushed to disk, and no guarantee whether the changes will or will not take effect. rem Some simple SELECTs select * from _col1; select rid, rownum, tname, colname from _col1; select count(*) from _col1; select ecount(*) from _col1; rem SELECTs with WHERE, perl and SQL style. select * from _tab1 where tname =~ m/col/; select * from _tab1 where tid < 5; select * from _tab1 where numcols > 3 AND numcols < 6; rem basic joins select * from _tab1, _col1 where _tab1.tid = _col1.tid and _tab1.tname =~ m/col/; rem Column aliases and Expressions select tid as Table_ID, tname Name, (tid+1)/2 from _tab1; rem Basic INSERT insert into test1 values ('a','b','c','d'); insert into test1(col2, col1) values ('a','b','c','d'); rem CREATE TABLE and INSERT...SELECT create table test2(col1 char, col2 char); insert into test2 (col1) select col1 from test1; rem DELETE with WHERE delete from test1 where col1 < 'bravo' and col2 > 5; rem UPDATE with WHERE (no subqueries supported) update test2 set col2 = 'foo' where col2 is null; commit By default, the database will autostart -- the gendba.pl command will connect to an active, updateable instance. The shutdown command will reset the database to a read-only mode that only provides read access to certain dictionary tables. The startup command will restart the database. The Genezzo parser also supports the Feeble query language, as described by gendba.pl -man. Briefly, the commands are ct (create table), dt (drop table), i (insert), u (update), d (delete), s (select). The following example creates a table EMP with two columns, ENAME (as character) and ID (as number), inserts five rows, and counts them. ct EMP ENAME=c ID=n i EMP bob 1 orville 2 wilbur 3 harry 4 luke 5 s EMP count s EMP ecount commit While the SQL update and delete commands support a WHERE clause, the Feeble "u" and "d" commands only work via rowid. For example: gendba 10> s EMP rid rownum * rid rownum ename id ___ ______ _____ __ 1/33/1 1 bob 1 1/33/2 2 orville 2 1/33/3 3 wilbur 3 1/33/4 4 harry 4 1/33/5 5 luke 5 5 rows selected. gendba 11> d EMP 1/33/1 deleted 1 rows from table emp. gendba 12> u EMP 1/33/5 margo 66 updated 1 rows in table emp. gendba 13> !10 s EMP rid rownum * rid rownum ename id ___ ______ _____ __ 1/33/2 1 orville 2 1/33/3 2 wilbur 3 1/33/4 3 harry 4 1/33/5 4 margo 66 4 rows selected. The default database is stored in a fixed-size 160K file. You can change the default filesize and the blocksize at database initialization time. For example: gendba.pl -init -define blocksize=8k -define dbsize=10M to create a database with a 10 megabyte datafile using 8K blocks. By default, this file is a fixed size, but it can be set to grow using the "increase_by" settings added in version 0.18. Also, you can add additional files to your database using the "addfile" command. By default, each file is double the size of the previous. Type "addfile help" for more options. NEW FEATURES - Indexed Tables Version 0.13 introduces an indexed-table type. An indexed table treats the first column in the table definition as a unique, not-null primary key. Note that an "Indexed Table" is distinct from an ordinary table with a separate index, where you can assign a separate index object to a particular column or group of columns. gendba 9> ct EMP index id=n name=c Create Table : emp with unique index option tablename : emp column id : n (primary key) column name : c table emp created gendba 10> i EMP 1 bob 2 orville 3 wilbur 4 harry 5 luke inserted 5 rows into table emp. gendba 11> s EMP rid * rid id name ___ __ ____ 1 1 bob 2 2 orville 3 3 wilbur 4 4 harry 5 5 luke 5 rows selected. gendba 12> i EMP 1 bob 2 orville 3 wilbur 4 harry 5 luke whisper: duplicate key 1 Failed to insert row 1 in table emp at lib/Genezzo/gendba.pl line 230 inserted 0 rows into table emp. gendba 13> select tid, tname, otype from _tab1 where tname = 'emp' tid tname otype ___ _____ _____ 14 emp IDXTAB 1 row selected. Notice that the first column is marked as a primary key when the table is created, and the rids for indexed tables are the primary key, not a physical block address. Indexed tables have an object type of IDXTAB in the data dictionary -- conventional tables are marked as TABLE. WHERE clause processing has been enhanced to use equality predicates (e.g. "SELECT * from EMP where id = 2") to drive index scans. Any series of "AND"ed predicates is a candidate for an efficient index search, but the use of an OR currently requires a full table scan. CAVEATS: The index performance is pretty poor, for a couple reasons. It should improve a lot. NEW FEATURES - Indexes Users can now define their own indexes. The dictionary automatically maintains indexes on its own tables. In order to create an index called EMP_I1 on the ID column of table EMP, the command is: Create Index EMP_I1 on EMP (ID) You can have arbitrary combinations of numeric and character keys. RESTRICTIONS: only one index per table currently. Index keys must be less than one-half of the db blocksize. In version 0.15, the query engine will use the index to drive the row fetch for a SELECT if the WHERE clause specifies an equality predicate on the primary key. For example: Select * from EMP where ID = 5 uses an index. In general, additional predicates may accompany the equality expression as long as they are ANDed: Select * from EMP where ID = 5 and ENAME =~ m/BOB/ However, the presence of ORs causes the plan to revert to a full table scan: Select * from EMP where ID = 5 or ID = 6 All user-defined indexes are non-unique. Unique indexes should be created as unique constraints (fixed in version 0.22) using ALTER TABLE. NEW FEATURES - automatic tablespace and file growth In version 0.18, datafiles are no longer a fixed size. The column "increase_by" in table _tsfiles controls the behavior: if this column is blank, null, or zero, the datafile size is fixed. If "increase_by" is a non-zero number, the datafile will attempt to grow by that many bytes if it runs out of free extents. If "increase_by" is a percentage like "50%", the datafile is extended by the current size multiplied by this percentage. The AddFile command now takes "increase_by" as an optional argument. Also in version 0.18, each tablespace can be configured to automatically create new datafiles when the current datafile runs out of free extents. The column "addfile" in table _tspace controls the behavior: if the column is blank or null, then the tablespace must be extended manually. If the "addfile" column contains a valid argument list for the AddFile command, then this command will get invoked when all of the current datafiles are full. Note that the datafiles can be of a fixed size, or use the new "increase_by" parameter to grow to some maximum size (currently 2G). The script "bigdb.sql" can be used to reset your database for automatic growth. It can be invoked as "@bigdb.sql" from the gendba command line. NEW FEATURES - ALTER TABLE ADD CONSTRAINT Version 0.20 supports ALTER TABLE ADD CONSTRAINT CHECK (). The check-text is any basic where clause predicate. Example: gendba 3> create table testcon (col1 c, col2 n); table testcon created gendba 4> alter table testcon add constraint cn1 check ((col2 > 5) AND col1 =~ m/(a|b|c)/) Added constraint cn1 to table testcon gendba 5> insert into testcon values ('a', 7); inserted 1 row into table testcon. gendba 6> insert into testcon values ('b', 8); inserted 1 row into table testcon. gendba 7> insert into testcon values ('d', 8); WARNING: violated constraint cn1 must satisfy (col2 > 5) AND col1 =~ m/(a|b|c)/ WARNING: Failed to insert row 1 in table testcon inserted 0 rows into table testcon. gendba 8> insert into testcon values ('a', 1); WARNING: violated constraint cn1 must satisfy (col2 > 5) AND col1 =~ m/(a|b|c)/ WARNING: Failed to insert row 1 in table testcon inserted 0 rows into table testcon. Version 0.22 supports ALTER TABLE ADD CONSTRAINT UNIQUE () and ALTER TABLE ADD CONSTRAINT PRIMARY KEY ( Version 0.22 only supports a single constraint or single index per table, but version 0.23 supports multiple constraints and indexes on a table. Need to extend to fully support NULL/NOT NULL, FOREIGN KEY, plus ENABLE/DISABLE. NEW FEATURES - System/User Extensibility Version 0.25 of Genezzo supports database extensibility. See Genezzo::Havok for more details. The script "havok.sql" (now deprecated) can be used to add extensibility tables to your database. In version 0.58 a new function HavokUse supercedes the havok.sql script. In particular, the table user_extend can be used to load new functions or perl modules into Genezzo. For example, the havok.sql script loads "Text::Soundex", so queries can use Text::Soundex::soundex in the WHERE clause. Version 0.28 of Genezzo slightly modifies the havok and user_extend tables, and it provides the capability to import functions into the default Genezzo namespace. For example, if the "soundex" function is imported, then the query can just call soundex(), versus using the fully-qualified name Text::Soundex::soundex. NEW FEATURES - Improved error reporting, handling Version 0.31 error reporting has been significantly reworked to be more compatible with DBI. In particular, Genezzo::GenDBI->connect now takes an optional attribute hash with PrintError and RaiseError, so it will print messages using warn() or die(). In addition, it also supports a GZERR argument is used to replace the internal Genezzo error handlers. More documentation exists in GenDBI.pm. The DBI standard err and errstr aren't supported yet. NEW FEATURES - backward compatiblity Version 0.32 supports a compatibility matrix - it can support version 0.31 databases. Genezzo can now maintain a list of which versions use the same database format. Need to extend this capability to get automatic upgrade between different database revisions. NEW FEATURES - database export, rollback, quoted strings, dynamic system extensions Version 0.33 introduces genexpl.pl, a very, very basic data export utility. A "rollback" command is now available, which will discard all in-memory changes and replace them with the previous committed results. Note that the buffer cache will automatically flush data to disk when it is full, so rollback will not be able to restore data in this case. Single-quoted strings in SQL expressions now support backslash quoting, so 'foo\\bar\'s' is interpreted as foo\bar's. Finally, a new Havok module called SysHook supports extensions to basic Genezzo database functions. SysHook is restructured in version 0.34. NEW FEATURES - multiline SQL statements Version 0.38 allows multiline SQL statements for SELECT, INSERT, UPDATE, DELETE, and EXPLAIN in the gendba interactive line mode. These statements must be terminated by a semicolon. NEW FEATURES - Improved Parser Version 0.39 uses a new parser based upon Parse::RecDescent for SELECT statements. The new parser is stricter about quoting conventions and identifier names -- standard identifiers must start with an alpha character, or be specified in double quotes. Only identifiers can be unquoted. String literals must be in single quotes. The parser does make an exception for dictionary tables like _tab1 and _col1, which do *not* need to quoted. Boolean operations must use AND and OR, not && and ||. || is the SQL string concatenation operation. The count estimation function ECOUNT must be specified as ecount(*) in the select list. While the parser is smarter, the execution phase is not improved. The only supported aggregation functions are count(*) or ecount(*) in the select list. Having, Group By, and Order By are still missing. Version 0.40 allows arbitrary expressions in SELECT lists, e.g: Select tid+5/2 from _tab1; Version 0.41 adds support for the literals TRUE, FALSE, and NULL, and extends the new parser to cover ALTER TABLE, DELETE, and INSERT. INSERT can now handle arbitrary expressions, e.g: Insert into mytable values (1+1, 2*2, 3/4); Version 0.42 adds arbitrary expression support to UPDATE, and marks the complete switchover to the new parser. Version 0.43 adds case-insensitive table and column names. NEW FEATURES - Raw Device support Version 0.44 and 0.45 add support for raw io devices. NEW FEATURES - Additional User-supplied Preferences and File Header Info In Version 0.47, users can supply arbitrary command-line definitions for the dictionary preference table or fileheader info using -define and -fileheader_define, respectively. NEW FEATURES - Object-Oriented SysHooks In Version 0.47, a single package can load multiple functions as a set of class methods, versus package methods. NEW FEATURES - Tablespaces In Version 0.51, users can create additional tablespaces (collections of datafiles for storage) for tables and indexes. By default, all tables and indexes are created in the SYSTEM tablespace. The command "CREATE TABLESPACE tablespace_name" creates a new, empty tablespace. You must add files to the tablespace use "addfile tsname=..." or "af tsname=..." before it can be used. See "addfile help" for more details. CREATE TABLE and CREATE INDEX now take a new TABLESPACE option. Note that key constraints on a table inherit its tablespace. A table or index can only be stored in a single tablespace, and they cannot be moved to a different tablespace. Additional files may be added to any tablespace at any time. Currently, there is no support for DROP TABLESPACE, or taking tablespace offline. Example: gendba 1> create tablespace ts1 gendba 2> addfile tsname=ts1 gendba 3> create table mytab (col1 char) tablespace ts1 NEW FEATURES - Joins In Version 0.52, basic joins are supported. NEW FEATURES - DUAL table In version 0.58, Genezzo provides a single-row DUAL table which is useful for evaluating expressions, e.g.: gendba 1> select (((5+5)/4)*32) from dual; (((5+5)/4)*32) ______________ 80 1 row selected. NEW FEATURES - HavokUse function for System/User Extensibility In version 0.58, the HavokUse function supercedes the havok.sql and syshook.sql scripts. HavokUse operates on Havok modules in a fashion analogous to the way perl "use" loads perl modules. Havok modules now use a metadata document similar to the MakeMaker META.yml which lists module dependencies, table definitions, etc. For example, to load Genezzo::Havok::SysHook, use the command: Select HavokUse('Genezzo::Havok::SysHook') from dual; If the havok and sys_hook tables do not exist, HavokUse will create and update these tables. If the tables already exist, HavokUse will do nothing, but return successfully. See the Havok module for more details. Note that the "select" statement is now invoking dictionary actions which may create and update new tables, and which may need to be committed. Also, HavokUse only performs partial initialization -- it is still necessary to commit, shutdown and restart to activate the new Havok modules. NEW FEATURES - Havok::UserFunctions, standard SQL scalar functions Previously, all standard perl functions could be invoked in a SQL statement, and additional SQL functions were optionally added via Havok::UserExtend. In version 0.59, Havok::UserExtend has been deprecated and replaced with Havok::UserFunctions, and Genezzo only supports functions defined in the new user_functions table. This table is preloaded when the database is created with many common perl and sql functions. The perl functions which are available in SQL are: chomp, chop, chr, crypt, index, lc, lcfirst, length, ord, pack, reverse, rindex, sprintf, substr, uc, ucfirst, abs, atan2, cos, exp, hex, int, log10 (to avoid confusion between natural log and base10), oct, rand, sin, sqrt, srand, and perl_join (which is renamed to avoid a conflict with the SQL JOIN operation). The SQL string functions are concat, greatest, initcap, least, lower, and upper. The SQL math functions are cosh, ceil, floor, ln, sinh, tan, and tanh (also note many standard SQL math functions are identical to their perl counterparts, which are listed above). The SQL conversion functions are ascii, instr, and nvl. NEW FEATURES - improved genexp.pl The version 0.60 database export (genexp.pl) now correctly handles table constraints and indexes. NEW FEATURES - Feeble/SQL compatible case-sensitivity Prior to version 0.63, Feeble commands (ct, d, u, s) did not match the SQL behavior for case-insensitive identifiers. For example, the Feeble command: "ct EMP COL1=c" would create a table called EMP (uppercase) with a column COL1, while the SQL command: "create table EMP (COL1 char(10))" would create a table called emp (lowercase) with a column col1. This discrepancy would require the use of quoted identifiers in SQL in order to reference uppercase or mixed case table or column names. To facilitate interoperablity between Feeble and SQL, Feeble now matches the SQL usage and legal naming restrictions for unquoted, case-insensitive names, and uppercase or mixed case names are only available using SQL quoted identifiers. Feeble does not support quoted names. INCOMPATIBLE CHANGES Version 0.64 dictionary and block format is incompatible with previous releases. No upgrade path is supported. Version 0.48 thru 0.50 are compatible. SEE ALSO The project homepage at http://www.genezzo.com AUTHOR Original author: Jeffrey I Cohen Copyright (c) 2003, 2004, 2005, 2006 Jeffrey I Cohen. All rights reserved. This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA Address bug reports and comments to: jcohen@genezzo.com For more information, please visit the Genezzo homepage at http://www.genezzo.com