Skip to content

An SQL engine for YottaDB and other M-like databases.

License

Notifications You must be signed in to change notification settings

chrisemunt/mgsql

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

mgsql

An SQL engine for YottaDB and other M-like databases.

Chris Munt [email protected]
8 June 2023, MGateway Ltd http://www.mgateway.com

  • Current Release: Version: 1.4; Revision 23
  • Release Notes can be found at the end of this document.

Contents

Overview

mgsql is an Open Source SQL engine developed primarily for the YottaDB database. It will also work with the GT.M database and other M-like databases.

SQL access is provided via the following routes:

  • Embedded SQL statements in M code.
  • REST.
  • ODBC.

Pre-requisites

The YottaDB database (or similar M database):

   https://yottadb.com/

Installing mgsql

YottaDB

The instructions given here assume a standard 'out of the box' installation of YottaDB deployed in the following location:

   /usr/local/lib/yottadb/r130

The primary default location for routines:

   /root/.yottadb/r1.30_x86_64/r

Copy all the routines (i.e. all files with an 'm' extension) held in the GitHub /yottadb directory to:

   /root/.yottadb/r1.30_x86_64/r

Change directory to the following location and start a YottaDB command shell:

   cd /usr/local/lib/yottadb/r130
   ./ydb

Link all the mgsql routines and check the installation:

   do ylink^%mgsql

   do ^%mgsql

   MGSQL by MGateway Ltd.
   Version: 1.4; Revision 23 (8 June 2023) %mgsql

Note that the version of mgsql is successfully displayed.

InterSystems Cache/IRIS

Log in to the %SYS Namespace and install the mgsql routines held in /isc/mgsql_isc.ro.

   do $system.OBJ.Load("/isc/mgsql_isc.ro","ck")

Change to your development Namespace and check the installation:

   do ^%mgsql

   MGSQL by MGateway Ltd.
   Version: 1.4; Revision 23 (8 June 2023) %mgsql

Other M systems

All routines are held in /m/mgsql.ro, use an appropriate utility to install them in the Manager UCI then change to your development UCI and check the installation:

   do ^%mgsql

   MGSQL by MGateway Ltd.
   Version: 1.4; Revision 23 (8 June 2023) %mgsql

Embedding SQL statements in M code

Before executing SQL statements do familiarise yourself with the M system resources (i.e. globals) used by mgsql. Refer to the Resources used by mgsql section.

The general form for executing SQL statements from within M code (or from the M command line) is as follows:

   set status=$$exec^%mgsql(<schema>,<sql statement>,.%zi,.%zo)

Where:

  • %zi is an M array representing data that needs to be input to the script.
  • %zo is an M array representing parameters controlling output from the script.

In the simplest case, the query output will be written to the primary device (for example, a terminal window), line by line with columns returned as a comma-separated list.

Examples

The top level routine %mgsql (physical file _mgsql.m) contains a number of sample SQL scripts. These work to a simple database representing hospital patients and their associated admissions. View the embedded scripts in this routine.

Create the test schema:

   do create^%mgsql

Insert a few test records:

   do insert^%mgsql

Run the various SQL retrieval scripts:

   do sel1^%mgsql

A number of SQL scripts are available at line labels sel1, sel2, sel3 ... to sel[n].

Supplying variable inputs to a query

You can supply variable inputs to the query via the inputs array - %zi in the examples. The rationale for doing this, as opposed to embedding values directly in the scripts, is that it reduces the amount of compilation time needed. Once the query is compiled the same generated code can be used for all sets of input variables.

Example:

   new %zi,%zo
   set %zi("number")=100000
   set status=$$exec^%mgsql("","select * from patient where num = :number",.%zi,.%zo)

Using the SQL output spool file

Rather than dumping query output to the current device, it is possible to mandate that output is directed to the spool file, which is a global called mgsqls. To do this, specify a query statement ID in the input array.

Example:

   new %zi,%zo
   set %zi(0,"stmt")="all patients"
   set status=$$exec^%mgsql("","select * from patient",.%zi,.%zo)

The query will execute silently and write the output to the spool file, the structure of which is as follows:

   ^mgsqls($Job, <statement_ID>, 0, <row_no>, <column_no>)=<value> 

Where $Job is the current process ID.

Using an SQL output callback

Finally, it is possible to mandate that query output is directed to a callback – an M function defined in the application. To do this, specify the name of the callback function (including the routine name) in the input array.

The callback function must be defined as follows:

   callback(%zi, %zo, rn)

Where:

  • %zi is the query input array.
  • %zo is the query output array.
  • rn is the current row number.

Example:

               new %zi,%zo
               set %zi(0,"callback")="allpatients^thisroutine"
               set status=$$exec^%mgsql("","select * from patient",.%zi,.%zo)
               quit
               ;
   allpatients(%zi,%zo,rn) ; query callback
               new stop
               set stop=0
               ; process the row of output here
               Kill %zo(rn) ; we don’t want to keep the current row
               quit stop

Where rn is the current row number and stop is a stop flag. Set the stop flag to 1 in the callback function to force the early termination of the query. The data for the row is held in the output array (%zo), the structure of which is as follows:

   %zo(<row_no>, <column_no>)=<value>

Starting the mgsql network Superserver

So far we have covered the basics of executing SQL statements from M code. In order to execute SQL queries over REST or ODBC the mgsql installation must be accessible over the network. The Superserver service described here will concurrently support access to mgsql via REST and ODBC. The default TCP server port for mgsql is 7041. If you wish to use an alternative port then modify the following instructions accordingly.

For most M systems, the mgsql Superserver can be started from the M command prompt. For YottaDB there is the option of starting Superserver child processes via the xinetd service.

  • Note that if you are using the generic MGateway Superserver (%zmgsi) then no action is required here as the generic Superserver is able to serve mgsql.

Starting the mgsql Superserver

Start the M-hosted concurrent TCP service in the Manager UCI:

   do start^%mgsql(0) 

To use a server TCP port other than 7041, specify it in the start-up command (as opposed to using zero to indicate the default port of 7041).

Starting YottaDB Superserver child processes via xinetd

Network connectivity to YottaDB is managed via the xinetd service. First create the following launch script (called mgsql_ydb here):

   /usr/local/lib/yottadb/r130/mgsql_ydb

Content:

   #!/bin/bash
   cd /usr/local/lib/yottadb/r130
   export ydb_dir=/root/.yottadb
   export ydb_dist=/usr/local/lib/yottadb/r130
   export ydb_routines="/root/.yottadb/r1.30_x86_64/o*(/root/.yottadb/r1.30_x86_64/r /root/.yottadb/r) /usr/local/lib/yottadb/r130/libyottadbutil.so"
   export ydb_gbldir="/root/.yottadb/r1.30_x86_64/g/yottadb.gld"
   $ydb_dist/ydb -r xinetd^%mgsql

Create the xinetd script (called mgsql_xinetd here):

   /etc/xinetd.d/mgsql_xinetd

Content:

   service mgsql_xinetd
   {
        disable         = no
        type            = UNLISTED
        port            = 7041
        socket_type     = stream
        wait            = no
        user            = root
        server          = /usr/local/lib/yottadb/r130/mgsql_ydb
   }
  • Note: sample copies of mgsql_xinetd and mgsql_ydb are included in the /unix directory.

Edit the services file:

   /etc/services

Add the following line to this file:

   mgsql_xinetd          7041/tcp                        # MGSQL

Finally restart the xinetd service:

   /etc/init.d/xinetd restart

Access to mgsql using REST

Now that the network service has been configured and deployed it is possible to execute SQL scripts via REST calls. Results are returned formatted as JSON.

For example, using the curl utility from the UNIX command line:

   curl -d "select * from patient" -H "Content-Type: text/sql" http://localhost:7041/mg.sql/execute

Assuming that the simple test database described previously has been created the above request will generate the following output:

   {"sqlcode": 0, "sqlstate": "00000", "error": "", "result": [{"num": "1","name": "Chris Munt","address": "Banstead"},{"num": "2","name": "Rob Tweed","address": "Redhill"},{"num": "3","name": "John Smith","address": "London"},{"num": "4","name": "Jane Doe","address": "Oxford"}]}

Simple invocation from a browser (Hint: Firefox does a good job of rendering JSON):

   http://127.0.0.1:7041/mgsql/mg.sql?sql=select * from patient

Alternatively, enter an SQL statement in the form generated by:

   http://127.0.0.1:7041/mgsql/mg.sql

In a live environment a production-grade web server should be used. For example, using the Apache server the mod_proxy module can be used to front the mgsql service.

Access to mgsql using ODBC

The ODBC driver is in the /odbc directory. Pre-built drivers for 32 and 64-bit Windows are in the /odbc/x86 and /odbc/x64 directories respectively. To install both drivers copy the contents of /odbc/x86 to:

   C:\Program Files (x86)\mgsql\

And copy the contents of /odbc/x64 to:

   C:\Program Files\mgsql\

You will have to create the /mgsql sub-directory if it doesn't already exist. To register both drivers, using Windows Explorer, double click on each of the following Registry files:

   C:\Program Files (x86)\mgsql\mgodbc32.reg
   C:\Program Files\mgsql\mgodbc64.reg

You can now configure an ODBC Data Source using the Windows Administrative tools for ODBC Data sources (accessed via the Windows Control Panel:

   Control Panel\System and Security\Administrative Tools\ODBC Data Sources

Under the System DSN tab. select Add... and choose one of the mgsql drivers as appropriate:

   MGSQL ODBC x86
   MGSQL ODBC x64

Complete the mgodbc configuration dialogue box and save:

  • Name: Your Data Source Name (DSN).
  • Description: An optional description.
  • Server: IP Address of your M server.
  • TCP Port: TCP Port (the default is 7041).
  • Directory or UCI: M UCI (leave blank for YottaDB).
  • Event Log File: Log file (including full path).
  • Event Log Level: Log level (a comma separated list of log directives).

Log Level Directives:

  • e: Log Errors.
  • ft: Log ODBC function call trace.
  • nt: Log all network buffers sent and received.

The data source created can now be used in Windows applications.

Transaction Processing

mgsql supports the standard SQL Transaction Processing commands:

   START TRANSACTION
   COMMIT
   ROLLBACK

The implementation of these commands is based on the underlying M commands: tstart, tcommit and trollback. SQL Transactions can be implemented in M code or via an external program connecting to mgsql via the ODBC driver.

The M Transaction Processing commands can be used directly in M code. For example:

   new %zi,%zo
   tstart
   set status=$$exec^%mgsql("",<sql update statement>,.%zi,.%zo)
   ; further update statements …
   tcommit

Alternatively, the corresponding SQL Transaction Processing commands can be used in SQL statements. For example:

   new %zi,%zo
   set status=$$exec^%mgsql("","transaction start",.%zi,.%zo)
   set status=$$exec^%mgsql("",<sql update statement>,.%zi,.%zo)
   ; further update statements …
   set status=$$exec^%mgsql("","commit",.%zi,.%zo)

It is also possible to implement transactions in M callback functions. This method is mandatory for YottaDB.

The transaction callback function must be defined as follows:

   callback(%zi, %zo)

Where:

  • %zi is the query input array.
  • %zo is the query output array.

For example:

   tp   ; transaction implemented in a callback function
        new %zi,%zo
        set %zi(0,"callback")="tpcb"
        set status=$$exec^%mgsql("","transaction start",.%zi,.%zo)
        quit
        ;
   tpcb(%zi,%zo) ; callback function
        set status=$$exec^%mgsql("",<sql update statement>,.%zi,.%zo)
        ; further update statements …
        set status=$$exec^%mgsql("","commit",.%zi,.%zo)
        quit status

SQL Query Optimisation

mgsql will attempt to find the most optimal route through the set of tables queried in accordance with the information that the optimiser can extract from the WHERE predicate together with any JOIN constraints specified.

In cases where mgsql does not come up with an optimal route through the tables it is possible to provide hints in the form of explicitly defining the indices to use. If an index is defined in the FROM statement, mgsql will attempt to process the tables in the order specified in the FROM statement.

The general form for explicitly optimising queries is as follows:

   select [columns] from table1:index1, table2 where [predicate]

Or if aliases are used:

   select [columns] from table1 a:index1, table2 b where [predicate]

In the above example, mgsql will attempt to parse table1 first using index1, followed by table2.

As a convenience, the primary key may be defined as index ‘0’:

   select [columns] from table1 a:0, table2 b:idx2 where [predicate]

In the above example, mgsql will attempt to parse table1 first using the primary key, followed by table2 using index idx2.

Resources used by mgsql

mgsql will write to the following globals

  • ^mgsqld: The catalogue or schema.

  • ^mgsqls: The spool file for SQL output.

  • ^mgsqlx: The cache of compiled queries.

  • ^mglog: The event Log.

  • ^mgtmp: A temporary file used by the SQL compiler.

  • ^mgtemp: A temporary sort file used when executing SQL queries.

  • mgsql will generate M Routines prefixed by 'x'.

License

Copyright (c) 2018-2023 MGateway Ltd, Surrey UK.
All rights reserved.

http://www.mgateway.com
Email: [email protected]

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

  http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

Release Notes

v1.0.7 (13 June 2019)

  • Initial Release

v1.0.8 (1 November 2019)

  • Greater flexibility in mapping mgsql tables to existing M global structures. The facility for specifying trailing M global subscripts is extended (see the 'separate' keyword in table creation).
  • A fault in the processing of outer join queries qualified with an 'on' clause has been corrected.

v1.1.9 (15 January 2020)

  • Introduce support for alternative date separators in the decode date function (ddate^%mgsqls). For example, although the separator is '/' in the US locale, it is is "." in the Czech locale.
  • Introduce support for Derived Fields. A Derived Field is defined in the schema as an M extrinsic function and can take any number of values from the same row as input parameters. For example, the calculation of a person's age from the stored 'date of birth' field can be implemented as a Derived Field - the value of 'age' being dependant on the time of data retrieval.
  • The embedded functions lower and upper have been implemented (Convert string to lower or upper-case, respectively.
  • A fault that led to some SQL queries of the form 'update ... set ... where' crashing has been corrected.
  • A fault that led to some secondary index names not being recognised has been corrected.
  • A fault that led to an error status (-1) being returned for some SQL DDL scripts even though the script completed successfully has been corrected.

v1.2.10 (14 April 2020)

  • Introduce support for the SQL TOP clause.
    • select top columns..., from table... (return just the first row of data).
    • select top 3 columns..., from table... (return just the first 3 rows of data).
  • The embedded functions trim, rtrim and ltrim have been implemented.
    • trim((string) input[, (string) characters]) (remove listed characters from the beginning and end of the supplied string).
    • rtrim((string) input[, (string) characters]) (remove listed characters from the end of the supplied string).
    • ltrim((string) input[, (string) characters]) (remove listed characters from the beginning of the supplied). string.
    • In all cases the default value for characters (if this argument is not supplied) is white space - ' '.
  • Correct a fault in the compilation of the SQL LIKE operator.
  • Finally, run the upgrade procedure. This will clean-up any temporary files previously used by mgsql and force the recompilation of all queries.
    • set status=$$upgrade^%mgsql(0)

v1.2.11 (28 May 2020)

  • Improve the parsing of expressions embedded in SQL scripts - notably the SQL WHERE predicate.
  • Remove the need to fully qualify column names with table names (or table name aliases) in SQL scripts except in cases where ambiguities would otherwise occur.

v1.2.12 (3 January 2021)

  • Correct a number of faults in the compilation of 'Select distinct ...' queries.

v1.2.13 (5 January 2021)

  • Correct a fault that led to duplicate values being erroneously returned for 'Select distinct ...' queries.
  • For this update, run the upgrade procedure after installation. This will clean-up any temporary files previously used by mgsql and force the recompilation of all queries.
    • set status=$$upgrade^%mgsql(0)

v1.2.14 (8 January 2021)

  • Introduce further options for embedding SQL queries in M code. Including:
    • The use of the spool file.
    • The use of callback functions.
  • Update the documentation.
  • Miscellaneous bug fixes.

v1.2.15 (10 January 2021)

  • Correct a regression in the processing of SQL queries containing 'in' or 'or' in the WHERE predicate.

v1.2.16 (13 January 2021)

  • Correct a fault in the compilation of INSERT queries for tables having secondary indices defined.

v1.2.17 (14 January 2021)

  • Correct a fault in the compilation of SELECT DISTINCT ... queries.

v1.2.18 (22 January 2021)

  • Correct a fault in the compilation of SELECT COUNT(DISTINCT ... queries.

v1.3.19 (22 February 2021)

  • Introduce SQL commands to support M Transaction Processing.
    • SQL Commands: START TRANSACTION, COMMIT and ROLLBACK
  • Introduce a native concurrent TCP server for YottaDB.
    • The Superserver can be started from the M command prompt using d start^%mgsql().
    • Invocation of Superserver child processes from the xinetd daemon is still supported.

v1.3.20 (25 June 2021)

  • Correct a fault that led to queries being recompiled on every call instead of reusing the code previously generated.
    • To force the recompilation of a particular query, set the recompile property - for example:
    • set %zi(0,"recompile")=1,status=$$exec^%mgsql("",[query],.%zi,.%zo)

v1.3.21 (30 September 2021)

  • Correct a fault that led to some queries containing an OUTER JOIN crashing with an 'undefined variable' error. Particularly affected were OUTER JOINs qualified with an ON condition.

v1.4.22 (3 February 2022)

  • Correct a fault that led to CREATE TABLE statements crashing if a SQL reserved word was used as a column name.
  • Correct a regression in the processing of the following operators in WHERE statements: OR, >, <.
  • Improved query optimisation together with a means through which queries can be explicitly optimised by specifying hints in the query text.
    • See the section on SQL Query Optimisation.
    • Run the upgrade procedure to force the recompilation (and re-optimisation) of all queries: set status=$$upgrade^%mgsql(0)

v1.4.23 (8 June 2023)

  • Documentation update,

About

An SQL engine for YottaDB and other M-like databases.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published