Singer tap that extracts data from a mssql database and produces JSON-formatted data following the Singer spec.
This is a PipelineWise compatible tap connector.
The recommended method of running this tap is to use it from PipelineWise. When running it from PipelineWise you don't need to configure this tap with JSON files and most of things are automated. Please check the related documentation at Tap mssql
If you want to run this Singer Tap independently please read further.
This section dives into basic usage of tap-mssql
by walking through extracting
data from a table. It assumes that you can connect to and read from a mssql
database.
First, make sure Python 3 is installed on your system or follow these installation instructions for Mac or Ubuntu.
It's recommended to use a virtualenv:
python3 -m venv venv
pip install tap-mssql
or
python3 -m venv venv
. venv/bin/activate
pip install --upgrade pip
pip install tap-mssql
In the event you encounter install issues on OS X stating fatal error: 'sqlfront.h' file not found
, install FreeTDS with SQL Server compatibility:
- Download the latest version from FreeTDS
- Extract and run
./configure --enable-msdblib
- Run
make && make install
There's some important business data siloed in this mssql database -- we need to extract it. Here's the table we'd like to sync:
mssql> select * from example_db.animals;
+----|----------|----------------------+
| id | name | likes_getting_petted |
+----|----------|----------------------+
| 1 | aardvark | 0 |
| 2 | bear | 0 |
| 3 | cow | 1 |
+----|----------|----------------------+
3 rows in set (0.00 sec)
Create a config file containing the database connection credentials, e.g.:
{
"host": "localhost",
"database": "db",
"port": "3306",
"user": "root",
"password": "password"
}
Recommended optional settings
"use_date_datatype": true
- This will emit true timestamps and handle the time datatype."use_singer_decimal": true
- This will help avoid numeric rounding issues emitting as a string with a format of singer.decimal."cursor_array_size": 10000
- This will help speed up extracts over a WAN or low latency network. The default is 1.
Windows Authentication is available! Don't provide a user or password and pymssql will use the user that is running the process on windows to login. e.g.:
{
"host": "localhost",
"database": "db"
}
Optional:
To filter the discovery to a particular schema within a database. This is useful if you have a large number of schemas and wish to speed up the discovery.
{
"filter_dbs": "your database schema name"
}
Optional:
To emit a date as a date without a time component or time without an UTC offset. This is helpful to avoid time conversions or to just work with a date datetype in the target database. If this boolean config item is not set, the default behaviour is false
i.e. emit date datatypes as a datetime. It is recommended to set this on if you have time datetypes and are having issues uploading into into a target database.
{
"use_date_datatype": true
}
Optional:
Set the version of TDS to use when communicating with MS SQL Server (the default is 7.3). This is used by pymssql with connecting and fetching data from SQL Server databases. See the pymssql documentation and FreeTDS documentation for more details.
{
"tds_version": "7.3"
}
Optional:
The characterset for the database / source system. The default is utf8
, however older databases might use a charactersets like cp1252 for the encoding. If you have errors with a UnicodeDecodeError: 'utf-8' codec can't decode byte ....
then a solution is examine the characterset of the source database / system and make an appropriate substitution for utf8 like cp1252.
{
"characterset": "utf8"
}
Optional:
The "conn_properties"
allows specific tweaking of database settings via SQL set statements to send to the database instance upon connection establishment. Can be a string or another kind of iterable of strings.
The default values set if this settings is not defined are:
"SET ARITHABORT ON; SET CONCAT_NULL_YIELDS_NULL ON; SET ANSI_NULLS ON; SET ANSI_NULL_DFLT_ON ON; SET ANSI_PADDING ON; SET ANSI_WARNINGS ON; SET ANSI_NULL_DFLT_ON ON; SET CURSOR_CLOSE_ON_COMMIT ON; SET QUOTED_IDENTIFIER ON; SET TEXTSIZE 2147483647;"
Example: override the built-in session properties supplied by pymssql by default, because one of the default settings (CURSOR_CLOSE_ON_COMMIT) is not available on PDW
{
"conn_properties": "SET ARITHABORT ON; SET CONCAT_NULL_YIELDS_NULL ON; SET ANSI_NULLS ON; SET ANSI_NULL_DFLT_ON ON; SET ANSI_PADDING ON; SET ANSI_WARNINGS ON; SET ANSI_NULL_DFLT_ON ON; SET QUOTED_IDENTIFIER ON; SET TEXTSIZE 2147483647;"
}
Optional:
The "enable_tds_logging"
When set it will dump out the underlying TDS driver logs. Useful for diagnosing issues if you are having connection issues to SQL Server databases. WARNING! this
does dump a lot of information and may log secure data, should be only used in Development
environments.
{
"enable_tds_logging": true
}
These are the same basic configuration properties used by the mssql command-line
client (mssql
).
Optional:
To emit all numeric values as strings and treat floats as string data types for the target, set use_singer_decimal to true. The resulting SCHEMA message will contain an attribute in additionalProperties containing the scale and precision of the discovered property:
"property": {
"inclusion": "available",
"format": "singer.decimal",
"type": [
"null",
"number"
],
"additionalProperties": {
"scale_precision": "(12,0)"
}
Usage:
{
"use_singer_decimal": true
}
Optional:
A numeric setting adjusting the internal buffersize. The common query tuning scenario is for SELECT statements that return a large number of rows over a slow network. Increasing arraysize can improve performance by reducing the number of round-trips to the database. However increasing this value increases the amount of memory required.
{
"cursor_array_size": 10000,
}
The tap can be invoked in discovery mode to find the available tables and columns in the database:
$ tap-mssql --config config.json --discover
A discovered catalog is output, with a JSON-schema description of each table. A source table directly corresponds to a Singer stream.
{
"streams": [
{
"tap_stream_id": "example_db-animals",
"table_name": "animals",
"schema": {
"type": "object",
"properties": {
"name": {
"inclusion": "available",
"type": ["null", "string"],
"maxLength": 255
},
"id": {
"inclusion": "automatic",
"minimum": -2147483648,
"maximum": 2147483647,
"type": ["null", "integer"]
},
"likes_getting_petted": {
"inclusion": "available",
"type": ["null", "boolean"]
}
}
},
"metadata": [
{
"breadcrumb": [],
"metadata": {
"row-count": 3,
"table-key-properties": ["id"],
"database-name": "example_db",
"selected-by-default": false,
"is-view": false
}
},
{
"breadcrumb": ["properties", "id"],
"metadata": {
"sql-datatype": "int(11)",
"selected-by-default": true
}
},
{
"breadcrumb": ["properties", "name"],
"metadata": {
"sql-datatype": "varchar(255)",
"selected-by-default": true
}
},
{
"breadcrumb": ["properties", "likes_getting_petted"],
"metadata": {
"sql-datatype": "tinyint(1)",
"selected-by-default": true
}
}
],
"stream": "animals"
}
]
}
In sync mode, tap-mssql
consumes the catalog and looks for tables and fields
have been marked as selected in their associated metadata entries.
Redirect output from the tap's discovery mode to a file so that it can be modified:
$ tap-mssql -c config.json --discover > properties.json
Then edit properties.json
to make selections. In this example we want the
animals
table. The stream's metadata entry (associated with "breadcrumb": []
)
gets a top-level selected
flag, as does its columns' metadata entries. Additionally,
we will mark the animals
table to replicate using a FULL_TABLE
strategy. For more,
information, see Replication methods and state file.
[
{
"breadcrumb": [],
"metadata": {
"row-count": 3,
"table-key-properties": ["id"],
"database-name": "example_db",
"selected-by-default": false,
"is-view": false,
"selected": true,
"replication-method": "FULL_TABLE"
}
},
{
"breadcrumb": ["properties", "id"],
"metadata": {
"sql-datatype": "int(11)",
"selected-by-default": true,
"selected": true
}
},
{
"breadcrumb": ["properties", "name"],
"metadata": {
"sql-datatype": "varchar(255)",
"selected-by-default": true,
"selected": true
}
},
{
"breadcrumb": ["properties", "likes_getting_petted"],
"metadata": {
"sql-datatype": "tinyint(1)",
"selected-by-default": true,
"selected": true
}
}
]
With a properties catalog that describes field and table selections, the tap can be invoked in sync mode:
$ tap-mssql -c config.json --properties properties.json
Messages are written to standard output following the Singer specification. The resultant stream of JSON data can be consumed by a Singer target.
{"value": {"currently_syncing": "example_db-animals"}, "type": "STATE"}
{"key_properties": ["id"], "stream": "animals", "schema": {"properties": {"name": {"inclusion": "available", "maxLength": 255, "type": ["null", "string"]}, "likes_getting_petted": {"inclusion": "available", "type": ["null", "boolean"]}, "id": {"inclusion": "automatic", "minimum": -2147483648, "type": ["null", "integer"], "maximum": 2147483647}}, "type": "object"}, "type": "SCHEMA"}
{"stream": "animals", "version": 1509133344771, "type": "ACTIVATE_VERSION"}
{"record": {"name": "aardvark", "likes_getting_petted": false, "id": 1}, "stream": "animals", "version": 1509133344771, "type": "RECORD"}
{"record": {"name": "bear", "likes_getting_petted": false, "id": 2}, "stream": "animals", "version": 1509133344771, "type": "RECORD"}
{"record": {"name": "cow", "likes_getting_petted": true, "id": 3}, "stream": "animals", "version": 1509133344771, "type": "RECORD"}
{"stream": "animals", "version": 1509133344771, "type": "ACTIVATE_VERSION"}
{"value": {"currently_syncing": "example_db-animals", "bookmarks": {"example_db-animals": {"initial_full_table_complete": true}}}, "type": "STATE"}
{"value": {"currently_syncing": null, "bookmarks": {"example_db-animals": {"initial_full_table_complete": true}}}, "type": "STATE"}
In the above example, we invoked tap-mssql
without providing a state file
and without specifying a replication method. The three ways to replicate a given
table are FULL_TABLE
, LOG_BASED
, and INCREMENTAL
.
Full-table replication extracts all data from the source table each time the tap is invoked.
Log_Based replication extracts change data from the MS SQL Server Change Data Capture (CDC) tables you have enrolled.
This method allows you to replicate just the changes to a table e.g. the Inserts, Deletes, and Updates. For this method to work you must enrol the database in question and tables that you wish to replicate.
See : https://docs.microsoft.com/en-us/sql/relational-databases/track-changes/enable-and-disable-change-data-capture-sql-server for more details.
Please Note: CDC is different to Change Tracking which is a older approach for tracking change. Log Based only works with CDC, it does not work with Change Tracking!
To find out more about setting up CDC, refer to this page MSSQL CDC Setup
Incremental replication works in conjunction with a state file to only extract new records each time the tap is invoked. This requires a replication key to be specified in the table's metadata as well.
Let's sync the animals
table again, but this time using incremental
replication. The replication method and replication key are set in the
table's metadata entry in properties file:
{
"streams": [
{
"tap_stream_id": "example_db-animals",
"table_name": "animals",
"schema": { ... },
"metadata": [
{
"breadcrumb": [],
"metadata": {
"row-count": 3,
"table-key-properties": [
"id"
],
"database-name": "example_db",
"selected-by-default": false,
"is-view": false,
"replication-method": "INCREMENTAL",
"replication-key": "id"
}
},
...
],
"stream": "animals"
}
]
}
We have no meaningful state so far, so just invoke the tap in sync mode again without a state file:
$ tap-mssql -c config.json --properties properties.json
The output messages look very similar to when the table was replicated using the
default FULL_TABLE
replication method. One important difference is that the
STATE
messages now contain a replication_key_value
-- a bookmark or
high-water mark -- for data that was extracted:
{"type": "STATE", "value": {"currently_syncing": "example_db-animals"}}
{"stream": "animals", "type": "SCHEMA", "schema": {"type": "object", "properties": {"id": {"type": ["null", "integer"], "minimum": -2147483648, "maximum": 2147483647, "inclusion": "automatic"}, "name": {"type": ["null", "string"], "inclusion": "available", "maxLength": 255}, "likes_getting_petted": {"type": ["null", "boolean"], "inclusion": "available"}}}, "key_properties": ["id"]}
{"stream": "animals", "type": "ACTIVATE_VERSION", "version": 1509135204169}
{"stream": "animals", "type": "RECORD", "version": 1509135204169, "record": {"id": 1, "name": "aardvark", "likes_getting_petted": false}}
{"stream": "animals", "type": "RECORD", "version": 1509135204169, "record": {"id": 2, "name": "bear", "likes_getting_petted": false}}
{"stream": "animals", "type": "RECORD", "version": 1509135204169, "record": {"id": 3, "name": "cow", "likes_getting_petted": true}}
{"type": "STATE", "value": {"bookmarks": {"example_db-animals": {"version": 1509135204169, "replication_key_value": 3, "replication_key": "id"}}, "currently_syncing": "example_db-animals"}}
{"type": "STATE", "value": {"bookmarks": {"example_db-animals": {"version": 1509135204169, "replication_key_value": 3, "replication_key": "id"}}, "currently_syncing": null}}
Note that the final STATE
message has a replication_key_value
of 3
,
reflecting that the extraction ended on a record that had an id
of 3
.
Subsequent invocations of the tap will pick up from this bookmark.
Normally, the target will echo the last STATE
after it's finished processing
data. For this example, let's manually write a state.json
file using the
STATE
message:
{
"bookmarks": {
"example_db-animals": {
"version": 1509135204169,
"replication_key_value": 3,
"replication_key": "id"
}
},
"currently_syncing": null
}
Let's add some more animals to our farm:
mssql> insert into animals (name, likes_getting_petted) values ('dog', true), ('elephant', true), ('frog', false);
$ tap-mssql -c config.json --properties properties.json --state state.json
This invocation extracts any data since (and including) the
replication_key_value
:
{"type": "STATE", "value": {"bookmarks": {"example_db-animals": {"replication_key": "id", "version": 1509135204169, "replication_key_value": 3}}, "currently_syncing": "example_db-animals"}}
{"key_properties": ["id"], "schema": {"properties": {"name": {"maxLength": 255, "inclusion": "available", "type": ["null", "string"]}, "id": {"maximum": 2147483647, "minimum": -2147483648, "inclusion": "automatic", "type": ["null", "integer"]}, "likes_getting_petted": {"inclusion": "available", "type": ["null", "boolean"]}}, "type": "object"}, "type": "SCHEMA", "stream": "animals"}
{"type": "ACTIVATE_VERSION", "version": 1509135204169, "stream": "animals"}
{"record": {"name": "cow", "id": 3, "likes_getting_petted": true}, "type": "RECORD", "version": 1509135204169, "stream": "animals"}
{"record": {"name": "dog", "id": 4, "likes_getting_petted": true}, "type": "RECORD", "version": 1509135204169, "stream": "animals"}
{"record": {"name": "elephant", "id": 5, "likes_getting_petted": true}, "type": "RECORD", "version": 1509135204169, "stream": "animals"}
{"record": {"name": "frog", "id": 6, "likes_getting_petted": false}, "type": "RECORD", "version": 1509135204169, "stream": "animals"}
{"type": "STATE", "value": {"bookmarks": {"example_db-animals": {"replication_key": "id", "version": 1509135204169, "replication_key_value": 6}}, "currently_syncing": "example_db-animals"}}
{"type": "STATE", "value": {"bookmarks": {"example_db-animals": {"replication_key": "id", "version": 1509135204169, "replication_key_value": 6}}, "currently_syncing": null}}
Based on Stitch documentation
The json output by this tap should be JSON schema compliant and also conform to the Singer Framework standard.
Noted: Below is some examples of the data transformation to present data in a standard manner for Singer Targets.
A special note for datetime2
and datetimeoffsets
MS SQL Server datatypes.
For datetime2
, it as a naïve datetime timestamp with no concept of an offset, datetime2
supports nanoseconds with up to seven decimal places. With datetimeoffset
datatypes, the data is normalized to UTC i.e. taking the offset as it is in MSSQL Server and converting it to a UTC so it is agnostic to the target platform - anything with a datetimeoffset can be cast into local time in the target.
To work-around issues with the PYMSSQL driver, the conversion takes place on the SQL Server source to ensure there is no loss of precision leading to errors as a result of rounding.
SQL DataType | Example Value | Min | Max | JSON Format | JSON Type | Example Result |
---|---|---|---|---|---|---|
tinyint | 254 | 0 | 255 | integer | 254 | |
smallint | -32768 | -32768 | 32767 | integer | -32768 | |
int | -2147483648 | -2147483648 | 2147483647 | integer | -2147483648 | |
bigint | -9223372036854775808 | -9223372036854775808 | 9223372036854775807 | integer | -9223372036854775808 | |
datetime2 | 9999-12-31 23:59:59.9999999 | date-time | string | 9999-12-31 23:59:59.9999999 | ||
datetimeoffset | 2023-08-22 08:24:32.1277000 +10:00 | date-time | string | 2023-08-21T22:24:32.1277000Z |
This section dives into basic commands to build tap-mssql
if an alteration is made to the code.
Run the following command each time you need to rebuild the tap.
python3 -m venv venv
. venv/bin/activate
pip install --upgrade pip
pip install .
To run the init.py python program in debug mode, you need to do the following two steps. Note: This was run within a Docker Container in Visual Studio Code.
- Create a .vscode/launch.json file. Note: The parameters config.json and properties.json should point to the files you have generated in previous steps above. If you want to test state, include the state parameter as shown below and prepare an appropriate state file as per the instructions in an earlier section.
{
"version": "0.2.0",
"configurations": [
{
"name": "Python: Program",
"type": "python",
"request": "launch",
"program": "${workspaceRoot}/tap_mssql/__init__.py",
"args": [
"-c", "config.json",
"--properties", "properties.json"
"--state", "state.json"
]
}
]
}
- Add a main entry to the init.py file to run interactively
Add the following lines to the end of the init.py in the tap_mssql directory.
if __name__ == '__main__':
main() # pylint: disable=no-value-for-parameter