Recce (rek-ee) is a database reconciliation tool for developers.
It is intended to make it easier to reconcile arbitrary datasets between source and target database schemas on an ongoing basis using configured SQL expressions without writing code.
You can read more about the ideas behind Recce at DESIGN.md or in the FAQ.
- 🔥 Trigger reconciliation of source and target database schemas using simple SQL expressions
- đź”’ Privacy - no sensitive data stored or output by default. By default, Recce compares hashes of data.
- ✏️ Flexible configuration to map groups of data to reconcile to different datasources and databases
- 🤝 Supports connectivity to MySQL, Postgres, MSSQL, MariaDB, AWS Aurora RDS MySQL, Oracle databases
- 💪🏾 Database agnostic when comparing database types - as long as it is coerced to a comparable JVM type consistently, it will be hashed identically
- ⏰ Schedule reconciliations for low-traffic periods against your data sources
- Recce Server
- Getting Started
- Configuration
- Development
- FAQ
Recce is currently published and available as a container image via GHCR.
However, it also
- requires its own Postgres database
- ...and to be practically useful you will want to configure it with connectivity to some data sources you wish to reconcile.
The below options require only JDK 11+ and Docker (or equivalent) installed locally.
-
Run Recce locally
- Either Build locally with an example source and target database (More info at DEVELOPMENT.md).
./batect -f examples/batect-petshop-mariadb.yml run
- Or use pre-validated Docker image locally, using this repository only for setting up a DB for Recce, and an example scenario.
# Run in one shell - starts a DB for Recce, and an example scenario ./batect -f examples/batect-petshop-mariadb.yml run-deps # Run in another shell - runs Recce docker run -p 8080:8080 \ -v $(pwd)/examples/scenario/petshop-mariadb:/examples/scenario/petshop-mariadb \ -e AUTH_USERNAME=admin -e AUTH_PASSWORD=admin \ -e MICRONAUT_CONFIG_FILES=/examples/scenario/petshop-mariadb/application-petshop-mariadb.yml \ -e DATABASE_HOST=host.docker.internal \ -e R2DBC_DATASOURCES_SOURCE_URL=r2dbc:pool:mariadb://host.docker.internal:8000/db \ -e R2DBC_DATASOURCES_TARGET_URL=r2dbc:pool:mariadb://host.docker.internal:8001/db \ ghcr.io/thoughtworks-sea/recce-server:latest
-
Explore and trigger runs via Recce's APIs, accessible via interactive UI at http://localhost:8080/rapidoc/. Some non-exhaustive examples are included below, but fuller documentation is available via the UI. Recce uses basic authentication to protect its API endpoints. In the example scenario, both username and password are set to "admin".
-
Synchronously trigger a run, waiting for it to complete via UI or
curl -X POST http://localhost:8080/runs -H 'Content-Type: application/json' -d '{ "datasetId": "categories" }' -u "admin:admin"
Expand example results
{ "completedDurationSeconds": 0.736356642, "completedTime": "2021-12-07T10:37:35.469576795Z", "createdTime": "2021-12-07T10:37:34.733220153Z", "datasetId": "categories", "id": 35, "summary": { "bothMatchedCount": 1, "bothMismatchedCount": 0, "source": { "meta": { "cols": [ { "javaType": "String", "name": "MigrationKey" }, { "javaType": "Long", "name": "count(distinct category)" } ] }, "onlyHereCount": 0, "totalCount": 1 }, "target": { "meta": { "cols": [ { "javaType": "String", "name": "MigrationKey" }, { "javaType": "Long", "name": "count(*)" } ] }, "onlyHereCount": 0, "totalCount": 1 }, "totalCount": 1 } }
-
Recce is configured by adding datasources and datasets that you wish to reconcile.
You can manage configuration as multiple files; or one single file. They will be merged together at runtime.
As a Micronaut application, much of Recce's configuration is open for hacking and can be expressed in multiple ways.
Recce-specific configuration (as opposed to all possible generic Micronaut configuration) is documented within the default config.
To make it easier to configure Recce's own DB, some dedicated properties are respected to construct the appropriate URL including connection pooling.
Env value | Default | Description |
---|---|---|
DATABASE_HOST | localhost |
Host your Postgres DB is on |
DATABASE_PORT | 9000 |
Port your Postgres DB is on |
DATABASE_NAME | db |
The name of the logical database within Postgres |
DATABASE_USERNAME | user |
Username to connect with |
DATABASE_PASSWORD | password |
Password to connect with |
In the conventional Micronaut way, additional settings can be configured using
R2DBC_DATASOURCES_DEFAULT_*
environment variables, orr2dbc.datasources.default.*
Java system properties or- Merging in an additional YAML configuration file (see below)
As a Micronaut application, configuration can be externalised in many ways.
However, the recommended way to add additional configuration for your own datasources and datasets to reconcile is to
mount a volume with your configuration and set MICRONAUT_CONFIG_FILES
to point Recce at your additional configuration
which will be merged into the base configuration. This allows you to separate and manage configuration as appropriate
for your environment.
mkdir -p my-dataset-configs
touch my-dataset-configs/config1.yml my-dataset-configs/config2.yml
## Run latest from docker
docker run -p 8080:8080 \
-v $(pwd)/my-dataset-configs:/config \
-e MICRONAUT_CONFIG_FILES=/config/config1.yml,/config/config2.yml \
ghcr.io/thoughtworks-sea/recce-server:latest
Recce supports only basic authentication for API endpoints out of the box. The credentials used in basic authentication
to protect the API endpoints can be configured in the auth
block of your config file.
auth:
username: some-username
password: some-password
or read from the environment:
export AUTH_USERNAME=some-username
export AUTH_PASSWORD=some-password
Arbitrary #s of data sources can be configured in the r2dbc.datasources
block of your config file.
r2dbc:
datasources:
my-source-db: # Name your datasource anything you want, other than "default"
# R2DBC URL for your database r2dbc:pool:DB_TYPE://DB_HOST:DB_PORT/DB_NAME
# DB_TYPE supported = mysql|postgresql|mssql|mariadb|oracle
url: r2dbc:pool:mysql://source-db:3306/db
username: user
password: password
my-target-db:
url: r2dbc:pool:mysql://target-db:3306/db
username: user
password: password
For configuration specific to a given driver/DB you can consult their documentation. Usually additional settings needs to be in the options:
block; or sometimes inside the connection URL.
- MySQL
- Postgres
- MS SQL Server
- MariaDB (can also be used for MySQL, as well as AWS RDS Aurora MySQL)
- Oracle
By default, Recce is deployed with r2dbc-pool to manage connection pooling to data sources. If you remove :pool
from the URL, this will be disabled.
You can thus customise the connection pool size, etc
r2dbc:
datasources:
my-source-db:
url: r2dbc:pool:mysql://source-db:3306/db
username: user
password: password
options: # Connection Pool options
initialSize: 1
maxSize: 5
maxLifeTime: 5
# etc, see https://github.com/r2dbc/r2dbc-pool
Datasets are the heart of your configuration. This tells Recce what and how to reconcile logically equivalent chunks of data.
Datasets are groupings of data which
- point to source and target datasources configured above
- express queries against source and target that will should produce identical output if the datasets are to be considered reconciled
- can use normal SQL to express an equivalent data representation between source and target schemas which accounts for intended differences in the way data has been structured and migrated
- can be scheduled to run on regular intervals
reconciliation:
datasets:
my-data-set: # Name your datasets however you would like
source:
# Reference to a datasource defined in `r2dbc.datasources`
datasourceRef: my-source-db
# Specify whether to load inline query or read from a query file
# If both `query` and `queryFile` are provided, inline query takes precedence
# If both query and queryFile are not specified,
# load the query statement from ${dataset-name}-${source-or-target}.sql in `queryFileBaseDir`
# Optional SQL query to evaluate against the source DB
query: >
SELECT id AS MigrationKey, * FROM my-table
# Optional path to file containing query to evaluate against the source DB
queryFile: examples/scenario/<{example-scenario}/queries/<{example-scenario}-my-data-set-source.yml
target:
# Reference to a datasource defined in `r2dbc.datasources`
datasourceRef: my-target-db
# Specify whether to load inline query or read from a query file
# If both `query` and `queryFile` are provided, inline query takes precedence
# If both query and queryFile are not specified,
# load the query statement from ${dataset-name}-${source-or-target}.sql in `queryFileBaseDir`
# Optional SQL query to evaluate against the target DB
query: >
SELECT id AS MigrationKey, * FROM my-table
# Optional path to file containing query to evaluate against the target DB
queryFile: examples/scenario/<{example-scenario}/queries/<{example-scenario}-my-data-set-target.yml
# Optional strategy for determining equivalence of column values
hashingStrategy: TypeLenient
# Optional scheduling of regular or one-of reconciliations
schedule:
# Must adhere to format https://docs.micronaut.io/latest/api/io/micronaut/scheduling/cron/CronExpression.html
# or https://crontab.guru/ (without seconds)
cronExpression: 0 0 * * *
Fuller example Recce-specific configuration is available here.
The general philosophy of Recce is that differences between source and target are best handled by the developer using regular SQL queries.
Recce needs to know which column represents a unique identifier for the row that should be consistent between source
and target
and implies these rows represent the same entity.
To do this, designate a column by naming it as MigrationKey
(case insensitive)
SELECT natural_id AS MigrationKey, some, other, columns
FROM my_table
Recce will complain if there is more than one column in your dataset with this name.
Currently Recce ignores names of columns other than the MigrationKey
column. That means that the order of columns is critical and must match between your two queries.
If the column in position 3 represents datum X
in the source
dataset, then the column in position 3 in the target
dataset should also represent the same datum.
The data types of the columns need not match exactly; however if the values produce different hashes, this will lead to a row mismatch. For example an integer of 111
in source will produce a different hash to a string of "111"
in target. If you want these values to be considered identical, you should use SQL to coerce the data types to match and express this intended difference.
If you would like to see a nameBased
column matching option, consider adding your thoughts to #55.
You should look to handle differences in types and semantics using the SQL expressions and functions available on the relevant database platform.
For example, if your source database represented an enumeration using integers (1, 2, 3
etc) whereas your target represented them using VARCHAR
s, you would use a CASE
statement to express this expected difference
-- source
SELECT id AS MigrationKey, CASE WHEN enumerated = 1 THEN 'Completed' ELSE 'PENDING' END
FROM my_table
-- target
SELECT id AS MigrationKey, enumerated_text
FROM my_table
When reconciling very large data sets, it may not be feasible to do row-by-row, column-by-column comparisons. In such cases it may be sufficient to use aggregate queries to get an idea whether you have lost data. e.g
--- check #s of login audit records for a month of data by location and type
SELECT user_location || user_type AS MigrationKey, count(*) AS Count
FROM user_login_audit
WHERE user_login_datetime >= timestamp('2021-11-01 00:00:00') AND
user_login_datetime < timestamp('2021-12-01 00:00:00')
GROUP BY user_location, user_type
See DEVELOPMENT.md to get started.
See FAQ.md.