PostgreSQL

Description

Change Data Capture (CDC) enables you to ingest data changes from tables in a PostgreSQL Data Store and publish them to a sink. In a CDC pipeline, data changes in the source relational table are captured at the row-level operations (INSERT, DELETE, UPDATE) based on the table's PRIMARY KEY. Then they're pushed to the downstream sink in real-time.

How a CDC Pipeline Works

Source Record Structure

DeltaStream uses Debezium to capture changes in the source relational table. This means when you define a CDC source backed by a given table in a POSTGRESQL store, you can use this JSON skeleton:

{
 "op": ...,
 "ts_ms": ...,
 "before": {...},
 "after": {...},
 "system": {...}
}

• op: Describes the row-level operation that caused a change in the source. It can be c for CREATE, u for UPDATE, d for DELETE or r for READ.

• ts_ms: Shows the time at which the change event is processed by the CDC pipeline, in the source.

• before and after: These specify the state of the row at which the change occurred, before and/or after the change, depending on the semantics of the operation.

• system: Shows the metadata in the source about the operation. Some of its included fields are:

  • db (String): The relational database name containing the source table.

  • schema (String): The relational schema name containing the source table.

  • table (String): The relational source table name.

  • lsn (BigInt): The log sequence number of the change.

Requirements for a PostgreSQL Source Store

Debezium uses PostgreSQL's logical decoding slot for streaming changes from the given relational source table.

• When you create the CDC pipeline, you need to specify a name for the replication slot. If a slot already exists with that name, the system uses the CDC job. If there's no pre-existing slot, the system creates a new replication slot and uses the slot name for the job. See below for more details.

• When you create the source PostgreSQL Data Store:

  • the username you provide must have enough privileges to use an existing replication slot or create a new one.

  • the source PostgreSQL store must use logical decoding with the write-ahead log (WAL). This means you should set wal_level to logical in the source relational database. (For more information, see PostgreSQL ALTER SYSTEM docs.)

  • PostgreSQL needs to have a publication created for streaming changes. This publication is created at start-up, if it does not already exist, and it includes all tables. For that, the Postgres user must have superuser permissions to create such a publication. Another option is to create a publication, named dbz_publication , for all tables in PostgreSQL before starting the CDC query so the connector uses that.

Working with PostgreSQL replication slots

Each DeltaStream query that reads rows from a PostgreSQL source for a CDC pipeline must specify a replication slot. Replication slots in PostgreSQL represent a stream of changes that can be replayed to a client in the order in which the changes were made on the original server. Each slot has the following characteristics:

• uniquely named across all databases in a PostgreSQL instance

• persists independently from the clients using it

• crash-proof

• contains its own independent state

You cannot assign a single replication slot in DeltaStream to multiple simultaneously-running queries. By default, PostgreSQL instances limit the number of replication slots to 10.

To list, create, or drop replication slots In the PostgreSQL CLI psql , use the following queries:

// Check list of existing replication slots
SELECT * FROM pg_replication_slots;

// Create a new logical replication slot
SELECT pg_create_logical_replication_slot('ds_cdc', 'pgoutput');

// Delete an existing replication slot
// Note: For logical slots, this must be called while connected to the same database the slot was created on.
SELECT pg_drop_replication_slot('ds_cdc');

Note For more information on replication slots, please see the following sections of the PostgreSQL documentation:

• Replication

• Replication Slots

• Logical Decoding and Replication Slots

Considerations

Storage TOAST

PostgreSQL uses a fixed page size (commonly 8 kb), so rows with large field values are compressed and/or broken up into multiple physical rows. This is the TOAST technique. See PostgreSQL documentation on the subject to learn more.

DeltaStream follows the Debezium default behavior when generating CDC events containing TOASTed fields. If the TOASTed field is unchanged, DeltaStream replaces the value in the CDC event with __debezium_unavailable_value.

How to Create a CDC Pipeline

You create a CDC pipeline in 2 steps:

1. Define the CDC source

2. Define the sink and the query to write CDC changes into it

Step 1. Define DDL for the CDC source

Using CREATE STREAM, define a Stream backed by the source relational table. The parameters below are used as CDC parameters in the WITH clause in CREATE STREAM:

Parameter Name	Description
postgresql.db.name	Name of the database in the POSTGRESQL store containing the source table. Required: No Type: String Valid values: See LIST ENTITIES
postgresql.schema.name	Name of the schema in the POSTGRESQL store containing the source table. Required: No Type: String Valid values: See LIST ENTITIES
postgresql.table.name	Name of the source table in the POSTGRESQL store. Required: No Type: String Valid values: See LIST ENTITIES
postgresql.cdc.table.list	Comma-separated list of fully qualified source tables to capture using CDC. Each entry must be specified as schema_name.table_name. If a table name contains special characters (such as a dot) or uppercase letters, enclose it in double quotes — for example, public."orders.v1". Regex patterns are supported to match multiple tables dynamically. (For example, public.orders_.* matches all tables in the public schema with names starting with orders_). Required: No Type: String Valid values: See LIST ENTITIES
value.format	Format of the CDC record coming from the source. Required: Yes Type: String Valid values: JSON
store	Name of the POSTGRESQL store that hosts the relational table backing the CDC source. Required: Yes Type: String Valid values: See LIST STORES

Step 2. Define the CREATE STREAM AS SELECT (CSAS) for CDC Sink Stream and Query

Using CREATE STREAM AS SELECT, define a Stream backed by the sink Streaming Entity and the query to insert CDC records into it.

Sink parameters

The CDC sink can write into an existing entity in the sink store or create a new entity. The below parameters are used in the WITH clause for the sink in CSAS to create the desired behavior:

Parameter Name	Description
topic	Name of the Streaming Entity into which the data for the CDC sink is written. If the entity doesn't exist, the system creates an entity with this name in the corresponding store. Required: No Default value: Lowercase sink name Type: String
topic.partitions	The number of partitions to use when creating the sink entity, if applicable. If the entity already exists, this value must equal the number of partitions in the existing entity. Required: Yes, unless topic already exists Type: Integer Valid values: [1, ...]
topic.replicas	The number of replicas to use when creating the sink entity, if applicable. If the entity already exists, this value must equal the number of replicas in the existing entity. Required: Yes, unless topic already exists Type: Integer Valid values: [1, ...]

Source parameters

The source stream in the CDC pipeline captures the changes from a relational table in the source POSTGRESQL Data Store. Here are the parameters you can specify in the WITH clause for the source in CSAS:

Parameter Name	Description
postgresql.slot.name	Name of the PostgreSQL replication slot to use for the CDC pipeline. Note that the username you provide for the source PostgreSQL store should have the required privileges to use this slot. Required: Yes Type: String
postgresql.cdc.scan.newly.added.table	Enables automatic discovery of new tables when resuming a Postgres CDC pipeline. When set to true, any tables that match the configured tables list patterns, but were not part of the previous run, are automatically detected. When set to false (default), the pipeline only continues reading the tables that were part of the original state and ignores any newly created matching tables. This option has an effect only when resuming a CDC pipeline from a previous run's state and does not change the behavior of a fresh initial run. Required: No Default value: false Type: Boolean
scan.startup.option	Startup mode for Postgres CDC consumer when starting. Required: No Default value: initial Type: String Valid values: initial, latest-offset, committed-offset, snapshot
postgresql.decoding.plugin.name	Name of the Postgres logical decoding plug-in on the source store. Supported values are decoderbufs, pgoutput, wal2json, wal2json_rds, wal2json_streaming, and wal2json_rds_streaming. Required: No Default value: pgoutput Type: String

Passthrough Source parameters

Passthrough CDC properties

Passthrough CDC properties are supported for the CDC source stream. These properties must be specified using the cdc. prefix in their names and must be assigned a string value. They are listed in the table below. Refer to the CDC connector options for more details.

Parameter Name (with prefix)	Description
cdc.split.size	The number of rows of table snapshot. Captured tables are split into multiple splits when read the snapshot of table. Required: No Default value: '1000' Type: String
cdc.fetch.size	The maximum fetch size for per poll when read table snapshot. Required: No Default value: '1024' Type: String
cdc.scan.incremental.close-idle-reader.enabled	Whether to close idle readers at the end of the snapshot phase. Required: No Default value: 'false' Type: String
cdc.heartbeat.interval.ms	The interval of sending heartbeat event for tracing the latest available replication slot offsets. Required: No Default value:'30s' Type: String
cdc.chunk-key.even-distribution.factor.lower-bound	The lower bound of chunk key distribution factor. The distribution factor is used to determine whether the table is evenly distribution or not. The table chunks would use evenly calculation optimization when the data distribution is even, and the query for splitting would happen when it is uneven. The distribution factor could be calculated by (MAX(id) - MIN(id) + 1) / rowCount. Required: No Default value: '0.05d' Type: String
cdc.chunk-key.even-distribution.factor.upper-bound	The upper bound of chunk key distribution factor. The distribution factor is used to determine whether the table is evenly distribution or not. The table chunks would use evenly calculation optimization when the data distribution is even, and the query for splitting would happen when it is uneven. The distribution factor could be calculated by (MAX(id) - MIN(id) + 1) / rowCount. Required: No Default value: '1000.0d' Type: String
cdc.connect.pool.size	The connection pool size. Required: No Default value: '30' Type: String
cdc.connect.max-retries	The max retry times that the connector should retry to build database server connection. Required: No Default value: '3' Type: String
cdc.connect.timeout	The maximum time that the connector should wait after trying to connect to the PostgreSQL database server before timing out. Required: No Default value: '30s' Type: String

Passthrough Debezium properties

CDC source also supports passthrough Debezium properties. These properties must be specified using the cdc.debezium. prefix. For additional details on Debezium passthrough options, refer to Debezium's Postgres Connector properties

Supported Data Types in a Postgres CDC Pipeline

PostgreSQL type	DeltaStream SQL type
BOOLEAN	BOOLEAN
SMALLINT INT2 SMALLSERIAL	SMALLINT
INTEGER INT INT4 SERIAL	INTEGER
BIGINT INT8 BIGSERIAL	BIGINT
REAL FLOAT4	FLOAT
FLOAT8 DOUBLE PRECISION	DOUBLE
NUMERIC(p, s) DECIMAL(p, s)	DECIMAL(p,s)
DATE	DATE
TIME [(p)] [WITHOUT TIMEZONE]	TIME[(p)]
TIMESTAMP [(p)] [WITHOUT TIMEZONE]	TIMESTAMP [(p)]
CHAR(n) CHARACTER(n) VARCHAR(n) CHARACTER VARYING(n) TEXT	VARCHAR
BYTEA	BYTES

Attention : PostgreSQL data types that are not listed in the table above are not supported by the DeltaStream Flink CDC Connector.

Example

Create CDC pipeline capturing changes from a Postgres table

Assume a POSTGRESQL Data Store is defined with the name pgstore and it has a table named pageviews under the schema named public in the report database. Here is how the data displays in pageviews. Each row has 3 columns — viewtime, userid, and pageid — and shows when a given page was visited by a specific user:

demodb.public/pgstore# PRINT ENTITY "public".pageviews;
+-----------------+-------------+-------------+
| viewtime        | userid      | pageid      |
+=================+=============+=============+
| 1694124853651   | User_4      | Page_29     |
+-----------------+-------------+-------------+
| 1694124856731   | User_1      | Page_59     |
+-----------------+-------------+-------------+
| 1694124857732   | User_1      | Page_63     |
+-----------------+-------------+-------------+

Use the DDL statement below to create a Stream named pageviews_cdc, backed by the pageviews table to capture its CDC records:

CREATE STREAM pageviews_cdc(
  op VARCHAR,
  ts_ms BIGINT,
  `before` STRUCT<viewtime BIGINT, userid VARCHAR, pageid VARCHAR>, 
  `after`  STRUCT<viewtime BIGINT, userid VARCHAR, pageid VARCHAR>, 
  `source` STRUCT<db VARCHAR, `schema` VARCHAR, `table` VARCHAR, `lsn` BIGINT>)
WITH (
  'store'='pgstore', 
  'value.format'='json',
  'postgresql.db.name'='report',
  'postgresql.schema.name'='public',
  'postgresql.table.name'='pageviews');

The pageviews_cdc stream is used as the source in the CDC pipeline. Whenever an INSERT, DELETE, or UPDATE happens on pageviews in the relational database, a corresponding record is generated in pageviews_cdc to capture the change. Now use the below CSAS statement to define a stream as the sink and run a query to write those changes into it:

CREATE STREAM cdc_sink WITH (
  'topic.partitions' = 1,
  'topic.replicas' = 1) AS 
SELECT * FROM pageviews_cdc WITH ('postgresql.slot.name' = 'slot001');

Note that in the above CSAS you must provide the topic properties, as a new topic is created in the sink store. To use a pre-existing topic such as cdc_logs in the sink store, you can replace topic properties with the topic name in the WITH clause:

CREATE STREAM cdc_sink WITH ('topic' = 'cdc_logs') AS 
SELECT * FROM pageviews_cdc WITH ('postgresql.slot.name' = 'slot002');

As an example, assume a new record is added to pageviews showing that User_5 visited Page_94. Following this INSERT operation, you see a record similar to the below in cdc_sink published via the CDC pipeline defined above:

{
 "op":"c",
 "ts_ms":1693430399726,
 "before":null,
 "after":{"viewtime":1693430399292,"userid":"User_5","pageid":"Page_94"},
 "source":{"db":"report","schema":"public","table":"pageviews","lsn":38990016}
}

Now imagine you're only interested in the DELETE events in the source and wish to write the userid and lsn for each DELETE. You can use the query below to create such a CDC pipeline:

CREATE STREAM cdc_sink WITH ('topic' = 'cdc_logs') AS 
SELECT ts_ms AS event_ts,
       `before`->userid AS uid,
       `source`->`lsn` AS seq_num
FROM pageviews_cdc WITH ('postgresql.slot.name' = 'slot003')
WHERE op = 'd';

Create CDC pipeline capturing changes from multiple tables

Now consider a scenario similar to the previous example, except that besides the pageviews table there is another table called users under the staff schema within the report database in the pgstore store.

Suppose you want to create a CDC pipeline to capture changes for both tables using a single replication slot. First, define a stream as the CDC source:

CREATE STREAM cdc_source(
  op VARCHAR,
  ts_ms BIGINT,
  `before` BYTES,
  `after`  BYTES, 
  `source` STRUCT<`schema` VARCHAR, `table` VARCHAR>)
WITH (
  'store'='pgstore', 
  'value.format'='json',
  'postgresql.cdc.table.list'='public.pageviews,staff.users');

The cdc_source stream acts as the CDC source in your pipeline. Whenever an INSERT, DELETE, or UPDATE happens on either the pageviews or users table in the source database, a corresponding change event is captured and written into this stream.

Now, use the following CSAS statement to define a downstream stream as the sink and run a query to write these changes into it. Each output record includes the operation type and timestamp (op and ts_ms), along with the source schema and table name, which indicate where the changes came from.

You can use these fields to filter or route changes for each table in subsequent queries that read from cdc_sink.

CREATE STREAM cdc_sink WITH (
  'topic.partitions' = 1,
  'topic.replicas' = 1) AS 
SELECT
  `source`->`schema` AS src_schema,
  `source`->`table` AS src_table,
  `before`,
  `after`,
  `op`,
  `ts_ms`
FROM cdc_source WITH ('postgresql.slot.name' = 'slot004');

Run one query per table to write the captured changes into their own dedicated Kafka topic.

// Query 1
CREATE pageviews_raw_cdc WITH ('store'='kafka_store') AS
SELECT `before`,
    `after`,
    `op`,
    `ts_ms`
FROM cdc_raw_sink

WHERE src_schema = ‘public’ AND `src_table` = ‘pageviews’;

// Query 2
CREATE orders_raw_cdc WITH ('store'='kafka_store') AS
SELECT `before`,
    `after`,
    `op`,
    `ts_ms`
FROM cdc_raw_sink
WHERE src_schema = ‘public’ AND `src_table` = ‘orders’;

Now run a DDL per topic to define a table-specific CDC source for the table.

CREATE STREAM pageviews_cdc(
op VARCHAR,

ts_ms BIGINT,
`before` STRUCT<viewtime BIGINT, userid VARCHAR, pageid VARCHAR>,

`after` STRUCT<viewtime BIGINT, userid VARCHAR, pageid VARCHAR>,
) WITH (
    'store'='kafka_store',
    'value.format'='json',
    'topic'='pageviews_raw_cdc’);

CREATE STREAM orders_cdc(
    op VARCHAR,
    ts_ms BIGINT,
    `before` STRUCT<ts BIGINT, order_id VARCHAR, price INTEGER>,
    `after` STRUCT<ts BIGINT, order_id VARCHAR, price INTEGER>,

) WITH (
    'store'='kafka_store',
    'value.format'='json',
    'topic'='orders_raw_cdc’);

You could write further queries on the above source relations, again defined per table on dedicated topics, to process and write results into the sink of your choice, such as a Snowflake table.