Transferring data from a PostgreSQL source endpoint

Yandex Data Transfer enables you to migrate data from a PostgreSQL database and implement various data transfer, processing, and transformation scenarios. To implement a transfer:

1. Explore possible data transfer scenarios.
2. Prepare the PostgreSQL database for the transfer.
3. Set up a source endpoint in Yandex Data Transfer.
4. Set up one of the supported data targets.
5. Create a transfer and start it.
6. Perform the required operations with the database and see how the transfer is going.
7. In case of any issues, use ready-made solutions to resolve them.

Scenarios for transferring data from PostgreSQLScenarios for transferring data from PostgreSQL

1. Migration: Moving data from one storage to another. Migration often means migrating a database from obsolete local databases to managed cloud ones.

   • Migrating a PostgreSQL cluster.
   • Migrating from AWS RDS for PostgreSQL.
   • Migration with change of storage from PostgreSQL to YDB.
   • Migration with change of storage from PostgreSQL to MySQL®.
   • Migration with change of storage from PostgreSQL to OpenSearch.

2. Data change capture means tracking changes to a database and delivering those changes to consumers. It is used for applications that are sensitive to real-time data changes.

   • Capturing changes from PostgreSQL and delivering to YDS.
   • Capturing changes from PostgreSQL and delivering to Apache Kafka®.

3. Uploading data to data marts is a process of transferring prepared data to storage for subsequent visualization.

   • Loading data from PostgreSQL to the ClickHouse® data mart.

4. Uploading data to scalable Object Storage storage allows you to save on data storage and simplifies the exchange with contractors.

   • Loading PostgreSQL data to the Object Storage scalable storage.

For a detailed description of possible Yandex Data Transfer scenarios, see Tutorials.

Preparing the source databasePreparing the source database

Configuration

1. In the postgresql.conf file, change the value of the wal_level parameter to logical:

   wal_level = logical
   

2. Restart PostgreSQL.

SELECT NOW() - query_start AS duration, query, state
FROM pg_stat_activity
WHERE state != 'idle' ORDER BY 1 DESC;

ignore_slots:
  - database: <database>
    name: <replication_slot>
    plugin: wal2json
    type: logical

Warn(Termination): unable to create new pg source: Replication slotID <replication_slot_name> does not exist.

Configuring the PostgreSQL source endpointConfiguring the PostgreSQL source endpoint

When creating or updating an endpoint, you can define:

• Yandex Managed Service for PostgreSQL cluster connection or custom installation settings, including those based on Yandex Compute Cloud VMs. These are required parameters.
• Additional parameters.

Before you get started, check the Service specifics for PostgreSQL sources and targets.

Managed Service for PostgreSQL clusterManaged Service for PostgreSQL cluster

Connecting to the database with the cluster ID specified in Yandex Cloud.

resource "yandex_datatransfer_endpoint" "<endpoint_name_in_Terraform>" {
  name = "<endpoint_name>"
  settings {
    postgres_source {
      security_groups = ["<list_of_security_group_IDs>"]
      connection {
        mdb_cluster_id = "<cluster_ID>"
      }
      database = "<name_of_database_to_migrate>"
      user     = "<username_for_connection>"
      password {
        raw = "<user_password>"
      }
      <additional_endpoint_settings>
    }
  }
}

Custom installationCustom installation

For OnPremise, all fields are filled in manually.

resource "yandex_datatransfer_endpoint" "<endpoint_name_in_Terraform>" {
  name = "<endpoint_name>"
  settings {
    postgres_source {
      security_groups = ["<list_of_security_group_IDs>"]
      connection {
        on_premise {
          hosts = ["<list_of_hosts>"]
          port  = <port_for_connection>
        }
      }
      database = "<name_of_database_to_migrate>"
      user     = "<username_for_connection>"
      password {
        raw = "<user_password>"
      }
      <additional_endpoint_settings>
    }
  }
}

Additional settingsAdditional settings

Settings for transferring a DB schema when enabling and disabling a transferSettings for transferring a DB schema when enabling and disabling a transfer

During a transfer, the database schema is transferred from the source to the target. The transfer is performed in two stages:

1. At the activation stage.

   This step is performed when the transfer is activated before copying or replicating data to create a schema on the target. You can choose which parts of the schema will be migrated. By default, these are: TABLE, VIEW, PRIMARY KEY, SEQUENCE, SEQUENCE OWNED BY, RULE, TYPE, FUNCTION, and DEFAULT.

2. At the deactivation stage.

   This step is performed at the end of the transfer operation when it is deactivated. If the transfer keeps running in replication mode, the final stage of the transfer will be performed only when replication stops. You can choose which parts of the schema will be migrated.

   At the final stage, it is assumed that when the transfer is deactivated, there is no writing load on the source. You can ensure this by switching to read-only mode. At this stage, the database schema on the target is brought to a state where it will be consistent with the schema on the source.

   It is recommended to include resource-intensive operations, e.g., index migration, in the final stage of migration. Migrating indexes at the beginning of the transfer can slow it down.

To transfer the schema at both the initial and final stages, one uses the pg_dump utility.

Replication cannot guarantee that sequence values are preserved, so we recommend updating the sequence on the target.

Configuring the data targetConfiguring the data target

Configure one of the supported data targets:

• PostgreSQL
• MySQL®
• ClickHouse®
• Greenplum®
• Yandex Managed Service for YDB
• Yandex Object Storage
• Apache Kafka®
• YDS
• Elasticsearch
• OpenSearch

For a complete list of supported sources and targets in Yandex Data Transfer, see Available transfers.

After configuring the data source and target, create and start the transfer.

Operations with the database during transferOperations with the database during transfer

For Snapshot and Snapshot and increment transfers:

• In the Copying status, changes to the data schema on the source and target are not allowed.

• In the Replicating status, any changes to the data schema on the source should be manually applied to the target, otherwise the transfer will not be able to continue.

  For example, suppose we added a new column to the test_table table on the source:

  ALTER TABLE test_table ADD COLUMN val2 TEXT;
  

  If writing to this table continues, the transfer will not be able to insert data on the target. For replication to continue, run a similar query to change the data schema on the target:

  ALTER TABLE test_table ADD COLUMN val2 TEXT;
  

  After that, the transfer can continue.

Troubleshooting data transfer issuesTroubleshooting data transfer issues

Known issues when using a PostgreSQL endpoint:

• Stopping a transfer's master transaction session.
• Exceeding the connection time-to-live quota.
• VIEW transfer error.
• Error when adding a table entry by constraint.
• Error when migrating all schema tables.
• Unable to create objects involving extension functions.
• Low transfer speed.
• Unable to transfer child tables.
• Insufficient replication slots in a source database.
• No data transfer after changing a source endpoint.
• Transfer error when changing a master host.
• Error when transferring nested transactions.
• Error transferring tables with deferred constraints.
• Cannot create a replication slot at the activation step.
• Excessive WAL size increase.
• Error when replicating from an external source.
• Error when transferring tables without primary keys.
• Error when dropping a table under the Drop cleanup policy.

For more troubleshooting tips, see the Troubleshooting section.

Stopping a transfer's master transaction sessionStopping a transfer's master transaction session

Error message:

Cannot set transaction snapshot:
ERROR: invalid snapshot identifier: "<snapshot_ID>" (SQLSTATE 22023).

Possible causes include:

• The source is running a cron job or other application that periodically terminates sessions that are too long.
• Someone manually terminated the master transaction.
• The source's CPU resources are insufficient to execute a query.

Solution: Disable this cron job and add more CPU resources to the source. After you make the changes, re-activate the transfer.

Exceeding the connection time-to-live quotaExceeding the connection time-to-live quota

Yandex Managed Service for PostgreSQL has a connection time-to-live quota of 12 hours.
​​
Solution: If transferring a database requires more time, [change the Session duration timeout setting of the Yandex Managed Service for PostgreSQL cluster](../../../../managed-postgresql/operations/update.md#change-postgresql-config).

VIEW transfer errorVIEW transfer error

Error message:

[ERROR] "... _view": no key columns found

Can't replicate new data from Views. Under PostgreSQL — PostgreSQL transfers, only those views are transferred which are listed in the Table filter → Included tables source endpoint parameter.

Solution: manually exclude all views from the transfer, list them in the Table filter → Included tables source endpoint parameter, and then reactivate the transfer.

Error when adding a table entry by constraintError when adding a table entry by constraint

Solution: prepare the source as described in Preparing for the transfer.

Error when transferring all schema tablesError when transferring all schema tables

Error message:

Unable to apply DDL of type 'TABLE', name '<schema>'.'<table>', error:
ERROR: schema "<schema name>" does not exist (SQLSTATE 3F000)

A transfer fails if tables of a specific schema are listed as <schema>.*. This occurs due to the specifics of pg_dump that is used for a schema transfer. When specifying tables of the entire schema in <schema>.* format in the source endpoint parameter Table filter → Included tables, the PostgreSQL types from this schema aren't extracted even if it contains tables that depend on these types.

Solution: Create PostgreSQL types in the target database manually.

Unable to create objects involving extension functionsUnable to create objects involving extension functions

Error message:

Unable to apply DDL of type 'TABLE', <object_name>, error:
failed to push non-row item 0 of kind "pg:DDL" in batch:
Push failed: ERROR: function <schema_name>.<function_name>() does not exist 
(SQLSTATE 42883)

In Managed Service for PostgreSQL, you cannot install an extension into a custom schema in a target database. This means the transfer will be interrupted if the user installation of Managed Service for PostgreSQL has extensions installed in a custom schema and these extensions are used in the DDLs of objects you are moving.

Solution: check the DDLs of objects whose names appear in the error. If these objects can call a function from a custom schema, manually create DDLs in the target that call functions without specifying a schema. Set the target endpoint cleanup policy to Truncate to prevent the transfer from deleting these objects.

Low transfer speedLow transfer speed

Snapshot or Snapshot and increment transfers from PostgreSQL to PostgreSQL may slow down.

Possible cause:

• Write protocol.

  Under normal operation, a transfer is performed via the fast copy protocol. However, if there's a batch write conflict, it switches to slow line-by-line writes. The more write conflicts, the lower the transfer speed.

  Solution: set the target endpoint's cleanup policy type to Drop and exclude other writing processes.

• Parallel table reading.

  Parallel reading is only available for tables that contain a primary key in serial mode.

  Solution: Set up parallel copy and reactivate the transfer.

Unable to transfer child tablesUnable to transfer child tables

Child tables are either not transferred or transferred without data if a table is partitioned.

Solution: set the following source endpoint parameters:

1. Select Merge inherited tables in advanced settings.
2. In the Included tables field, specify all child tables to transfer data from.
3. Make sure the user has access to the child tables.

To speed up the transfer of child tables, set up parallel copy.

Insufficient replication slots in a source databaseInsufficient replication slots in a source database

Error message:

Warn(Activate): failed to create a replication slot "<transfer_ID>" at source:
failed to create a replication slot:
failed to create a replication slot:
ERROR: all replication slots are in use
(SQLSTATE 53400)

Solution: Increase the number of replication slots in the source database (by default, 10).

No data transfer after changing a source endpointNo data transfer after changing a source endpoint

After adding tables to the List of included tables in the source endpoint parameters, a data transfer restarted and stopped.

Solution:

• Create tables in the target database manually.

  1. Create new tables with a Primary key and without a Foreign key in the target database.
  2. Add the new tables to the List of included tables in the source endpoint parameters.
  3. Transfer a dump with historical data to the target database.
  4. If there are errors in the logs, fix them depending on a specific error.
  5. If there are no errors but the logs are empty, contact technical support or your account manager to dump goroutines. This may help resume replication without restarting the transfer.

• Deactivate and reactivate the transfer.

• Create a separate transfer of the Snapshot type for the new tables. Deactivating the original transfer is not required.

Transfer error when changing a master hostTransfer error when changing a master host

The error occurs in Replication or Snapshot and increment transfers due to required Write-Ahead Log (WAL) parts missing. This occurs when the Write Ahead Log (WAL) logical replication lag between the current master and a replica is greater than the allowable WAL size on other hosts, so, when switching from the master to the replica, the replication slot can't sync to the WAL on the new master.

Solution: set a limit in the additional target endpoint parameter Maximum WAL size for the replication slot and reactivate the transfer.

Error when transferring nested transactionsError when transferring nested transactions

Transfers of PostgreSQL below version 14 don't support a transfer of tables with committed transactions that are nested more than 1024 times with replication changes at each nesting level. The degree of nesting depends on the number of nested begin; .. commit; statements.

Solution:

• Use PostgreSQL 14 or higher.
• Exclude transactions with a nesting level like this from a transfer.

Error transferring tables with deferred constraintsError transferring tables with deferred constraints

The error arises in the transfers of type Replication or Snapshot and increment because updating tables and transactions with DEFERRABLE constraints isn't supported by Data Transfer. To learn more about deferred constraints, see the PostgreSQL documentation.

Solution: Change the constraint type in such tables to IMMEDIATE and activate the transfer again.

Could not create a replication slot at the activation stepCould not create a replication slot at the activation step

In the beginning of the transfer, one or more replication slots are created in the source database. The database objects are locked at this point. If some object is locked by another transaction, this results in a competing lock, which will terminate the transfer with an error.

Solution:

1. Get the PID of the process that competes for locks with the transfer:

   /* Get PID of the transfer */
   SELECT active_pid
     FROM pg_replication_slots
     WHERE slot_name = '<transfer_ID>';
   
   /* search the PID of the locking process */
   SELECT pid, pg_blocking_pids(pid) as blocked_by
     FROM pg_stat_activity
     WHERE cardinality(pg_blocking_pids(pid)) > 0;
   

           pid      | blocked_by
   -----------------+-------------------
    <transfer_PID> | {<locking_transaction_PID>}
   (1 row)
   

2. Look up the locking query:

   SELECT query, username
     FROM pg_stat_activity
     WHERE pid = <locking_transaction_PID>;
   

3. (Optional) Stop the transaction by the command:

   SELECT pg_terminate_backend(<locking_transaction_PID>);
   

4. Reactivate the transfer.

Excessive WAL size increaseExcessive WAL size increase

In the PostgreSQL source database, the size of the Write-Ahead Log (WAL) may reach up to dozens of GB due to the queries running for over five minutes. These queries block the WAL and do not allow it to move forward and reset.

You can see the WAL size has gone up when:

• The space used by the source database has increased.
• The Read buffer size chart in Data Transfer monitoring is rising.

Solution:

1. Find the query sessions running for more than five minutes:

   SELECT now()-xact_start, query, pid FROM pg_stat_activity
   WHERE (now()-xact_start)>'5minute'::interval AND STATE != 'idle'
   ORDER BY 1 DESC;
   

2. Terminate the sessions found. Make sure to avoid such queries moving forward.

Error when replicating data from an external sourceError when replicating data from an external source

Error message:

[XX000] ERROR: could not connect to the publisher:
SSL error: certificate verify failed FATAL:
no pg_hba.conf entry for replication connection
from host "<PostgreSQL_host_IP_address>", user "postgres", SSL off

Solution: Prepare the source as described in Preparing for the transfer.

Error when transfering tables without primary keysError when transfering tables without primary keys

Error message:

Primary key check failed: 14 Tables errors: Table no key columns found

For Replication and Snapshot and increment transfers, tables without primary keys are not transferred.

Solution: Prepare the source as described in Preparing for the transfer.

Error when dropping a table under the Drop cleanup policyError when dropping a table under the Drop cleanup policy

Error message:

ERROR: cannot drop table <table_name> because other objects depend on it (SQLSTATE 2BP01)

With the Drop cleanup policy, a transfer drops tables in multiple iterations:

1. The transfer tries to drop all tables one by one. Cascade delete is not used, as it may delete tables that are not included in the transfer. If a table cannot be dropped, for example, due to its links to external keys, an error occurs, but the transfer will continue dropping tables.

2. During the next iteration, the transfer will try to drop the remaining tables. If the blocking child tables were dropped in the previous iteration, the table linked to external keys will also be dropped. In this case, the error is fixed while Data Transfer is running and no other actions are required.

3. If the transfer does not drop any table during another iteration, table dropping stops. In which case:

   • The transfer will continue running if all tables were dropped.
   • The transfer will be aborted with an error if there are any tables left.

Solution:

• If child tables are not involved in other transfers and their transfer is not in conflict with the transfer goals, add these tables:

  • To the list of included tables in the source endpoint parameters.
  • To the list of objects to be transferred in the transfer parameters.

• Delete the blocking child tables in the target database manually.

• Use the Truncate cleanup policy.

• Recreate the target database.

  Warning

  This will lead to losing all data in the database.