Logical Decoding PostgreSQL provides infrastructure to stream the modifications performed via SQL to external consumers. This functionality can be used for a variety of purposes, including replication solutions and auditing. Changes are sent out in streams identified by logical replication slots. The format in which those changes are streamed is determined by the output plugin used. An example plugin is provided in the PostgreSQL distribution. Additional plugins can be written to extend the choice of available formats without modifying any core code. Every output plugin has access to each individual new row produced by INSERT and the new row version created by UPDATE. Availability of old row versions for UPDATE and DELETE depends on the configured replica identity (see ). Changes can be consumed either using the streaming replication protocol (see and ), or by calling functions via SQL (see ). It is also possible to write additional methods of consuming the output of a replication slot without modifying core code (see ). The following example demonstrates controlling logical decoding using the SQL interface. Before you can use logical decoding, you must set to logical and to at least 1. Then, you should connect to the target database (in the example below, postgres) as a superuser. postgres=# -- Create a slot named 'regression_slot' using the output plugin 'test_decoding' postgres=# SELECT * FROM pg_create_logical_replication_slot('regression_slot', 'test_decoding', false, true); slot_name | lsn -----------------+----------- regression_slot | 0/16B1970 (1 row) postgres=# SELECT slot_name, plugin, slot_type, database, active, restart_lsn, confirmed_flush_lsn FROM pg_replication_slots; slot_name | plugin | slot_type | database | active | restart_lsn | confirmed_flush_lsn -----------------+---------------+-----------+----------+--------+-------------+----------------- regression_slot | test_decoding | logical | postgres | f | 0/16A4408 | 0/16A4440 (1 row) postgres=# -- There are no changes to see yet postgres=# SELECT * FROM pg_logical_slot_get_changes('regression_slot', NULL, NULL); lsn | xid | data -----+-----+------ (0 rows) postgres=# CREATE TABLE data(id serial primary key, data text); CREATE TABLE postgres=# -- DDL isn't replicated, so all you'll see is the transaction postgres=# SELECT * FROM pg_logical_slot_get_changes('regression_slot', NULL, NULL); lsn | xid | data -----------+-------+-------------- 0/BA2DA58 | 10297 | BEGIN 10297 0/BA5A5A0 | 10297 | COMMIT 10297 (2 rows) postgres=# -- Once changes are read, they're consumed and not emitted postgres=# -- in a subsequent call: postgres=# SELECT * FROM pg_logical_slot_get_changes('regression_slot', NULL, NULL); lsn | xid | data -----+-----+------ (0 rows) postgres=# BEGIN; postgres=*# INSERT INTO data(data) VALUES('1'); postgres=*# INSERT INTO data(data) VALUES('2'); postgres=*# COMMIT; postgres=# SELECT * FROM pg_logical_slot_get_changes('regression_slot', NULL, NULL); lsn | xid | data -----------+-------+--------------------------------------------------------- 0/BA5A688 | 10298 | BEGIN 10298 0/BA5A6F0 | 10298 | table public.data: INSERT: id[integer]:1 data[text]:'1' 0/BA5A7F8 | 10298 | table public.data: INSERT: id[integer]:2 data[text]:'2' 0/BA5A8A8 | 10298 | COMMIT 10298 (4 rows) postgres=# INSERT INTO data(data) VALUES('3'); postgres=# -- You can also peek ahead in the change stream without consuming changes postgres=# SELECT * FROM pg_logical_slot_peek_changes('regression_slot', NULL, NULL); lsn | xid | data -----------+-------+--------------------------------------------------------- 0/BA5A8E0 | 10299 | BEGIN 10299 0/BA5A8E0 | 10299 | table public.data: INSERT: id[integer]:3 data[text]:'3' 0/BA5A990 | 10299 | COMMIT 10299 (3 rows) postgres=# -- The next call to pg_logical_slot_peek_changes() returns the same changes again postgres=# SELECT * FROM pg_logical_slot_peek_changes('regression_slot', NULL, NULL); lsn | xid | data -----------+-------+--------------------------------------------------------- 0/BA5A8E0 | 10299 | BEGIN 10299 0/BA5A8E0 | 10299 | table public.data: INSERT: id[integer]:3 data[text]:'3' 0/BA5A990 | 10299 | COMMIT 10299 (3 rows) postgres=# -- options can be passed to output plugin, to influence the formatting postgres=# SELECT * FROM pg_logical_slot_peek_changes('regression_slot', NULL, NULL, 'include-timestamp', 'on'); lsn | xid | data -----------+-------+--------------------------------------------------------- 0/BA5A8E0 | 10299 | BEGIN 10299 0/BA5A8E0 | 10299 | table public.data: INSERT: id[integer]:3 data[text]:'3' 0/BA5A990 | 10299 | COMMIT 10299 (at 2017-05-10 12:07:21.272494-04) (3 rows) postgres=# -- Remember to destroy a slot you no longer need to stop it consuming postgres=# -- server resources: postgres=# SELECT pg_drop_replication_slot('regression_slot'); pg_drop_replication_slot ----------------------- (1 row) The following example shows how logical decoding is controlled over the streaming replication protocol, using the program included in the PostgreSQL distribution. This requires that client authentication is set up to allow replication connections (see ) and that max_wal_senders is set sufficiently high to allow an additional connection. $ pg_recvlogical -d postgres --slot=test --create-slot $ pg_recvlogical -d postgres --slot=test --start -f - ControlZ $ psql -d postgres -c "INSERT INTO data(data) VALUES('4');" $ fg BEGIN 693 table public.data: INSERT: id[integer]:4 data[text]:'4' COMMIT 693 ControlC $ pg_recvlogical -d postgres --slot=test --drop-slot The following example shows SQL interface that can be used to decode prepared transactions. Before you use two-phase commit commands, you must set max_prepared_transactions to at least 1. You must also have set the two-phase parameter as 'true' while creating the slot using pg_create_logical_replication_slot Note that we will stream the entire transaction after the commit if it is not already decoded. postgres=# BEGIN; postgres=*# INSERT INTO data(data) VALUES('5'); postgres=*# PREPARE TRANSACTION 'test_prepared1'; postgres=# SELECT * FROM pg_logical_slot_get_changes('regression_slot', NULL, NULL); lsn | xid | data -----------+-----+--------------------------------------------------------- 0/1689DC0 | 529 | BEGIN 529 0/1689DC0 | 529 | table public.data: INSERT: id[integer]:3 data[text]:'5' 0/1689FC0 | 529 | PREPARE TRANSACTION 'test_prepared1', txid 529 (3 rows) postgres=# COMMIT PREPARED 'test_prepared1'; postgres=# select * from pg_logical_slot_get_changes('regression_slot', NULL, NULL); lsn | xid | data -----------+-----+-------------------------------------------- 0/168A060 | 529 | COMMIT PREPARED 'test_prepared1', txid 529 (4 row) postgres=#-- you can also rollback a prepared transaction postgres=# BEGIN; postgres=*# INSERT INTO data(data) VALUES('6'); postgres=*# PREPARE TRANSACTION 'test_prepared2'; postgres=# select * from pg_logical_slot_get_changes('regression_slot', NULL, NULL); lsn | xid | data -----------+-----+--------------------------------------------------------- 0/168A180 | 530 | BEGIN 530 0/168A1E8 | 530 | table public.data: INSERT: id[integer]:4 data[text]:'6' 0/168A430 | 530 | PREPARE TRANSACTION 'test_prepared2', txid 530 (3 rows) postgres=# ROLLBACK PREPARED 'test_prepared2'; postgres=# select * from pg_logical_slot_get_changes('regression_slot', NULL, NULL); lsn | xid | data -----------+-----+---------------------------------------------- 0/168A4B8 | 530 | ROLLBACK PREPARED 'test_prepared2', txid 530 (1 row) Logical Decoding Logical decoding is the process of extracting all persistent changes to a database's tables into a coherent, easy to understand format which can be interpreted without detailed knowledge of the database's internal state. In PostgreSQL, logical decoding is implemented by decoding the contents of the write-ahead log, which describe changes on a storage level, into an application-specific form such as a stream of tuples or SQL statements. replication slot logical replication In the context of logical replication, a slot represents a stream of changes that can be replayed to a client in the order they were made on the origin server. Each slot streams a sequence of changes from a single database. PostgreSQL also has streaming replication slots (see ), but they are used somewhat differently there. A replication slot has an identifier that is unique across all databases in a PostgreSQL cluster. Slots persist independently of the connection using them and are crash-safe. A logical slot will emit each change just once in normal operation. The current position of each slot is persisted only at checkpoint, so in the case of a crash the slot may return to an earlier LSN, which will then cause recent changes to be sent again when the server restarts. Logical decoding clients are responsible for avoiding ill effects from handling the same message more than once. Clients may wish to record the last LSN they saw when decoding and skip over any repeated data or (when using the replication protocol) request that decoding start from that LSN rather than letting the server determine the start point. The Replication Progress Tracking feature is designed for this purpose, refer to replication origins. Multiple independent slots may exist for a single database. Each slot has its own state, allowing different consumers to receive changes from different points in the database change stream. For most applications, a separate slot will be required for each consumer. A logical replication slot knows nothing about the state of the receiver(s). It's even possible to have multiple different receivers using the same slot at different times; they'll just get the changes following on from when the last receiver stopped consuming them. Only one receiver may consume changes from a slot at any given time. Replication slots persist across crashes and know nothing about the state of their consumer(s). They will prevent removal of required resources even when there is no connection using them. This consumes storage because neither required WAL nor required rows from the system catalogs can be removed by VACUUM as long as they are required by a replication slot. In extreme cases this could cause the database to shut down to prevent transaction ID wraparound (see ). So if a slot is no longer required it should be dropped. Output plugins transform the data from the write-ahead log's internal representation into the format the consumer of a replication slot desires. When a new replication slot is created using the streaming replication interface (see ), a snapshot is exported (see ), which will show exactly the state of the database after which all changes will be included in the change stream. This can be used to create a new replica by using SET TRANSACTION SNAPSHOT to read the state of the database at the moment the slot was created. This transaction can then be used to dump the database's state at that point in time, which afterwards can be updated using the slot's contents without losing any changes. Creation of a snapshot is not always possible. In particular, it will fail when connected to a hot standby. Applications that do not require snapshot export may suppress it with the NOEXPORT_SNAPSHOT option. The commands CREATE_REPLICATION_SLOT slot_name LOGICAL output_plugin DROP_REPLICATION_SLOT slot_name WAIT START_REPLICATION SLOT slot_name LOGICAL ... are used to create, drop, and stream changes from a replication slot, respectively. These commands are only available over a replication connection; they cannot be used via SQL. See for details on these commands. The command can be used to control logical decoding over a streaming replication connection. (It uses these commands internally.) See for detailed documentation on the SQL-level API for interacting with logical decoding. Synchronous replication (see ) is only supported on replication slots used over the streaming replication interface. The function interface and additional, non-core interfaces do not support synchronous replication. The pg_replication_slots view and the pg_stat_replication view provide information about the current state of replication slots and streaming replication connections respectively. These views apply to both physical and logical replication. The pg_stat_replication_slots view provides statistics information about the logical replication slots. An example output plugin can be found in the contrib/test_decoding subdirectory of the PostgreSQL source tree. _PG_output_plugin_init An output plugin is loaded by dynamically loading a shared library with the output plugin's name as the library base name. The normal library search path is used to locate the library. To provide the required output plugin callbacks and to indicate that the library is actually an output plugin it needs to provide a function named _PG_output_plugin_init. This function is passed a struct that needs to be filled with the callback function pointers for individual actions. typedef struct OutputPluginCallbacks { LogicalDecodeStartupCB startup_cb; LogicalDecodeBeginCB begin_cb; LogicalDecodeChangeCB change_cb; LogicalDecodeTruncateCB truncate_cb; LogicalDecodeCommitCB commit_cb; LogicalDecodeMessageCB message_cb; LogicalDecodeFilterByOriginCB filter_by_origin_cb; LogicalDecodeShutdownCB shutdown_cb; LogicalDecodeFilterPrepareCB filter_prepare_cb; LogicalDecodeBeginPrepareCB begin_prepare_cb; LogicalDecodePrepareCB prepare_cb; LogicalDecodeCommitPreparedCB commit_prepared_cb; LogicalDecodeRollbackPreparedCB rollback_prepared_cb; LogicalDecodeStreamStartCB stream_start_cb; LogicalDecodeStreamStopCB stream_stop_cb; LogicalDecodeStreamAbortCB stream_abort_cb; LogicalDecodeStreamPrepareCB stream_prepare_cb; LogicalDecodeStreamCommitCB stream_commit_cb; LogicalDecodeStreamChangeCB stream_change_cb; LogicalDecodeStreamMessageCB stream_message_cb; LogicalDecodeStreamTruncateCB stream_truncate_cb; } OutputPluginCallbacks; typedef void (*LogicalOutputPluginInit) (struct OutputPluginCallbacks *cb); The begin_cb, change_cb and commit_cb callbacks are required, while startup_cb, filter_by_origin_cb, truncate_cb, and shutdown_cb are optional. If truncate_cb is not set but a TRUNCATE is to be decoded, the action will be ignored. An output plugin may also define functions to support streaming of large, in-progress transactions. The stream_start_cb, stream_stop_cb, stream_abort_cb, stream_commit_cb, stream_change_cb, and stream_prepare_cb are required, while stream_message_cb and stream_truncate_cb are optional. An output plugin may also define functions to support two-phase commits, which allows actions to be decoded on the PREPARE TRANSACTION. The begin_prepare_cb, prepare_cb, stream_prepare_cb, commit_prepared_cb and rollback_prepared_cb callbacks are required, while filter_prepare_cb is optional. To decode, format and output changes, output plugins can use most of the backend's normal infrastructure, including calling output functions. Read only access to relations is permitted as long as only relations are accessed that either have been created by initdb in the pg_catalog schema, or have been marked as user provided catalog tables using ALTER TABLE user_catalog_table SET (user_catalog_table = true); CREATE TABLE another_catalog_table(data text) WITH (user_catalog_table = true); Note that access to user catalog tables or regular system catalog tables in the output plugins has to be done via the systable_* scan APIs only. Access via the heap_* scan APIs will error out. Additionally, any actions leading to transaction ID assignment are prohibited. That, among others, includes writing to tables, performing DDL changes, and calling pg_current_xact_id(). Output plugin callbacks can pass data to the consumer in nearly arbitrary formats. For some use cases, like viewing the changes via SQL, returning data in a data type that can contain arbitrary data (e.g., bytea) is cumbersome. If the output plugin only outputs textual data in the server's encoding, it can declare that by setting OutputPluginOptions.output_type to OUTPUT_PLUGIN_TEXTUAL_OUTPUT instead of OUTPUT_PLUGIN_BINARY_OUTPUT in the startup callback. In that case, all the data has to be in the server's encoding so that a text datum can contain it. This is checked in assertion-enabled builds. An output plugin gets notified about changes that are happening via various callbacks it needs to provide. Concurrent transactions are decoded in commit order, and only changes belonging to a specific transaction are decoded between the begin and commit callbacks. Transactions that were rolled back explicitly or implicitly never get decoded. Successful savepoints are folded into the transaction containing them in the order they were executed within that transaction. A transaction that is prepared for a two-phase commit using PREPARE TRANSACTION will also be decoded if the output plugin callbacks needed for decoding them are provided. It is possible that the current prepared transaction which is being decoded is aborted concurrently via a ROLLBACK PREPARED command. In that case, the logical decoding of this transaction will be aborted too. All the changes of such a transaction are skipped once the abort is detected and the prepare_cb callback is invoked. Thus even in case of a concurrent abort, enough information is provided to the output plugin for it to properly deal with ROLLBACK PREPARED once that is decoded. Only transactions that have already safely been flushed to disk will be decoded. That can lead to a COMMIT not immediately being decoded in a directly following pg_logical_slot_get_changes() when synchronous_commit is set to off. The optional startup_cb callback is called whenever a replication slot is created or asked to stream changes, independent of the number of changes that are ready to be put out. typedef void (*LogicalDecodeStartupCB) (struct LogicalDecodingContext *ctx, OutputPluginOptions *options, bool is_init); The is_init parameter will be true when the replication slot is being created and false otherwise. options points to a struct of options that output plugins can set: typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; } OutputPluginOptions; output_type has to either be set to OUTPUT_PLUGIN_TEXTUAL_OUTPUT or OUTPUT_PLUGIN_BINARY_OUTPUT. See also . If receive_rewrites is true, the output plugin will also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. The startup callback should validate the options present in ctx->output_plugin_options. If the output plugin needs to have a state, it can use ctx->output_plugin_private to store it. The optional shutdown_cb callback is called whenever a formerly active replication slot is not used anymore and can be used to deallocate resources private to the output plugin. The slot isn't necessarily being dropped, streaming is just being stopped. typedef void (*LogicalDecodeShutdownCB) (struct LogicalDecodingContext *ctx); The required begin_cb callback is called whenever a start of a committed transaction has been decoded. Aborted transactions and their contents never get decoded. typedef void (*LogicalDecodeBeginCB) (struct LogicalDecodingContext *ctx, ReorderBufferTXN *txn); The txn parameter contains meta information about the transaction, like the time stamp at which it has been committed and its XID. The required commit_cb callback is called whenever a transaction commit has been decoded. The change_cb callbacks for all modified rows will have been called before this, if there have been any modified rows. typedef void (*LogicalDecodeCommitCB) (struct LogicalDecodingContext *ctx, ReorderBufferTXN *txn, XLogRecPtr commit_lsn); The required change_cb callback is called for every individual row modification inside a transaction, may it be an INSERT, UPDATE, or DELETE. Even if the original command modified several rows at once the callback will be called individually for each row. The change_cb callback may access system or user catalog tables to aid in the process of outputting the row modification details. In case of decoding a prepared (but yet uncommitted) transaction or decoding of an uncommitted transaction, this change callback might also error out due to simultaneous rollback of this very same transaction. In that case, the logical decoding of this aborted transaction is stopped gracefully. typedef void (*LogicalDecodeChangeCB) (struct LogicalDecodingContext *ctx, ReorderBufferTXN *txn, Relation relation, ReorderBufferChange *change); The ctx and txn parameters have the same contents as for the begin_cb and commit_cb callbacks, but additionally the relation descriptor relation points to the relation the row belongs to and a struct change describing the row modification are passed in. Only changes in user defined tables that are not unlogged (see ) and not temporary (see ) can be extracted using logical decoding. The truncate_cb callback is called for a TRUNCATE command. typedef void (*LogicalDecodeTruncateCB) (struct LogicalDecodingContext *ctx, ReorderBufferTXN *txn, int nrelations, Relation relations[], ReorderBufferChange *change); The parameters are analogous to the change_cb callback. However, because TRUNCATE actions on tables connected by foreign keys need to be executed together, this callback receives an array of relations instead of just a single one. See the description of the statement for details. The optional filter_by_origin_cb callback is called to determine whether data that has been replayed from origin_id is of interest to the output plugin. typedef bool (*LogicalDecodeFilterByOriginCB) (struct LogicalDecodingContext *ctx, RepOriginId origin_id); The ctx parameter has the same contents as for the other callbacks. No information but the origin is available. To signal that changes originating on the passed in node are irrelevant, return true, causing them to be filtered away; false otherwise. The other callbacks will not be called for transactions and changes that have been filtered away. This is useful when implementing cascading or multidirectional replication solutions. Filtering by the origin allows to prevent replicating the same changes back and forth in such setups. While transactions and changes also carry information about the origin, filtering via this callback is noticeably more efficient. The optional message_cb callback is called whenever a logical decoding message has been decoded. typedef void (*LogicalDecodeMessageCB) (struct LogicalDecodingContext *ctx, ReorderBufferTXN *txn, XLogRecPtr message_lsn, bool transactional, const char *prefix, Size message_size, const char *message); The txn parameter contains meta information about the transaction, like the time stamp at which it has been committed and its XID. Note however that it can be NULL when the message is non-transactional and the XID was not assigned yet in the transaction which logged the message. The lsn has WAL location of the message. The transactional says if the message was sent as transactional or not. Similar to the change callback, in case of decoding a prepared (but yet uncommitted) transaction or decoding of an uncommitted transaction, this message callback might also error out due to simultaneous rollback of this very same transaction. In that case, the logical decoding of this aborted transaction is stopped gracefully. The prefix is arbitrary null-terminated prefix which can be used for identifying interesting messages for the current plugin. And finally the message parameter holds the actual message of message_size size. Extra care should be taken to ensure that the prefix the output plugin considers interesting is unique. Using name of the extension or the output plugin itself is often a good choice. The optional filter_prepare_cb callback is called to determine whether data that is part of the current two-phase commit transaction should be considered for decoding at this prepare stage or later as a regular one-phase transaction at COMMIT PREPARED time. To signal that decoding should be skipped, return true; false otherwise. When the callback is not defined, false is assumed (i.e. no filtering, all transactions using two-phase commit are decoded in two phases as well). typedef bool (*LogicalDecodeFilterPrepareCB) (struct LogicalDecodingContext *ctx, TransactionId xid, const char *gid); The ctx parameter has the same contents as for the other callbacks. The parameters xid and gid provide two different ways to identify the transaction. The later COMMIT PREPARED or ROLLBACK PREPARED carries both identifiers, providing an output plugin the choice of what to use. The callback may be invoked multiple times per transaction to decode and must provide the same static answer for a given pair of xid and gid every time it is called. The required begin_prepare_cb callback is called whenever the start of a prepared transaction has been decoded. The gid field, which is part of the txn parameter, can be used in this callback to check if the plugin has already received this PREPARE in which case it can either error out or skip the remaining changes of the transaction. typedef void (*LogicalDecodeBeginPrepareCB) (struct LogicalDecodingContext *ctx, ReorderBufferTXN *txn); The required prepare_cb callback is called whenever a transaction which is prepared for two-phase commit has been decoded. The change_cb callback for all modified rows will have been called before this, if there have been any modified rows. The gid field, which is part of the txn parameter, can be used in this callback. typedef void (*LogicalDecodePrepareCB) (struct LogicalDecodingContext *ctx, ReorderBufferTXN *txn, XLogRecPtr prepare_lsn); The required commit_prepared_cb callback is called whenever a transaction COMMIT PREPARED has been decoded. The gid field, which is part of the txn parameter, can be used in this callback. typedef void (*LogicalDecodeCommitPreparedCB) (struct LogicalDecodingContext *ctx, ReorderBufferTXN *txn, XLogRecPtr commit_lsn); The required rollback_prepared_cb callback is called whenever a transaction ROLLBACK PREPARED has been decoded. The gid field, which is part of the txn parameter, can be used in this callback. The parameters prepare_end_lsn and prepare_time can be used to check if the plugin has received this PREPARE TRANSACTION in which case it can apply the rollback, otherwise, it can skip the rollback operation. The gid alone is not sufficient because the downstream node can have a prepared transaction with same identifier. typedef void (*LogicalDecodeRollbackPreparedCB) (struct LogicalDecodingContext *ctx, ReorderBufferTXN *txn, XLogRecPtr prepare_end_lsn, TimestampTz prepare_time); The stream_start_cb callback is called when opening a block of streamed changes from an in-progress transaction. typedef void (*LogicalDecodeStreamStartCB) (struct LogicalDecodingContext *ctx, ReorderBufferTXN *txn); The stream_stop_cb callback is called when closing a block of streamed changes from an in-progress transaction. typedef void (*LogicalDecodeStreamStopCB) (struct LogicalDecodingContext *ctx, ReorderBufferTXN *txn); The stream_abort_cb callback is called to abort a previously streamed transaction. typedef void (*LogicalDecodeStreamAbortCB) (struct LogicalDecodingContext *ctx, ReorderBufferTXN *txn, XLogRecPtr abort_lsn); The stream_prepare_cb callback is called to prepare a previously streamed transaction as part of a two-phase commit. typedef void (*LogicalDecodeStreamPrepareCB) (struct LogicalDecodingContext *ctx, ReorderBufferTXN *txn, XLogRecPtr prepare_lsn); The stream_commit_cb callback is called to commit a previously streamed transaction. typedef void (*LogicalDecodeStreamCommitCB) (struct LogicalDecodingContext *ctx, ReorderBufferTXN *txn, XLogRecPtr commit_lsn); The stream_change_cb callback is called when sending a change in a block of streamed changes (demarcated by stream_start_cb and stream_stop_cb calls). The actual changes are not displayed as the transaction can abort at a later point in time and we don't decode changes for aborted transactions. typedef void (*LogicalDecodeStreamChangeCB) (struct LogicalDecodingContext *ctx, ReorderBufferTXN *txn, Relation relation, ReorderBufferChange *change); The stream_message_cb callback is called when sending a generic message in a block of streamed changes (demarcated by stream_start_cb and stream_stop_cb calls). The message contents for transactional messages are not displayed as the transaction can abort at a later point in time and we don't decode changes for aborted transactions. typedef void (*LogicalDecodeStreamMessageCB) (struct LogicalDecodingContext *ctx, ReorderBufferTXN *txn, XLogRecPtr message_lsn, bool transactional, const char *prefix, Size message_size, const char *message); The stream_truncate_cb callback is called for a TRUNCATE command in a block of streamed changes (demarcated by stream_start_cb and stream_stop_cb calls). typedef void (*LogicalDecodeStreamTruncateCB) (struct LogicalDecodingContext *ctx, ReorderBufferTXN *txn, int nrelations, Relation relations[], ReorderBufferChange *change); The parameters are analogous to the stream_change_cb callback. However, because TRUNCATE actions on tables connected by foreign keys need to be executed together, this callback receives an array of relations instead of just a single one. See the description of the statement for details. To actually produce output, output plugins can write data to the StringInfo output buffer in ctx->out when inside the begin_cb, commit_cb, or change_cb callbacks. Before writing to the output buffer, OutputPluginPrepareWrite(ctx, last_write) has to be called, and after finishing writing to the buffer, OutputPluginWrite(ctx, last_write) has to be called to perform the write. The last_write indicates whether a particular write was the callback's last write. The following example shows how to output data to the consumer of an output plugin: OutputPluginPrepareWrite(ctx, true); appendStringInfo(ctx->out, "BEGIN %u", txn->xid); OutputPluginWrite(ctx, true); It is possible to add more output methods for logical decoding. For details, see src/backend/replication/logical/logicalfuncs.c. Essentially, three functions need to be provided: one to read WAL, one to prepare writing output, and one to write the output (see ). Logical decoding can be used to build synchronous replication solutions with the same user interface as synchronous replication for streaming replication. To do this, the streaming replication interface (see ) must be used to stream out data. Clients have to send Standby status update (F) (see ) messages, just like streaming replication clients do. A synchronous replica receiving changes via logical decoding will work in the scope of a single database. Since, in contrast to that, synchronous_standby_names currently is server wide, this means this technique will not work properly if more than one database is actively used. In synchronous replication setup, a deadlock can happen, if the transaction has locked [user] catalog tables exclusively. See for information on user catalog tables. This is because logical decoding of transactions can lock catalog tables to access them. To avoid this users must refrain from taking an exclusive lock on [user] catalog tables. This can happen in the following ways: Issuing an explicit LOCK on pg_class in a transaction. Perform CLUSTER on pg_class in a transaction. PREPARE TRANSACTION after LOCK command on pg_class and allow logical decoding of two-phase transactions. PREPARE TRANSACTION after CLUSTER command on pg_trigger and allow logical decoding of two-phase transactions. This will lead to deadlock only when published table have a trigger. Executing TRUNCATE on [user] catalog table in a transaction. Note that these commands that can cause deadlock apply to not only explicitly indicated system catalog tables above but also to any other [user] catalog table. The basic output plugin callbacks (e.g., begin_cb, change_cb, commit_cb and message_cb) are only invoked when the transaction actually commits. The changes are still decoded from the transaction log, but are only passed to the output plugin at commit (and discarded if the transaction aborts). This means that while the decoding happens incrementally, and may spill to disk to keep memory usage under control, all the decoded changes have to be transmitted when the transaction finally commits (or more precisely, when the commit is decoded from the transaction log). Depending on the size of the transaction and network bandwidth, the transfer time may significantly increase the apply lag. To reduce the apply lag caused by large transactions, an output plugin may provide additional callback to support incremental streaming of in-progress transactions. There are multiple required streaming callbacks (stream_start_cb, stream_stop_cb, stream_abort_cb, stream_commit_cb and stream_change_cb) and two optional callbacks (stream_message_cb and stream_truncate_cb). When streaming an in-progress transaction, the changes (and messages) are streamed in blocks demarcated by stream_start_cb and stream_stop_cb callbacks. Once all the decoded changes are transmitted, the transaction can be committed using the the stream_commit_cb callback (or possibly aborted using the stream_abort_cb callback). If two-phase commits are supported, the transaction can be prepared using the stream_prepare_cb callback, COMMIT PREPARED using the commit_prepared_cb callback or aborted using the rollback_prepared_cb. One example sequence of streaming callback calls for one transaction may look like this: stream_start_cb(...); <-- start of first block of changes stream_change_cb(...); stream_change_cb(...); stream_message_cb(...); stream_change_cb(...); ... stream_change_cb(...); stream_stop_cb(...); <-- end of first block of changes stream_start_cb(...); <-- start of second block of changes stream_change_cb(...); stream_change_cb(...); stream_change_cb(...); ... stream_message_cb(...); stream_change_cb(...); stream_stop_cb(...); <-- end of second block of changes stream_commit_cb(...); <-- commit of the streamed transaction The actual sequence of callback calls may be more complicated, of course. There may be blocks for multiple streamed transactions, some of the transactions may get aborted, etc. Similar to spill-to-disk behavior, streaming is triggered when the total amount of changes decoded from the WAL (for all in-progress transactions) exceeds the limit defined by logical_decoding_work_mem setting. At that point, the largest top-level transaction (measured by the amount of memory currently used for decoded changes) is selected and streamed. However, in some cases we still have to spill to disk even if streaming is enabled because we exceed the memory threshold but still have not decoded the complete tuple e.g., only decoded toast table insert but not the main table insert. Even when streaming large transactions, the changes are still applied in commit order, preserving the same guarantees as the non-streaming mode. With the basic output plugin callbacks (eg., begin_cb, change_cb, commit_cb and message_cb) two-phase commit commands like PREPARE TRANSACTION, COMMIT PREPARED and ROLLBACK PREPARED are not decoded. While the PREPARE TRANSACTION is ignored, COMMIT PREPARED is decoded as a COMMIT and ROLLBACK PREPARED is decoded as a ROLLBACK. To support the streaming of two-phase commands, an output plugin needs to provide additional callbacks. There are multiple two-phase commit callbacks that are required, (begin_prepare_cb, prepare_cb, commit_prepared_cb, rollback_prepared_cb and stream_prepare_cb) and an optional callback (filter_prepare_cb). If the output plugin callbacks for decoding two-phase commit commands are provided, then on PREPARE TRANSACTION, the changes of that transaction are decoded, passed to the output plugin, and the prepare_cb callback is invoked. This differs from the basic decoding setup where changes are only passed to the output plugin when a transaction is committed. The start of a prepared transaction is indicated by the begin_prepare_cb callback. When a prepared transaction is rolled back using the ROLLBACK PREPARED, then the rollback_prepared_cb callback is invoked and when the prepared transaction is committed using COMMIT PREPARED, then the commit_prepared_cb callback is invoked. Optionally the output plugin can define filtering rules via filter_prepare_cb to decode only specific transaction in two phases. This can be achieved by pattern matching on the gid or via lookups using the xid. The users that want to decode prepared transactions need to be careful about below mentioned points: If the prepared transaction has locked [user] catalog tables exclusively then decoding prepare can block till the main transaction is committed. The logical replication solution that builds distributed two phase commit using this feature can deadlock if the prepared transaction has locked [user] catalog tables exclusively. To avoid this users must refrain from having locks on catalog tables (e.g. explicit LOCK command) in such transactions. See for the details.