Perform apply of large transactions by parallel workers.

Currently, for large transactions, the publisher sends the data in
multiple streams (changes divided into chunks depending upon
logical_decoding_work_mem), and then on the subscriber-side, the apply
worker writes the changes into temporary files and once it receives the
commit, it reads from those files and applies the entire transaction. To
improve the performance of such transactions, we can instead allow them to
be applied via parallel workers.

In this approach, we assign a new parallel apply worker (if available) as
soon as the xact's first stream is received and the leader apply worker
will send changes to this new worker via shared memory. The parallel apply
worker will directly apply the change instead of writing it to temporary
files. However, if the leader apply worker times out while attempting to
send a message to the parallel apply worker, it will switch to
"partial serialize" mode -  in this mode, the leader serializes all
remaining changes to a file and notifies the parallel apply workers to
read and apply them at the end of the transaction. We use a non-blocking
way to send the messages from the leader apply worker to the parallel
apply to avoid deadlocks. We keep this parallel apply assigned till the
transaction commit is received and also wait for the worker to finish at
commit. This preserves commit ordering and avoid writing to and reading
from files in most cases. We still need to spill if there is no worker
available.

This patch also extends the SUBSCRIPTION 'streaming' parameter so that the
user can control whether to apply the streaming transaction in a parallel
apply worker or spill the change to disk. The user can set the streaming
parameter to 'on/off', or 'parallel'. The parameter value 'parallel' means
the streaming will be applied via a parallel apply worker, if available.
The parameter value 'on' means the streaming transaction will be spilled
to disk. The default value is 'off' (same as current behaviour).

In addition, the patch extends the logical replication STREAM_ABORT
message so that abort_lsn and abort_time can also be sent which can be
used to update the replication origin in parallel apply worker when the
streaming transaction is aborted. Because this message extension is needed
to support parallel streaming, parallel streaming is not supported for
publications on servers < PG16.

Author: Hou Zhijie, Wang wei, Amit Kapila with design inputs from Sawada Masahiko
Reviewed-by: Sawada Masahiko, Peter Smith, Dilip Kumar, Shi yu, Kuroda Hayato, Shveta Mallik
Discussion: https://postgr.es/m/CAA4eK1+wyN6zpaHUkCLorEWNx75MG0xhMwcFhvjqm2KURZEAGw@mail.gmail.com

Author: Amit Kapila

doc/src/sgml/catalogs.sgml

@@ -7913,11 +7913,16 @@ SCRAM-SHA-256$<replaceable>&lt;iteration count&gt;</replaceable>:<replaceable>&l
 
      <row>
       <entry role="catalog_table_entry"><para role="column_definition">
-       <structfield>substream</structfield> <type>bool</type>
+       <structfield>substream</structfield> <type>char</type>
       </para>
       <para>
-       If true, the subscription will allow streaming of in-progress
-       transactions
+       Controls how to handle the streaming of in-progress transactions:
+       <literal>f</literal> = disallow streaming of in-progress transactions,
+       <literal>t</literal> = spill the changes of in-progress transactions to
+       disk and apply at once after the transaction is committed on the
+       publisher and received by the subscriber,
+       <literal>p</literal> = apply changes directly using a parallel apply
+       worker if available (same as 't' if no worker is available)
       </para></entry>
      </row>
 

doc/src/sgml/config.sgml

@@ -4968,7 +4968,8 @@ ANY <replaceable class="parameter">num_sync</replaceable> ( <replaceable class="
       <listitem>
        <para>
         Specifies maximum number of logical replication workers. This includes
-        both apply workers and table synchronization workers.
+        leader apply workers, parallel apply workers, and table synchronization
+        workers.
        </para>
        <para>
         Logical replication workers are taken from the pool defined by
@@ -5008,6 +5009,31 @@ ANY <replaceable class="parameter">num_sync</replaceable> ( <replaceable class="
       </listitem>
      </varlistentry>
 
+     <varlistentry id="guc-max-parallel-apply-workers-per-subscription" xreflabel="max_parallel_apply_workers_per_subscription">
+      <term><varname>max_parallel_apply_workers_per_subscription</varname> (<type>integer</type>)
+      <indexterm>
+       <primary><varname>max_parallel_apply_workers_per_subscription</varname> configuration parameter</primary>
+      </indexterm>
+      </term>
+      <listitem>
+       <para>
+        Maximum number of parallel apply workers per subscription. This
+        parameter controls the amount of parallelism for streaming of
+        in-progress transactions with subscription parameter
+        <literal>streaming = parallel</literal>.
+       </para>
+       <para>
+        The parallel apply workers are taken from the pool defined by
+        <varname>max_logical_replication_workers</varname>.
+       </para>
+       <para>
+        The default value is 2. This parameter can only be set in the
+        <filename>postgresql.conf</filename> file or on the server command
+        line.
+       </para>
+      </listitem>
+     </varlistentry>
+
      </variablelist>
     </sect2>
 

doc/src/sgml/logical-replication.sgml

@@ -1501,6 +1501,16 @@ CONTEXT:  processing remote data for replication origin "pg_16395" during "INSER
    might not violate any constraint.  This can easily make the subscriber
    inconsistent.
   </para>
+
+  <para>
+   When the streaming mode is <literal>parallel</literal>, the finish LSN of
+   failed transactions may not be logged. In that case, it may be necessary to
+   change the streaming mode to <literal>on</literal> or <literal>off</literal> and
+   cause the same conflicts again so the finish LSN of the failed transaction will
+   be written to the server log. For the usage of finish LSN, please refer to <link
+   linkend="sql-altersubscription"><command>ALTER SUBSCRIPTION ...
+   SKIP</command></link>.
+  </para>
  </sect1>
 
  <sect1 id="logical-replication-restrictions">
@@ -1809,8 +1819,9 @@ CONTEXT:  processing remote data for replication origin "pg_16395" during "INSER
 
    <para>
     <link linkend="guc-max-logical-replication-workers"><varname>max_logical_replication_workers</varname></link>
-    must be set to at least the number of subscriptions (for apply workers), plus
-    some reserve for the table synchronization workers.
+    must be set to at least the number of subscriptions (for leader apply
+    workers), plus some reserve for the table synchronization workers and
+    parallel apply workers.
    </para>
 
    <para>
@@ -1827,6 +1838,13 @@ CONTEXT:  processing remote data for replication origin "pg_16395" during "INSER
      subscription initialization or when new tables are added.
    </para>
 
+   <para>
+    <link linkend="guc-max-parallel-apply-workers-per-subscription"><varname>max_parallel_apply_workers_per_subscription</varname></link>
+     controls the amount of parallelism for streaming of in-progress
+     transactions with subscription parameter
+     <literal>streaming = parallel</literal>.
+   </para>
+
    <para>
     Logical replication workers are also affected by
     <link linkend="guc-wal-receiver-timeout"><varname>wal_receiver_timeout</varname></link>,

doc/src/sgml/monitoring.sgml

@@ -1858,6 +1858,11 @@ postgres   27093  0.0  0.0  30096  2752 ?        Ss   11:34   0:00 postgres: ser
       <entry><literal>advisory</literal></entry>
       <entry>Waiting to acquire an advisory user lock.</entry>
      </row>
+     <row>
+      <entry><literal>applytransaction</literal></entry>
+      <entry>Waiting to acquire a lock on a remote transaction being applied
+      by a logical replication subscriber.</entry>
+     </row>
      <row>
       <entry><literal>extend</literal></entry>
       <entry>Waiting to extend a relation.</entry>

doc/src/sgml/protocol.sgml

@@ -3103,7 +3103,7 @@ psql "dbname=postgres replication=database" -c "IDENTIFY_SYSTEM;"
      <listitem>
       <para>
        Protocol version. Currently versions <literal>1</literal>, <literal>2</literal>,
-       and <literal>3</literal> are supported.
+       <literal>3</literal>, and <literal>4</literal> are supported.
       </para>
       <para>
        Version <literal>2</literal> is supported only for server version 14
@@ -3113,6 +3113,11 @@ psql "dbname=postgres replication=database" -c "IDENTIFY_SYSTEM;"
        Version <literal>3</literal> is supported only for server version 15
        and above, and it allows streaming of two-phase commits.
       </para>
+      <para>
+       Version <literal>4</literal> is supported only for server version 16
+       and above, and it allows streams of large in-progress transactions to
+       be applied in parallel.
+      </para>
      </listitem>
     </varlistentry>
 
@@ -6883,6 +6888,28 @@ psql "dbname=postgres replication=database" -c "IDENTIFY_SYSTEM;"
         </para>
        </listitem>
       </varlistentry>
+
+      <varlistentry>
+       <term>Int64 (XLogRecPtr)</term>
+       <listitem>
+        <para>
+         The LSN of the abort. This field is available since protocol version
+         4.
+        </para>
+       </listitem>
+      </varlistentry>
+
+      <varlistentry>
+       <term>Int64 (TimestampTz)</term>
+       <listitem>
+        <para>
+         Abort timestamp of the transaction. The value is in number
+         of microseconds since PostgreSQL epoch (2000-01-01). This field is
+         available since protocol version 4.
+        </para>
+       </listitem>
+      </varlistentry>
+
      </variablelist>
     </listitem>
    </varlistentry>

doc/src/sgml/ref/create_subscription.sgml

@@ -228,13 +228,29 @@ CREATE SUBSCRIPTION <replaceable class="parameter">subscription_name</replaceabl
        </varlistentry>
 
        <varlistentry>
-        <term><literal>streaming</literal> (<type>boolean</type>)</term>
+        <term><literal>streaming</literal> (<type>enum</type>)</term>
         <listitem>
          <para>
           Specifies whether to enable streaming of in-progress transactions
-          for this subscription.  By default, all transactions
-          are fully decoded on the publisher and only then sent to the
-          subscriber as a whole.
+          for this subscription.  The default value is <literal>off</literal>,
+          meaning all transactions are fully decoded on the publisher and only
+          then sent to the subscriber as a whole.
+         </para>
+
+         <para>
+          If set to <literal>on</literal>, the incoming changes are written to
+          temporary files and then applied only after the transaction is
+          committed on the publisher and received by the subscriber.
+         </para>
+
+         <para>
+          If set to <literal>parallel</literal>, incoming changes are directly
+          applied via one of the parallel apply workers, if available. If no
+          parallel apply worker is free to handle streaming transactions then
+          the changes are written to temporary files and applied after the
+          transaction is committed. Note that if an error happens in a
+          parallel apply worker, the finish LSN of the remote transaction
+          might not be reported in the server log.
          </para>
         </listitem>
        </varlistentry>

doc/src/sgml/system-views.sgml

@@ -1379,8 +1379,9 @@
        <literal>virtualxid</literal>,
        <literal>spectoken</literal>,
        <literal>object</literal>,
-       <literal>userlock</literal>, or
-       <literal>advisory</literal>.
+       <literal>userlock</literal>,
+       <literal>advisory</literal>, or
+       <literal>applytransaction</literal>.
        (See also <xref linkend="wait-event-lock-table"/>.)
       </para></entry>
      </row>
@@ -1594,6 +1595,15 @@
    so the <structfield>database</structfield> column is meaningful for an advisory lock.
   </para>
 
+  <para>
+   Apply transaction locks are used in parallel mode to apply the transaction
+   in logical replication. The remote transaction id is displayed in the
+   <structfield>transactionid</structfield> column. The <structfield>objsubid</structfield>
+   displays the lock subtype which is 0 for the lock used to synchronize the
+   set of changes, and 1 for the lock used to wait for the transaction to
+   finish to ensure commit order.
+  </para>
+
   <para>
    <structname>pg_locks</structname> provides a global view of all locks
    in the database cluster, not only those relevant to the current database.

src/backend/access/transam/xact.c

@@ -1713,6 +1713,7 @@ RecordTransactionAbort(bool isSubXact)
 	int			nchildren;
 	TransactionId *children;
 	TimestampTz xact_time;
+	bool		replorigin;
 
 	/*
 	 * If we haven't been assigned an XID, nobody will care whether we aborted
@@ -1743,6 +1744,13 @@ RecordTransactionAbort(bool isSubXact)
 		elog(PANIC, "cannot abort transaction %u, it was already committed",
 			 xid);
 
+	/*
+	 * Are we using the replication origins feature?  Or, in other words, are
+	 * we replaying remote actions?
+	 */
+	replorigin = (replorigin_session_origin != InvalidRepOriginId &&
+				  replorigin_session_origin != DoNotReplicateId);
+
 	/* Fetch the data we need for the abort record */
 	nrels = smgrGetPendingDeletes(false, &rels);
 	nchildren = xactGetCommittedChildren(&children);
@@ -1766,6 +1774,11 @@ RecordTransactionAbort(bool isSubXact)
 					   MyXactFlags, InvalidTransactionId,
 					   NULL);
 
+	if (replorigin)
+		/* Move LSNs forward for this replication origin */
+		replorigin_session_advance(replorigin_session_origin_lsn,
+								   XactLastRecEnd);
+
 	/*
 	 * Report the latest async abort LSN, so that the WAL writer knows to
 	 * flush this abort. There's nothing to be gained by delaying this, since
@@ -5873,11 +5886,10 @@ XactLogAbortRecord(TimestampTz abort_time,
 	}
 
 	/*
-	 * Dump transaction origin information only for abort prepared. We need
-	 * this during recovery to update the replication origin progress.
+	 * Dump transaction origin information. We need this during recovery to
+	 * update the replication origin progress.
 	 */
-	if ((replorigin_session_origin != InvalidRepOriginId) &&
-		TransactionIdIsValid(twophase_xid))
+	if (replorigin_session_origin != InvalidRepOriginId)
 	{
 		xl_xinfo.xinfo |= XACT_XINFO_HAS_ORIGIN;
 
@@ -5934,8 +5946,8 @@ XactLogAbortRecord(TimestampTz abort_time,
 	if (xl_xinfo.xinfo & XACT_XINFO_HAS_ORIGIN)
 		XLogRegisterData((char *) (&xl_origin), sizeof(xl_xact_origin));
 
-	if (TransactionIdIsValid(twophase_xid))
-		XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
+	/* Include the replication origin */
+	XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
 
 	return XLogInsert(RM_XACT_ID, info);
 }

src/backend/commands/subscriptioncmds.c

@@ -85,7 +85,7 @@ typedef struct SubOpts
 	bool		copy_data;
 	bool		refresh;
 	bool		binary;
-	bool		streaming;
+	char		streaming;
 	bool		twophase;
 	bool		disableonerr;
 	char	   *origin;
@@ -139,7 +139,7 @@ parse_subscription_options(ParseState *pstate, List *stmt_options,
 	if (IsSet(supported_opts, SUBOPT_BINARY))
 		opts->binary = false;
 	if (IsSet(supported_opts, SUBOPT_STREAMING))
-		opts->streaming = false;
+		opts->streaming = LOGICALREP_STREAM_OFF;
 	if (IsSet(supported_opts, SUBOPT_TWOPHASE_COMMIT))
 		opts->twophase = false;
 	if (IsSet(supported_opts, SUBOPT_DISABLE_ON_ERR))
@@ -242,7 +242,7 @@ parse_subscription_options(ParseState *pstate, List *stmt_options,
 				errorConflictingDefElem(defel, pstate);
 
 			opts->specified_opts |= SUBOPT_STREAMING;
-			opts->streaming = defGetBoolean(defel);
+			opts->streaming = defGetStreamingMode(defel);
 		}
 		else if (strcmp(defel->defname, "two_phase") == 0)
 		{
@@ -630,7 +630,7 @@ CreateSubscription(ParseState *pstate, CreateSubscriptionStmt *stmt,
 	values[Anum_pg_subscription_subowner - 1] = ObjectIdGetDatum(owner);
 	values[Anum_pg_subscription_subenabled - 1] = BoolGetDatum(opts.enabled);
 	values[Anum_pg_subscription_subbinary - 1] = BoolGetDatum(opts.binary);
-	values[Anum_pg_subscription_substream - 1] = BoolGetDatum(opts.streaming);
+	values[Anum_pg_subscription_substream - 1] = CharGetDatum(opts.streaming);
 	values[Anum_pg_subscription_subtwophasestate - 1] =
 		CharGetDatum(opts.twophase ?
 					 LOGICALREP_TWOPHASE_STATE_PENDING :
@@ -1099,7 +1099,7 @@ AlterSubscription(ParseState *pstate, AlterSubscriptionStmt *stmt,
 				if (IsSet(opts.specified_opts, SUBOPT_STREAMING))
 				{
 					values[Anum_pg_subscription_substream - 1] =
-						BoolGetDatum(opts.streaming);
+						CharGetDatum(opts.streaming);
 					replaces[Anum_pg_subscription_substream - 1] = true;
 				}
 
@@ -2128,3 +2128,60 @@ merge_publications(List *oldpublist, List *newpublist, bool addpub, const char *
 
 	return oldpublist;
 }
+
+/*
+ * Extract the streaming mode value from a DefElem.  This is like
+ * defGetBoolean() but also accepts the special value of "parallel".
+ */
+char
+defGetStreamingMode(DefElem *def)
+{
+	/*
+	 * If no parameter value given, assume "true" is meant.
+	 */
+	if (!def->arg)
+		return LOGICALREP_STREAM_ON;
+
+	/*
+	 * Allow 0, 1, "false", "true", "off", "on" or "parallel".
+	 */
+	switch (nodeTag(def->arg))
+	{
+		case T_Integer:
+			switch (intVal(def->arg))
+			{
+				case 0:
+					return LOGICALREP_STREAM_OFF;
+				case 1:
+					return LOGICALREP_STREAM_ON;
+				default:
+					/* otherwise, error out below */
+					break;
+			}
+			break;
+		default:
+			{
+				char	   *sval = defGetString(def);
+
+				/*
+				 * The set of strings accepted here should match up with the
+				 * grammar's opt_boolean_or_string production.
+				 */
+				if (pg_strcasecmp(sval, "false") == 0 ||
+					pg_strcasecmp(sval, "off") == 0)
+					return LOGICALREP_STREAM_OFF;
+				if (pg_strcasecmp(sval, "true") == 0 ||
+					pg_strcasecmp(sval, "on") == 0)
+					return LOGICALREP_STREAM_ON;
+				if (pg_strcasecmp(sval, "parallel") == 0)
+					return LOGICALREP_STREAM_PARALLEL;
+			}
+			break;
+	}
+
+	ereport(ERROR,
+			(errcode(ERRCODE_SYNTAX_ERROR),
+			 errmsg("%s requires a Boolean value or \"parallel\"",
+					def->defname)));
+	return LOGICALREP_STREAM_OFF;	/* keep compiler quiet */
+}