Hot Standby feedback for avoidance of cleanup conflicts on standby.

Standby optionally sends back information about oldestXmin of queries
which is then checked and applied to the WALSender's proc->xmin.
GetOldestXmin() is modified slightly to agree with GetSnapshotData(),
so that all backends on primary include WALSender within their snapshots.
Note this does nothing to change the snapshot xmin on either master or
standby. Feedback piggybacks on the standby reply message.
vacuum_defer_cleanup_age is no longer used on standby, though parameter
still exists on primary, since some use cases still exist.

Simon Riggs, review comments from Fujii Masao, Heikki Linnakangas, Robert Haas

Author: simonat2ndQuadrant

doc/src/sgml/config.sgml

@@ -2006,6 +2006,10 @@ SET ENABLE_SEQSCAN TO OFF;
         This parameter can only be set in the <filename>postgresql.conf</>
         file or on the server command line.
        </para>
+       <para>
+        You should also consider setting <varname>hot_standby_feedback</>
+        as an alternative to using this parameter.
+       </para>
       </listitem>
      </varlistentry>
      </variablelist>
@@ -2121,6 +2125,22 @@ SET ENABLE_SEQSCAN TO OFF;
       </listitem>
      </varlistentry>
 
+     <varlistentry id="guc-hot-standby-feedback" xreflabel="hot_standby">
+      <term><varname>hot_standby_feedback</varname> (<type>boolean</type>)</term>
+      <indexterm>
+       <primary><varname>hot_standby_feedback</> configuration parameter</primary>
+      </indexterm>
+      <listitem>
+       <para>
+        Specifies whether or not a hot standby will send feedback to the primary
+        about queries currently executing on the standby. This parameter can
+        be used to eliminate query cancels caused by cleanup records, though
+        it can cause database bloat on the primary for some workloads.
+        The default value is <literal>off</literal>.
+       </para>
+      </listitem>
+     </varlistentry>
+
      </variablelist>
     </sect2>
    </sect1>

doc/src/sgml/high-availability.sgml

@@ -1483,23 +1483,6 @@ if (!triggered)
     on the primary, if it delays application of WAL records.
    </para>
 
-   <para>
-    The most common reason for conflict between standby queries and WAL replay
-    is <quote>early cleanup</>.  Normally, <productname>PostgreSQL</> allows
-    cleanup of old row versions when there are no transactions that need to
-    see them to ensure correct visibility of data according to MVCC rules.
-    However, this rule can only be applied for transactions executing on the
-    master.  So it is possible that cleanup on the master will remove row
-    versions that are still visible to a transaction on the standby.
-   </para>
-
-   <para>
-    Experienced users should note that both row version cleanup and row version
-    freezing will potentially conflict with standby queries. Running a manual
-    <command>VACUUM FREEZE</> is likely to cause conflicts even on tables with
-    no updated or deleted rows.
-   </para>
-
    <para>
     Once the delay specified by <varname>max_standby_archive_delay</> or
     <varname>max_standby_streaming_delay</> has been exceeded, conflicting
@@ -1526,6 +1509,23 @@ if (!triggered)
     as a result of being unable to keep up with a heavy update load.
    </para>
 
+   <para>
+    The most common reason for conflict between standby queries and WAL replay
+    is <quote>early cleanup</>.  Normally, <productname>PostgreSQL</> allows
+    cleanup of old row versions when there are no transactions that need to
+    see them to ensure correct visibility of data according to MVCC rules.
+    However, this rule can only be applied for transactions executing on the
+    master.  So it is possible that cleanup on the master will remove row
+    versions that are still visible to a transaction on the standby.
+   </para>
+
+   <para>
+    Experienced users should note that both row version cleanup and row version
+    freezing will potentially conflict with standby queries. Running a manual
+    <command>VACUUM FREEZE</> is likely to cause conflicts even on tables with
+    no updated or deleted rows.
+   </para>
+
    <para>
     Users should be clear that tables that are regularly and heavily updated
     on the primary server will quickly cause cancellation of longer running
@@ -1537,12 +1537,10 @@ if (!triggered)
 
    <para>
     Remedial possibilities exist if the number of standby-query cancellations
-    is found to be unacceptable.  The first option is to connect to the
-    primary server and keep a query active for as long as needed to
-    run queries on the standby. This prevents <command>VACUUM</> from removing
-    recently-dead rows and so cleanup conflicts do not occur.
-    This could be done using <xref linkend="dblink"> and
-    <function>pg_sleep()</>, or via other mechanisms. If you do this, you
+    is found to be unacceptable.  The first option is to set the parameter
+    <varname>hot_standby_feedback</>, which prevents <command>VACUUM</> from
+    removing recently-dead rows and so cleanup conflicts do not occur.
+    If you do this, you
     should note that this will delay cleanup of dead rows on the primary,
     which may result in undesirable table bloat. However, the cleanup
     situation will be no worse than if the standby queries were running

src/backend/access/transam/xlog.c

@@ -158,6 +158,11 @@ static XLogRecPtr LastRec;
  * known, need to check the shared state".
  */
 static bool LocalRecoveryInProgress = true;
+/*
+ * Local copy of SharedHotStandbyActive variable. False actually means "not
+ * known, need to check the shared state".
+ */
+static bool LocalHotStandbyActive = false;
 
 /*
  * Local state for XLogInsertAllowed():
@@ -405,6 +410,12 @@ typedef struct XLogCtlData
 	 */
 	bool		SharedRecoveryInProgress;
 
+	/*
+	 * SharedHotStandbyActive indicates if we're still in crash or archive
+	 * recovery.  Protected by info_lck.
+	 */
+	bool		SharedHotStandbyActive;
+
 	/*
 	 * recoveryWakeupLatch is used to wake up the startup process to
 	 * continue WAL replay, if it is waiting for WAL to arrive or failover
@@ -4917,6 +4928,7 @@ XLOGShmemInit(void)
 	 */
 	XLogCtl->XLogCacheBlck = XLOGbuffers - 1;
 	XLogCtl->SharedRecoveryInProgress = true;
+	XLogCtl->SharedHotStandbyActive = false;
 	XLogCtl->Insert.currpage = (XLogPageHeader) (XLogCtl->pages);
 	SpinLockInit(&XLogCtl->info_lck);
 	InitSharedLatch(&XLogCtl->recoveryWakeupLatch);
@@ -6790,8 +6802,6 @@ StartupXLOG(void)
 static void
 CheckRecoveryConsistency(void)
 {
-	static bool backendsAllowed = false;
-
 	/*
 	 * Have we passed our safe starting point?
 	 */
@@ -6811,11 +6821,19 @@ CheckRecoveryConsistency(void)
 	 * enabling connections.
 	 */
 	if (standbyState == STANDBY_SNAPSHOT_READY &&
-		!backendsAllowed &&
+		!LocalHotStandbyActive &&
 		reachedMinRecoveryPoint &&
 		IsUnderPostmaster)
 	{
-		backendsAllowed = true;
+		/* use volatile pointer to prevent code rearrangement */
+		volatile XLogCtlData *xlogctl = XLogCtl;
+
+		SpinLockAcquire(&xlogctl->info_lck);
+		xlogctl->SharedHotStandbyActive = true;
+		SpinLockRelease(&xlogctl->info_lck);
+
+		LocalHotStandbyActive = true;
+
 		SendPostmasterSignal(PMSIGNAL_BEGIN_HOT_STANDBY);
 	}
 }
@@ -6862,6 +6880,38 @@ RecoveryInProgress(void)
 	}
 }
 
+/*
+ * Is HotStandby active yet? This is only important in special backends
+ * since normal backends won't ever be able to connect until this returns
+ * true. Postmaster knows this by way of signal, not via shared memory.
+ *
+ * Unlike testing standbyState, this works in any process that's connected to
+ * shared memory.
+ */
+bool
+HotStandbyActive(void)
+{
+	/*
+	 * We check shared state each time only until Hot Standby is active. We
+	 * can't de-activate Hot Standby, so there's no need to keep checking after
+	 * the shared variable has once been seen true.
+	 */
+	if (LocalHotStandbyActive)
+		return true;
+	else
+	{
+		/* use volatile pointer to prevent code rearrangement */
+		volatile XLogCtlData *xlogctl = XLogCtl;
+
+		/* spinlock is essential on machines with weak memory ordering! */
+		SpinLockAcquire(&xlogctl->info_lck);
+		LocalHotStandbyActive = xlogctl->SharedHotStandbyActive;
+		SpinLockRelease(&xlogctl->info_lck);
+
+		return LocalHotStandbyActive;
+	}
+}
+
 /*
  * Is this process allowed to insert new WAL records?
  *

src/backend/replication/walreceiver.c

@@ -38,13 +38,15 @@
 #include <signal.h>
 #include <unistd.h>
 
+#include "access/transam.h"
 #include "access/xlog_internal.h"
 #include "libpq/pqsignal.h"
 #include "miscadmin.h"
 #include "replication/walprotocol.h"
 #include "replication/walreceiver.h"
 #include "storage/ipc.h"
 #include "storage/pmsignal.h"
+#include "storage/procarray.h"
 #include "utils/builtins.h"
 #include "utils/guc.h"
 #include "utils/memutils.h"
@@ -56,6 +58,7 @@ bool		am_walreceiver;
 
 /* GUC variable */
 int			wal_receiver_status_interval;
+bool		hot_standby_feedback;
 
 /* libpqreceiver hooks to these when loaded */
 walrcv_connect_type walrcv_connect = NULL;
@@ -610,16 +613,43 @@ XLogWalRcvSendReply(void)
 			wal_receiver_status_interval * 1000))
 		return;
 
-	/* Construct a new message. */
+	/* Construct a new message */
 	reply_message.write = LogstreamResult.Write;
 	reply_message.flush = LogstreamResult.Flush;
 	reply_message.apply = GetXLogReplayRecPtr();
 	reply_message.sendTime = now;
 
-	elog(DEBUG2, "sending write %X/%X flush %X/%X apply %X/%X",
+	/*
+	 * Get the OldestXmin and its associated epoch
+	 */
+	if (hot_standby_feedback && HotStandbyActive())
+	{
+		TransactionId	nextXid;
+		uint32			nextEpoch;
+
+		reply_message.xmin = GetOldestXmin(true, false);
+
+		/*
+		 * Get epoch and adjust if nextXid and oldestXmin are different
+		 * sides of the epoch boundary.
+		 */
+		GetNextXidAndEpoch(&nextXid, &nextEpoch);
+		if (nextXid < reply_message.xmin)
+			nextEpoch--;
+		reply_message.epoch = nextEpoch;
+	}
+	else
+	{
+		reply_message.xmin = InvalidTransactionId;
+		reply_message.epoch = 0;
+	}
+
+	elog(DEBUG2, "sending write %X/%X flush %X/%X apply %X/%X xmin %u epoch %u",
 				 reply_message.write.xlogid, reply_message.write.xrecoff,
 				 reply_message.flush.xlogid, reply_message.flush.xrecoff,
-				 reply_message.apply.xlogid, reply_message.apply.xrecoff);
+				 reply_message.apply.xlogid, reply_message.apply.xrecoff,
+				 reply_message.xmin,
+				 reply_message.epoch);
 
 	/* Prepend with the message type and send it. */
 	buf[0] = 'r';

src/backend/replication/walsender.c

@@ -53,6 +53,7 @@
 #include "storage/ipc.h"
 #include "storage/pmsignal.h"
 #include "storage/proc.h"
+#include "storage/procarray.h"
 #include "tcop/tcopprot.h"
 #include "utils/builtins.h"
 #include "utils/guc.h"
@@ -502,6 +503,7 @@ ProcessStandbyReplyMessage(void)
 {
 	StandbyReplyMessage	reply;
 	char msgtype;
+	TransactionId newxmin = InvalidTransactionId;
 
 	resetStringInfo(&reply_message);
 
@@ -531,10 +533,12 @@ ProcessStandbyReplyMessage(void)
 
 	pq_copymsgbytes(&reply_message, (char *) &reply, sizeof(StandbyReplyMessage));
 
-	elog(DEBUG2, "write %X/%X flush %X/%X apply %X/%X ",
+	elog(DEBUG2, "write %X/%X flush %X/%X apply %X/%X xmin %u epoch %u",
 		 reply.write.xlogid, reply.write.xrecoff,
 		 reply.flush.xlogid, reply.flush.xrecoff,
-		 reply.apply.xlogid, reply.apply.xrecoff);
+		 reply.apply.xlogid, reply.apply.xrecoff,
+		 reply.xmin,
+		 reply.epoch);
 
 	/*
 	 * Update shared state for this WalSender process
@@ -550,6 +554,69 @@ ProcessStandbyReplyMessage(void)
 		walsnd->apply = reply.apply;
 		SpinLockRelease(&walsnd->mutex);
 	}
+
+	/*
+	 * Update the WalSender's proc xmin to allow it to be visible
+	 * to snapshots. This will hold back the removal of dead rows
+	 * and thereby prevent the generation of cleanup conflicts
+	 * on the standby server.
+	 */
+	if (TransactionIdIsValid(reply.xmin))
+	{
+		TransactionId	nextXid;
+		uint32			nextEpoch;
+		bool			epochOK;
+
+		GetNextXidAndEpoch(&nextXid, &nextEpoch);
+
+		/*
+		 * Epoch of oldestXmin should be same as standby or
+		 * if the counter has wrapped, then one less than reply.
+		 */
+		if (reply.xmin <= nextXid)
+		{
+			if (reply.epoch == nextEpoch)
+				epochOK = true;
+		}
+		else
+		{
+			if (nextEpoch > 0 && reply.epoch == nextEpoch - 1)
+				epochOK = true;
+		}
+
+		/*
+		 * Feedback from standby must not go backwards, nor should it go
+		 * forwards further than our most recent xid.
+		 */
+		if (epochOK && TransactionIdPrecedesOrEquals(reply.xmin, nextXid))
+		{
+			if (!TransactionIdIsValid(MyProc->xmin))
+			{
+				TransactionId oldestXmin = GetOldestXmin(true, true);
+				if (TransactionIdPrecedes(oldestXmin, reply.xmin))
+					newxmin = reply.xmin;
+				else
+					newxmin = oldestXmin;
+			}
+			else
+			{
+				if (TransactionIdPrecedes(MyProc->xmin, reply.xmin))
+					newxmin = reply.xmin;
+				else
+					newxmin = MyProc->xmin; /* stay the same */
+			}
+		}
+	}
+
+	/*
+	 * Grab the ProcArrayLock to set xmin, or invalidate for bad reply
+	 */
+	if (MyProc->xmin != newxmin)
+	{
+		LWLockAcquire(ProcArrayLock, LW_SHARED);
+		MyProc->xmin = newxmin;
+		LWLockRelease(ProcArrayLock);
+	}
 }
 
 /* Main loop of walsender process */