Logging running transactions every 15 seconds.

Previously, we did this just once per checkpoint, but that could make
Hot Standby take a long time to initialize.  To avoid busying an
otherwise-idle system, we don't do this if no WAL has been written
since we did it last.

Andres Freund

src/backend/postmaster/bgwriter.c

@@ -54,9 +54,11 @@
 #include "storage/shmem.h"
 #include "storage/smgr.h"
 #include "storage/spin.h"
+#include "storage/standby.h"
 #include "utils/guc.h"
 #include "utils/memutils.h"
 #include "utils/resowner.h"
+#include "utils/timestamp.h"
 
 
 /*
@@ -70,6 +72,20 @@ int			BgWriterDelay = 200;
  */
 #define HIBERNATE_FACTOR			50
 
+/*
+ * Interval in which standby snapshots are logged into the WAL stream, in
+ * milliseconds.
+ */
+#define LOG_SNAPSHOT_INTERVAL_MS 15000
+
+/*
+ * LSN and timestamp at which we last issued a LogStandbySnapshot(), to avoid
+ * doing so too often or repeatedly if there has been no other write activity
+ * in the system.
+ */
+static TimestampTz last_snapshot_ts;
+static XLogRecPtr last_snapshot_lsn = InvalidXLogRecPtr;
+
 /*
  * Flags set by interrupt handlers for later service in the main loop.
  */
@@ -141,6 +157,12 @@ BackgroundWriterMain(void)
 	 */
 	CurrentResourceOwner = ResourceOwnerCreate(NULL, "Background Writer");
 
+	/*
+	 * We just started, assume there has been either a shutdown or
+	 * end-of-recovery snapshot.
+	 */
+	last_snapshot_ts = GetCurrentTimestamp();
+
 	/*
 	 * Create a memory context that we will do all our work in.  We do this so
 	 * that we can reset the context during error recovery and thereby avoid
@@ -275,6 +297,46 @@ BackgroundWriterMain(void)
 			smgrcloseall();
 		}
 
+		/*
+		 * Log a new xl_running_xacts every now and then so replication can get
+		 * into a consistent state faster (think of suboverflowed snapshots)
+		 * and clean up resources (locks, KnownXids*) more frequently. The
+		 * costs of this are relatively low, so doing it 4 times
+		 * (LOG_SNAPSHOT_INTERVAL_MS) a minute seems fine.
+		 *
+		 * We assume the interval for writing xl_running_xacts is
+		 * significantly bigger than BgWriterDelay, so we don't complicate the
+		 * overall timeout handling but just assume we're going to get called
+		 * often enough even if hibernation mode is active. It's not that
+		 * important that log_snap_interval_ms is met strictly. To make sure
+		 * we're not waking the disk up unneccesarily on an idle system we
+		 * check whether there has been any WAL inserted since the last time
+		 * we've logged a running xacts.
+		 *
+		 * We do this logging in the bgwriter as its the only process thats
+		 * run regularly and returns to its mainloop all the
+		 * time. E.g. Checkpointer, when active, is barely ever in its
+		 * mainloop and thus makes it hard to log regularly.
+		 */
+		if (XLogStandbyInfoActive() && !RecoveryInProgress())
+		{
+			TimestampTz timeout = 0;
+			TimestampTz now = GetCurrentTimestamp();
+			timeout = TimestampTzPlusMilliseconds(last_snapshot_ts,
+												  LOG_SNAPSHOT_INTERVAL_MS);
+
+			/*
+			 * only log if enough time has passed and some xlog record has been
+			 * inserted.
+			 */
+			if (now >= timeout &&
+				last_snapshot_lsn != GetXLogInsertRecPtr())
+			{
+				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_ts = now;
+			}
+		}
+
 		/*
 		 * Sleep until we are signaled or BgWriterDelay has elapsed.
 		 *

src/backend/storage/ipc/standby.c

@@ -42,7 +42,7 @@ static void ResolveRecoveryConflictWithVirtualXIDs(VirtualTransactionId *waitlis
 									   ProcSignalReason reason);
 static void ResolveRecoveryConflictWithLock(Oid dbOid, Oid relOid);
 static void SendRecoveryConflictWithBufferPin(ProcSignalReason reason);
-static void LogCurrentRunningXacts(RunningTransactions CurrRunningXacts);
+static XLogRecPtr LogCurrentRunningXacts(RunningTransactions CurrRunningXacts);
 static void LogAccessExclusiveLocks(int nlocks, xl_standby_lock *locks);
 
 
@@ -853,10 +853,13 @@ standby_redo(XLogRecPtr lsn, XLogRecord *record)
  * currently running xids, performed by StandbyReleaseOldLocks().
  * Zero xids should no longer be possible, but we may be replaying WAL
  * from a time when they were possible.
+ *
+ * Returns the RecPtr of the last inserted record.
  */
-void
+XLogRecPtr
 LogStandbySnapshot(void)
 {
+	XLogRecPtr recptr;
 	RunningTransactions running;
 	xl_standby_lock *locks;
 	int			nlocks;
@@ -876,9 +879,12 @@ LogStandbySnapshot(void)
 	 * record we write, because standby will open up when it sees this.
 	 */
 	running = GetRunningTransactionData();
-	LogCurrentRunningXacts(running);
+	recptr = LogCurrentRunningXacts(running);
+
 	/* GetRunningTransactionData() acquired XidGenLock, we must release it */
 	LWLockRelease(XidGenLock);
+
+	return recptr;
 }
 
 /*
@@ -889,7 +895,7 @@ LogStandbySnapshot(void)
  * is a contiguous chunk of memory and never exists fully until it is
  * assembled in WAL.
  */
-static void
+static XLogRecPtr
 LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 {
 	xl_running_xacts xlrec;
@@ -939,6 +945,19 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 			 CurrRunningXacts->oldestRunningXid,
 			 CurrRunningXacts->latestCompletedXid,
 			 CurrRunningXacts->nextXid);
+
+	/*
+	 * Ensure running_xacts information is synced to disk not too far in the
+	 * future. We don't want to stall anything though (i.e. use XLogFlush()),
+	 * so we let the wal writer do it during normal
+	 * operation. XLogSetAsyncXactLSN() conveniently will mark the LSN as
+	 * to-be-synced and nudge the WALWriter into action if sleeping. Check
+	 * XLogBackgroundFlush() for details why a record might not be flushed
+	 * without it.
+	 */
+	XLogSetAsyncXactLSN(recptr);
+
+	return recptr;
 }
 
 /*

src/include/storage/standby.h

@@ -113,6 +113,6 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern void LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(void);
 
 #endif   /* STANDBY_H */