agora inbox for pgsql-hackers@postgresql.org  

help / color / mirror / Atom feed

[PATCH 4/5] Use only xlogreader.c:XLogRead()
8+ messages / 1 participants
[nested] [flat]

* [PATCH 4/5] Use only xlogreader.c:XLogRead()
@ 2019-07-09 09:54  Antonin Houska <ah@cybertec.at>
  0 siblings, 0 replies; 8+ messages in thread

From: Antonin Houska @ 2019-07-09 09:54 UTC (permalink / raw)

The implementations in xlogutils.c and walsender.c are just renamed now, to be
removed by the following diff.
---
 src/backend/access/transam/xlogreader.c | 128 ++++++++++++++++++++++++++++++++
 src/backend/access/transam/xlogutils.c  |  40 ++++++++--
 src/backend/replication/walsender.c     | 125 ++++++++++++++++++++++++++++++-
 src/bin/pg_waldump/pg_waldump.c         |  59 ++++++++++++++-
 src/include/access/xlogreader.h         |  19 +++++
 5 files changed, 359 insertions(+), 12 deletions(-)

diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c
index bbe383240d..3f8db2f25d 100644
--- a/src/backend/access/transam/xlogreader.c
+++ b/src/backend/access/transam/xlogreader.c
@@ -17,6 +17,8 @@
  */
 #include "postgres.h"
 
+#include <unistd.h>
+
 #include "access/transam.h"
 #include "access/xlogrecord.h"
 #include "access/xlog_internal.h"
@@ -26,6 +28,7 @@
 #include "replication/origin.h"
 
 #ifndef FRONTEND
+#include "pgstat.h"
 #include "utils/memutils.h"
 #endif
 
@@ -1009,7 +1012,132 @@ XLogSegmentInit(XLogSegment *seg, int size)
 	seg->tli = 0;
 	seg->dir = NULL;
 	seg->size = size;
+	seg->last_req = 0;
+}
+
+/*
+ * Read 'count' bytes from WAL into 'buf', starting at location 'startptr'. If
+ * tli is passed, get the data from timeline *tli. 'pos' is the current
+ * position in the XLOG file and openSegment is a callback that opens the next
+ * segment for reading.
+ *
+ * Returns true if the call succeeded, false if it failed. Caller should check
+ * errno in the case of failure. seg->last_req might also be useful for error
+ * messages.
+ *
+ * XXX probably this should be improved to suck data directly from the
+ * WAL buffers when possible.
+ */
+bool
+XLogRead(char *buf, XLogRecPtr startptr, Size count,
+		 TimeLineID *tli, XLogSegment *seg, XLogOpenSegment openSegment)
+{
+	char	   *p;
+	XLogRecPtr	recptr;
+	Size		nbytes;
+
+	p = buf;
+	recptr = startptr;
+	nbytes = count;
+
+	while (nbytes > 0)
+	{
+		int			readbytes;
+
+		seg->off = XLogSegmentOffset(recptr, seg->size);
+
+		if (seg->file < 0 ||
+			!XLByteInSeg(recptr, seg->num, seg->size) ||
+			(tli != NULL && *tli != seg->tli))
+		{
+			XLogSegNo	nextSegNo;
+
+			/* Switch to another logfile segment */
+			if (seg->file >= 0)
+				close(seg->file);
+
+			XLByteToSeg(recptr, nextSegNo, seg->size);
+
+			/* Open the next segment in the caller's way. */
+			openSegment(nextSegNo, tli, seg);
+
+			/*
+			 * If the function is called by the XLOG reader, the reader will
+			 * eventually set both "num" and "off". However we need to care
+			 * about them too because the function can also be used directly,
+			 * see walsender.c.
+			 */
+			seg->num = nextSegNo;
+			seg->off = 0;
+		}
+
+		/* How many bytes are within this segment? */
+		if (nbytes > (seg->size - seg->off))
+			seg->last_req = seg->size - seg->off;
+		else
+			seg->last_req = nbytes;
+
+#ifndef FRONTEND
+		pgstat_report_wait_start(WAIT_EVENT_WAL_READ);
+#endif
+
+		/*
+		 * Failure to read the data does not necessarily imply non-zero errno.
+		 * Set it to zero so that caller can distinguish the failure that does
+		 * not affect errno.
+		 */
+		errno = 0;
+
+		readbytes = pg_pread(seg->file, p, seg->last_req, seg->off);
+
+#ifndef FRONTEND
+		pgstat_report_wait_end();
+#endif
+
+		if (readbytes <= 0)
+			return false;
+
+		/* Update state for read */
+		recptr += readbytes;
+		nbytes -= readbytes;
+		p += readbytes;
+
+		/*
+		 * If the function is called by the XLOG reader, the reader will
+		 * eventually set this field. However we need to care about it too
+		 * because the function can also be used directly (see walsender.c).
+		 */
+		seg->off += readbytes;
+	}
+
+	return true;
+}
+
+#ifndef FRONTEND
+/*
+ * Backend-specific code to handle errors encountered by XLogRead().
+ */
+void
+XLogReadProcessError(XLogSegment *seg)
+{
+	if (errno != 0)
+	{
+		ereport(ERROR,
+				(errcode_for_file_access(),
+				 errmsg("could not read from log segment %s, offset %u, length %zu: %m",
+						XLogFileNameP(seg->tli, seg->num), seg->off,
+						(Size) seg->last_req)));
+	}
+	else
+	{
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_CORRUPTED),
+				 errmsg("could not read from log segment %s, offset %u: length %zu",
+						XLogFileNameP(seg->tli, seg->num), seg->off,
+						(Size) seg->last_req)));
+	}
 }
+#endif
 
 /* ----------------------------------------
  * Functions for decoding the data and block references in a record.
diff --git a/src/backend/access/transam/xlogutils.c b/src/backend/access/transam/xlogutils.c
index 836c2e2927..899bf1b551 100644
--- a/src/backend/access/transam/xlogutils.c
+++ b/src/backend/access/transam/xlogutils.c
@@ -653,8 +653,8 @@ XLogTruncateRelation(RelFileNode rnode, ForkNumber forkNum,
  * frontend).  Probably these should be merged at some point.
  */
 static void
-XLogRead(char *buf, int segsize, TimeLineID tli, XLogRecPtr startptr,
-		 Size count)
+XLogReadOld(char *buf, int segsize, TimeLineID tli, XLogRecPtr startptr,
+			Size count)
 {
 	char	   *p;
 	XLogRecPtr	recptr;
@@ -896,6 +896,35 @@ XLogReadDetermineTimeline(XLogReaderState *state, XLogRecPtr wantPage, uint32 wa
 	}
 }
 
+/*
+ * Callback for XLogRead() to open the next segment.
+ */
+static void
+read_local_xlog_page_open_segment(XLogSegNo nextSegNo, TimeLineID *tli,
+								  XLogSegment *seg)
+{
+	char		path[MAXPGPATH];
+
+	XLogFilePath(path, *tli, nextSegNo, seg->size);
+	seg->file = BasicOpenFile(path, O_RDONLY | PG_BINARY);
+
+	if (seg->file < 0)
+	{
+		if (errno == ENOENT)
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("requested WAL segment %s has already been removed",
+							path)));
+		else
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("could not open file \"%s\": %m",
+							path)));
+	}
+
+	seg->tli = *tli;
+}
+
 /*
  * read_page callback for reading local xlog files
  *
@@ -1022,10 +1051,9 @@ read_local_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr,
 	 * as 'count', read the whole page anyway. It's guaranteed to be
 	 * zero-padded up to the page boundary if it's incomplete.
 	 */
-	XLogRead(cur_page, state->seg.size, state->seg.tli, targetPagePtr,
-			 XLOG_BLCKSZ);
-	state->seg.tli = pageTLI;
-
+	if (!XLogRead(cur_page, targetPagePtr, XLOG_BLCKSZ, &pageTLI,
+				  &state->seg, read_local_xlog_page_open_segment))
+		XLogReadProcessError(&state->seg);
 	/* number of valid bytes in the buffer */
 	return count;
 }
diff --git a/src/backend/replication/walsender.c b/src/backend/replication/walsender.c
index 3b5fe2cb94..c64a4bf9f8 100644
--- a/src/backend/replication/walsender.c
+++ b/src/backend/replication/walsender.c
@@ -247,7 +247,9 @@ static void LagTrackerWrite(XLogRecPtr lsn, TimestampTz local_flush_time);
 static TimeOffset LagTrackerRead(int head, XLogRecPtr lsn, TimestampTz now);
 static bool TransactionIdInRecentPast(TransactionId xid, uint32 epoch);
 
-static void XLogRead(char *buf, XLogRecPtr startptr, Size count);
+static void WalSndOpenSegment(XLogSegNo nextSegNo, TimeLineID *tli,
+							  XLogSegment *seg);
+static void XLogReadOld(char *buf, XLogRecPtr startptr, Size count);
 
 
 /* Initialize walsender process before entering the main command loop */
@@ -782,7 +784,9 @@ logical_read_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr, int req
 		count = flushptr - targetPagePtr;	/* part of the page available */
 
 	/* now actually read the data, we know it's there */
-	XLogRead(cur_page, targetPagePtr, XLOG_BLCKSZ);
+	if (!XLogRead(cur_page, targetPagePtr, XLOG_BLCKSZ, NULL, sendSeg,
+				  WalSndOpenSegment))
+		XLogReadProcessError(sendSeg);
 
 	return count;
 }
@@ -2359,7 +2363,7 @@ WalSndKill(int code, Datum arg)
  * more than one.
  */
 static void
-XLogRead(char *buf, XLogRecPtr startptr, Size count)
+XLogReadOld(char *buf, XLogRecPtr startptr, Size count)
 {
 	char	   *p;
 	XLogRecPtr	recptr;
@@ -2532,6 +2536,76 @@ retry:
 	}
 }
 
+/*
+ * Callback for XLogRead() to open the next segment.
+ */
+void
+WalSndOpenSegment(XLogSegNo nextSegNo, TimeLineID *tli, XLogSegment *seg)
+{
+	char		path[MAXPGPATH];
+
+	/*
+	 * The timeline is determined below, caller should not do anything about
+	 * it.
+	 */
+	Assert(tli == NULL);
+
+	/*-------
+	 * When reading from a historic timeline, and there is a timeline switch
+	 * within this segment, read from the WAL segment belonging to the new
+	 * timeline.
+	 *
+	 * For example, imagine that this server is currently on timeline 5, and
+	 * we're streaming timeline 4. The switch from timeline 4 to 5 happened at
+	 * 0/13002088. In pg_wal, we have these files:
+	 *
+	 * ...
+	 * 000000040000000000000012
+	 * 000000040000000000000013
+	 * 000000050000000000000013
+	 * 000000050000000000000014
+	 * ...
+	 *
+	 * In this situation, when requested to send the WAL from segment 0x13, on
+	 * timeline 4, we read the WAL from file 000000050000000000000013. Archive
+	 * recovery prefers files from newer timelines, so if the segment was
+	 * restored from the archive on this server, the file belonging to the old
+	 * timeline, 000000040000000000000013, might not exist. Their contents are
+	 * equal up to the switchpoint, because at a timeline switch, the used
+	 * portion of the old segment is copied to the new file.  -------
+	 */
+	seg->tli = sendTimeLine;
+	if (sendTimeLineIsHistoric)
+	{
+		XLogSegNo	endSegNo;
+
+		XLByteToSeg(sendTimeLineValidUpto, endSegNo, seg->size);
+		if (seg->num == endSegNo)
+			seg->tli = sendTimeLineNextTLI;
+	}
+
+	XLogFilePath(path, seg->tli, nextSegNo, seg->size);
+	seg->file = BasicOpenFile(path, O_RDONLY | PG_BINARY);
+
+	if (seg->file < 0)
+	{
+		/*
+		 * If the file is not found, assume it's because the standby asked for
+		 * a too old WAL segment that has already been removed or recycled.
+		 */
+		if (errno == ENOENT)
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("requested WAL segment %s has already been removed",
+							XLogFileNameP(seg->tli, seg->num))));
+		else
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("could not open file \"%s\": %m",
+							path)));
+	}
+}
+
 /*
  * Send out the WAL in its normal physical/stored form.
  *
@@ -2549,6 +2623,7 @@ XLogSendPhysical(void)
 	XLogRecPtr	startptr;
 	XLogRecPtr	endptr;
 	Size		nbytes;
+	XLogSegNo	segno;
 
 	/* If requested switch the WAL sender to the stopping state. */
 	if (got_STOPPING)
@@ -2764,7 +2839,49 @@ XLogSendPhysical(void)
 	 * calls.
 	 */
 	enlargeStringInfo(&output_message, nbytes);
-	XLogRead(&output_message.data[output_message.len], startptr, nbytes);
+
+retry:
+	if (!XLogRead(&output_message.data[output_message.len], startptr, nbytes,
+				  NULL,			/* WalSndOpenSegment will determine TLI */
+				  sendSeg,
+				  WalSndOpenSegment))
+		XLogReadProcessError(sendSeg);
+
+	/*
+	 * After reading into the buffer, check that what we read was valid. We do
+	 * this after reading, because even though the segment was present when we
+	 * opened it, it might get recycled or removed while we read it. The
+	 * read() succeeds in that case, but the data we tried to read might
+	 * already have been overwritten with new WAL records.
+	 */
+	XLByteToSeg(startptr, segno, wal_segment_size);
+	CheckXLogRemoved(segno, ThisTimeLineID);
+
+	/*
+	 * During recovery, the currently-open WAL file might be replaced with the
+	 * file of the same name retrieved from archive. So we always need to
+	 * check what we read was valid after reading into the buffer. If it's
+	 * invalid, we try to open and read the file again.
+	 */
+	if (am_cascading_walsender)
+	{
+		WalSnd	   *walsnd = MyWalSnd;
+		bool		reload;
+
+		SpinLockAcquire(&walsnd->mutex);
+		reload = walsnd->needreload;
+		walsnd->needreload = false;
+		SpinLockRelease(&walsnd->mutex);
+
+		if (reload && sendSeg->file >= 0)
+		{
+			close(sendSeg->file);
+			sendSeg->file = -1;
+
+			goto retry;
+		}
+	}
+
 	output_message.len += nbytes;
 	output_message.data[output_message.len] = '\0';
 
diff --git a/src/bin/pg_waldump/pg_waldump.c b/src/bin/pg_waldump/pg_waldump.c
index a16793bb8b..3e09519f88 100644
--- a/src/bin/pg_waldump/pg_waldump.c
+++ b/src/bin/pg_waldump/pg_waldump.c
@@ -296,6 +296,45 @@ identify_target_directory(XLogDumpPrivate *private, char *directory,
 		fatal_error("could not find any WAL file");
 }
 
+static void
+XLogDumpOpenSegment(XLogSegNo nextSegNo, TimeLineID *tli, XLogSegment *seg)
+{
+	char		fname[MAXPGPATH];
+	int			tries;
+
+	XLogFileName(fname, *tli, nextSegNo, seg->size);
+
+	/*
+	 * In follow mode there is a short period of time after the server has
+	 * written the end of the previous file before the new file is available.
+	 * So we loop for 5 seconds looking for the file to appear before giving
+	 * up.
+	 */
+	for (tries = 0; tries < 10; tries++)
+	{
+		seg->file = open_file_in_directory(seg->dir, fname);
+		if (seg->file >= 0)
+			break;
+		if (errno == ENOENT)
+		{
+			int			save_errno = errno;
+
+			/* File not there yet, try again */
+			pg_usleep(500 * 1000);
+
+			errno = save_errno;
+			continue;
+		}
+		/* Any other error, fall through and fail */
+		break;
+	}
+
+	if (seg->file < 0)
+		fatal_error("could not find file \"%s\": %s",
+					fname, strerror(errno));
+	seg->tli = *tli;
+}
+
 /*
  * Read count bytes from a segment file in the specified directory, for the
  * given timeline, containing the specified record pointer; store the data in
@@ -441,8 +480,24 @@ XLogDumpReadPage(XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen,
 		}
 	}
 
-	XLogDumpXLogRead(private->inpath, private->timeline, targetPagePtr,
-					 readBuff, count);
+	if (!XLogRead(readBuff, targetPagePtr, count, &private->timeline,
+				  &state->seg, XLogDumpOpenSegment))
+	{
+		XLogSegment *seg = &state->seg;
+		int			err = errno;
+		char		fname[MAXPGPATH];
+		int			save_errno = errno;
+
+		XLogFileName(fname, seg->tli, seg->num, seg->size);
+		errno = save_errno;
+
+		if (errno != 0)
+			fatal_error("could not read from log file %s, offset %u, length %zu: %s",
+						fname, seg->off, (Size) seg->last_req, strerror(err));
+		else
+			fatal_error("could not read from log file %s, offset %u: length: %zu",
+						fname, seg->off, (Size) seg->last_req);
+	}
 
 	return count;
 }
diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h
index 26da3613f7..856e1b2e00 100644
--- a/src/include/access/xlogreader.h
+++ b/src/include/access/xlogreader.h
@@ -39,6 +39,7 @@ typedef struct XLogSegment
 
 	char	   *dir;			/* directory (only needed by frontends) */
 	int			size;			/* segment size */
+	int			last_req;		/* the amount of data requested last time */
 } XLogSegment;
 
 typedef struct XLogReaderState XLogReaderState;
@@ -221,7 +222,25 @@ extern bool XLogReaderValidatePageHeader(XLogReaderState *state,
 extern XLogRecPtr XLogFindNextRecord(XLogReaderState *state, XLogRecPtr RecPtr);
 #endif							/* FRONTEND */
 
+/*
+ * Callback to open the specified XLOG segment nextSegNo in timeline *tli for
+ * reading, and assign the descriptor to ->file. BasicOpenFile() is the
+ * preferred way to open the segment file in backend code, whereas open(2)
+ * should be used in frontend.
+ *
+ * If NULL is passed for tli, the callback must determine the timeline
+ * itself. In any case it's supposed to eventually set ->tli.
+ */
+typedef void (*XLogOpenSegment) (XLogSegNo nextSegNo, TimeLineID *tli,
+								 XLogSegment *seg);
+
 extern void XLogSegmentInit(XLogSegment *seg, int size);
+extern bool XLogRead(char *buf, XLogRecPtr startptr, Size count,
+					 TimeLineID *tli, XLogSegment *seg,
+					 XLogOpenSegment openSegment);
+#ifndef FRONTEND
+void		XLogReadProcessError(XLogSegment *seg);
+#endif
 
 /* Functions for decoding an XLogRecord */
 
-- 
2.16.4


--=-=-=
Content-Type: text/x-diff
Content-Disposition: attachment;
 filename=v04-0005-Remove-the-old-implemenations-of-XLogRead.patch



^ permalink  raw  reply  [nested|flat] 8+ messages in thread

* [PATCH 3/4] Use only xlogreader.c:XLogRead()
@ 2019-09-09 09:53  Antonin Houska <ah@cybertec.at>
  0 siblings, 0 replies; 8+ messages in thread

From: Antonin Houska @ 2019-09-09 09:53 UTC (permalink / raw)

The implementations in xlogutils.c and walsender.c are just renamed now, to be
removed by the following diff.
---
 src/backend/access/transam/xlogreader.c | 128 ++++++++++++++++++++++++
 src/backend/access/transam/xlogutils.c  |  40 ++++++--
 src/backend/replication/walsender.c     | 125 ++++++++++++++++++++++-
 src/bin/pg_waldump/pg_waldump.c         |  59 ++++++++++-
 src/include/access/xlogreader.h         |  19 ++++
 5 files changed, 359 insertions(+), 12 deletions(-)

diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c
index 7b4ec81493..2a80bc5823 100644
--- a/src/backend/access/transam/xlogreader.c
+++ b/src/backend/access/transam/xlogreader.c
@@ -17,6 +17,8 @@
  */
 #include "postgres.h"
 
+#include <unistd.h>
+
 #include "access/transam.h"
 #include "access/xlogrecord.h"
 #include "access/xlog_internal.h"
@@ -27,6 +29,7 @@
 
 #ifndef FRONTEND
 #include "miscadmin.h"
+#include "pgstat.h"
 #include "utils/memutils.h"
 #endif
 
@@ -1011,7 +1014,132 @@ XLogSegmentInit(XLogSegment *seg, int size)
 	seg->tli = 0;
 	seg->dir = NULL;
 	seg->size = size;
+	seg->last_req = 0;
+}
+
+/*
+ * Read 'count' bytes from WAL into 'buf', starting at location 'startptr'. If
+ * tli is passed, get the data from timeline *tli. 'pos' is the current
+ * position in the XLOG file and openSegment is a callback that opens the next
+ * segment for reading.
+ *
+ * Returns true if the call succeeded, false if it failed. Caller should check
+ * errno in the case of failure. seg->last_req might also be useful for error
+ * messages.
+ *
+ * XXX probably this should be improved to suck data directly from the
+ * WAL buffers when possible.
+ */
+bool
+XLogRead(char *buf, XLogRecPtr startptr, Size count,
+		 TimeLineID *tli, XLogSegment *seg, XLogOpenSegment openSegment)
+{
+	char	   *p;
+	XLogRecPtr	recptr;
+	Size		nbytes;
+
+	p = buf;
+	recptr = startptr;
+	nbytes = count;
+
+	while (nbytes > 0)
+	{
+		int			readbytes;
+
+		seg->off = XLogSegmentOffset(recptr, seg->size);
+
+		if (seg->file < 0 ||
+			!XLByteInSeg(recptr, seg->num, seg->size) ||
+			(tli != NULL && *tli != seg->tli))
+		{
+			XLogSegNo	nextSegNo;
+
+			/* Switch to another logfile segment */
+			if (seg->file >= 0)
+				close(seg->file);
+
+			XLByteToSeg(recptr, nextSegNo, seg->size);
+
+			/* Open the next segment in the caller's way. */
+			openSegment(nextSegNo, tli, seg);
+
+			/*
+			 * If the function is called by the XLOG reader, the reader will
+			 * eventually set both "num" and "off". However we need to care
+			 * about them too because the function can also be used directly,
+			 * see walsender.c.
+			 */
+			seg->num = nextSegNo;
+			seg->off = 0;
+		}
+
+		/* How many bytes are within this segment? */
+		if (nbytes > (seg->size - seg->off))
+			seg->last_req = seg->size - seg->off;
+		else
+			seg->last_req = nbytes;
+
+#ifndef FRONTEND
+		pgstat_report_wait_start(WAIT_EVENT_WAL_READ);
+#endif
+
+		/*
+		 * Failure to read the data does not necessarily imply non-zero errno.
+		 * Set it to zero so that caller can distinguish the failure that does
+		 * not affect errno.
+		 */
+		errno = 0;
+
+		readbytes = pg_pread(seg->file, p, seg->last_req, seg->off);
+
+#ifndef FRONTEND
+		pgstat_report_wait_end();
+#endif
+
+		if (readbytes <= 0)
+			return false;
+
+		/* Update state for read */
+		recptr += readbytes;
+		nbytes -= readbytes;
+		p += readbytes;
+
+		/*
+		 * If the function is called by the XLOG reader, the reader will
+		 * eventually set this field. However we need to care about it too
+		 * because the function can also be used directly (see walsender.c).
+		 */
+		seg->off += readbytes;
+	}
+
+	return true;
+}
+
+#ifndef FRONTEND
+/*
+ * Backend-specific code to handle errors encountered by XLogRead().
+ */
+void
+XLogReadProcessError(XLogSegment *seg)
+{
+	if (errno != 0)
+	{
+		ereport(ERROR,
+				(errcode_for_file_access(),
+				 errmsg("could not read from log segment %s, offset %u, length %zu: %m",
+						XLogFileNameP(seg->tli, seg->num), seg->off,
+						(Size) seg->last_req)));
+	}
+	else
+	{
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_CORRUPTED),
+				 errmsg("could not read from log segment %s, offset %u: length %zu",
+						XLogFileNameP(seg->tli, seg->num), seg->off,
+						(Size) seg->last_req)));
+	}
 }
+#endif
 
 /* ----------------------------------------
  * Functions for decoding the data and block references in a record.
diff --git a/src/backend/access/transam/xlogutils.c b/src/backend/access/transam/xlogutils.c
index 424bb06919..83b014e04c 100644
--- a/src/backend/access/transam/xlogutils.c
+++ b/src/backend/access/transam/xlogutils.c
@@ -653,8 +653,8 @@ XLogTruncateRelation(RelFileNode rnode, ForkNumber forkNum,
  * frontend).  Probably these should be merged at some point.
  */
 static void
-XLogRead(char *buf, int segsize, TimeLineID tli, XLogRecPtr startptr,
-		 Size count)
+XLogReadOld(char *buf, int segsize, TimeLineID tli, XLogRecPtr startptr,
+			Size count)
 {
 	char	   *p;
 	XLogRecPtr	recptr;
@@ -896,6 +896,35 @@ XLogReadDetermineTimeline(XLogReaderState *state, XLogRecPtr wantPage, uint32 wa
 	}
 }
 
+/*
+ * Callback for XLogRead() to open the next segment.
+ */
+static void
+read_local_xlog_page_open_segment(XLogSegNo nextSegNo, TimeLineID *tli,
+								  XLogSegment *seg)
+{
+	char		path[MAXPGPATH];
+
+	XLogFilePath(path, *tli, nextSegNo, seg->size);
+	seg->file = BasicOpenFile(path, O_RDONLY | PG_BINARY);
+
+	if (seg->file < 0)
+	{
+		if (errno == ENOENT)
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("requested WAL segment %s has already been removed",
+							path)));
+		else
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("could not open file \"%s\": %m",
+							path)));
+	}
+
+	seg->tli = *tli;
+}
+
 /*
  * read_page callback for reading local xlog files
  *
@@ -1022,10 +1051,9 @@ read_local_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr,
 	 * as 'count', read the whole page anyway. It's guaranteed to be
 	 * zero-padded up to the page boundary if it's incomplete.
 	 */
-	XLogRead(cur_page, state->seg.size, state->seg.tli, targetPagePtr,
-			 XLOG_BLCKSZ);
-	state->seg.tli = pageTLI;
-
+	if (!XLogRead(cur_page, targetPagePtr, XLOG_BLCKSZ, &pageTLI,
+				  &state->seg, read_local_xlog_page_open_segment))
+		XLogReadProcessError(&state->seg);
 	/* number of valid bytes in the buffer */
 	return count;
 }
diff --git a/src/backend/replication/walsender.c b/src/backend/replication/walsender.c
index f5630a63cf..0685c320b4 100644
--- a/src/backend/replication/walsender.c
+++ b/src/backend/replication/walsender.c
@@ -247,7 +247,9 @@ static void LagTrackerWrite(XLogRecPtr lsn, TimestampTz local_flush_time);
 static TimeOffset LagTrackerRead(int head, XLogRecPtr lsn, TimestampTz now);
 static bool TransactionIdInRecentPast(TransactionId xid, uint32 epoch);
 
-static void XLogRead(char *buf, XLogRecPtr startptr, Size count);
+static void WalSndOpenSegment(XLogSegNo nextSegNo, TimeLineID *tli,
+							  XLogSegment *seg);
+static void XLogReadOld(char *buf, XLogRecPtr startptr, Size count);
 
 
 /* Initialize walsender process before entering the main command loop */
@@ -782,7 +784,9 @@ logical_read_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr, int req
 		count = flushptr - targetPagePtr;	/* part of the page available */
 
 	/* now actually read the data, we know it's there */
-	XLogRead(cur_page, targetPagePtr, XLOG_BLCKSZ);
+	if (!XLogRead(cur_page, targetPagePtr, XLOG_BLCKSZ, NULL, sendSeg,
+				  WalSndOpenSegment))
+		XLogReadProcessError(sendSeg);
 
 	return count;
 }
@@ -2359,7 +2363,7 @@ WalSndKill(int code, Datum arg)
  * more than one.
  */
 static void
-XLogRead(char *buf, XLogRecPtr startptr, Size count)
+XLogReadOld(char *buf, XLogRecPtr startptr, Size count)
 {
 	char	   *p;
 	XLogRecPtr	recptr;
@@ -2532,6 +2536,76 @@ retry:
 	}
 }
 
+/*
+ * Callback for XLogRead() to open the next segment.
+ */
+void
+WalSndOpenSegment(XLogSegNo nextSegNo, TimeLineID *tli, XLogSegment *seg)
+{
+	char		path[MAXPGPATH];
+
+	/*
+	 * The timeline is determined below, caller should not do anything about
+	 * it.
+	 */
+	Assert(tli == NULL);
+
+	/*-------
+	 * When reading from a historic timeline, and there is a timeline switch
+	 * within this segment, read from the WAL segment belonging to the new
+	 * timeline.
+	 *
+	 * For example, imagine that this server is currently on timeline 5, and
+	 * we're streaming timeline 4. The switch from timeline 4 to 5 happened at
+	 * 0/13002088. In pg_wal, we have these files:
+	 *
+	 * ...
+	 * 000000040000000000000012
+	 * 000000040000000000000013
+	 * 000000050000000000000013
+	 * 000000050000000000000014
+	 * ...
+	 *
+	 * In this situation, when requested to send the WAL from segment 0x13, on
+	 * timeline 4, we read the WAL from file 000000050000000000000013. Archive
+	 * recovery prefers files from newer timelines, so if the segment was
+	 * restored from the archive on this server, the file belonging to the old
+	 * timeline, 000000040000000000000013, might not exist. Their contents are
+	 * equal up to the switchpoint, because at a timeline switch, the used
+	 * portion of the old segment is copied to the new file.  -------
+	 */
+	seg->tli = sendTimeLine;
+	if (sendTimeLineIsHistoric)
+	{
+		XLogSegNo	endSegNo;
+
+		XLByteToSeg(sendTimeLineValidUpto, endSegNo, seg->size);
+		if (seg->num == endSegNo)
+			seg->tli = sendTimeLineNextTLI;
+	}
+
+	XLogFilePath(path, seg->tli, nextSegNo, seg->size);
+	seg->file = BasicOpenFile(path, O_RDONLY | PG_BINARY);
+
+	if (seg->file < 0)
+	{
+		/*
+		 * If the file is not found, assume it's because the standby asked for
+		 * a too old WAL segment that has already been removed or recycled.
+		 */
+		if (errno == ENOENT)
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("requested WAL segment %s has already been removed",
+							XLogFileNameP(seg->tli, seg->num))));
+		else
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("could not open file \"%s\": %m",
+							path)));
+	}
+}
+
 /*
  * Send out the WAL in its normal physical/stored form.
  *
@@ -2549,6 +2623,7 @@ XLogSendPhysical(void)
 	XLogRecPtr	startptr;
 	XLogRecPtr	endptr;
 	Size		nbytes;
+	XLogSegNo	segno;
 
 	/* If requested switch the WAL sender to the stopping state. */
 	if (got_STOPPING)
@@ -2764,7 +2839,49 @@ XLogSendPhysical(void)
 	 * calls.
 	 */
 	enlargeStringInfo(&output_message, nbytes);
-	XLogRead(&output_message.data[output_message.len], startptr, nbytes);
+
+retry:
+	if (!XLogRead(&output_message.data[output_message.len], startptr, nbytes,
+				  NULL,			/* WalSndOpenSegment will determine TLI */
+				  sendSeg,
+				  WalSndOpenSegment))
+		XLogReadProcessError(sendSeg);
+
+	/*
+	 * After reading into the buffer, check that what we read was valid. We do
+	 * this after reading, because even though the segment was present when we
+	 * opened it, it might get recycled or removed while we read it. The
+	 * read() succeeds in that case, but the data we tried to read might
+	 * already have been overwritten with new WAL records.
+	 */
+	XLByteToSeg(startptr, segno, wal_segment_size);
+	CheckXLogRemoved(segno, ThisTimeLineID);
+
+	/*
+	 * During recovery, the currently-open WAL file might be replaced with the
+	 * file of the same name retrieved from archive. So we always need to
+	 * check what we read was valid after reading into the buffer. If it's
+	 * invalid, we try to open and read the file again.
+	 */
+	if (am_cascading_walsender)
+	{
+		WalSnd	   *walsnd = MyWalSnd;
+		bool		reload;
+
+		SpinLockAcquire(&walsnd->mutex);
+		reload = walsnd->needreload;
+		walsnd->needreload = false;
+		SpinLockRelease(&walsnd->mutex);
+
+		if (reload && sendSeg->file >= 0)
+		{
+			close(sendSeg->file);
+			sendSeg->file = -1;
+
+			goto retry;
+		}
+	}
+
 	output_message.len += nbytes;
 	output_message.data[output_message.len] = '\0';
 
diff --git a/src/bin/pg_waldump/pg_waldump.c b/src/bin/pg_waldump/pg_waldump.c
index a16793bb8b..3e09519f88 100644
--- a/src/bin/pg_waldump/pg_waldump.c
+++ b/src/bin/pg_waldump/pg_waldump.c
@@ -296,6 +296,45 @@ identify_target_directory(XLogDumpPrivate *private, char *directory,
 		fatal_error("could not find any WAL file");
 }
 
+static void
+XLogDumpOpenSegment(XLogSegNo nextSegNo, TimeLineID *tli, XLogSegment *seg)
+{
+	char		fname[MAXPGPATH];
+	int			tries;
+
+	XLogFileName(fname, *tli, nextSegNo, seg->size);
+
+	/*
+	 * In follow mode there is a short period of time after the server has
+	 * written the end of the previous file before the new file is available.
+	 * So we loop for 5 seconds looking for the file to appear before giving
+	 * up.
+	 */
+	for (tries = 0; tries < 10; tries++)
+	{
+		seg->file = open_file_in_directory(seg->dir, fname);
+		if (seg->file >= 0)
+			break;
+		if (errno == ENOENT)
+		{
+			int			save_errno = errno;
+
+			/* File not there yet, try again */
+			pg_usleep(500 * 1000);
+
+			errno = save_errno;
+			continue;
+		}
+		/* Any other error, fall through and fail */
+		break;
+	}
+
+	if (seg->file < 0)
+		fatal_error("could not find file \"%s\": %s",
+					fname, strerror(errno));
+	seg->tli = *tli;
+}
+
 /*
  * Read count bytes from a segment file in the specified directory, for the
  * given timeline, containing the specified record pointer; store the data in
@@ -441,8 +480,24 @@ XLogDumpReadPage(XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen,
 		}
 	}
 
-	XLogDumpXLogRead(private->inpath, private->timeline, targetPagePtr,
-					 readBuff, count);
+	if (!XLogRead(readBuff, targetPagePtr, count, &private->timeline,
+				  &state->seg, XLogDumpOpenSegment))
+	{
+		XLogSegment *seg = &state->seg;
+		int			err = errno;
+		char		fname[MAXPGPATH];
+		int			save_errno = errno;
+
+		XLogFileName(fname, seg->tli, seg->num, seg->size);
+		errno = save_errno;
+
+		if (errno != 0)
+			fatal_error("could not read from log file %s, offset %u, length %zu: %s",
+						fname, seg->off, (Size) seg->last_req, strerror(err));
+		else
+			fatal_error("could not read from log file %s, offset %u: length: %zu",
+						fname, seg->off, (Size) seg->last_req);
+	}
 
 	return count;
 }
diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h
index c2724fff74..4731023ccc 100644
--- a/src/include/access/xlogreader.h
+++ b/src/include/access/xlogreader.h
@@ -43,6 +43,7 @@ typedef struct XLogSegment
 
 	char	   *dir;			/* directory (only needed by frontends) */
 	int			size;			/* segment size */
+	int			last_req;		/* the amount of data requested last time */
 } XLogSegment;
 
 typedef struct XLogReaderState XLogReaderState;
@@ -225,7 +226,25 @@ extern bool XLogReaderValidatePageHeader(XLogReaderState *state,
 extern XLogRecPtr XLogFindNextRecord(XLogReaderState *state, XLogRecPtr RecPtr);
 #endif							/* FRONTEND */
 
+/*
+ * Callback to open the specified XLOG segment nextSegNo in timeline *tli for
+ * reading, and assign the descriptor to ->file. BasicOpenFile() is the
+ * preferred way to open the segment file in backend code, whereas open(2)
+ * should be used in frontend.
+ *
+ * If NULL is passed for tli, the callback must determine the timeline
+ * itself. In any case it's supposed to eventually set ->tli.
+ */
+typedef void (*XLogOpenSegment) (XLogSegNo nextSegNo, TimeLineID *tli,
+								 XLogSegment *seg);
+
 extern void XLogSegmentInit(XLogSegment *seg, int size);
+extern bool XLogRead(char *buf, XLogRecPtr startptr, Size count,
+					 TimeLineID *tli, XLogSegment *seg,
+					 XLogOpenSegment openSegment);
+#ifndef FRONTEND
+void		XLogReadProcessError(XLogSegment *seg);
+#endif
 
 /* Functions for decoding an XLogRecord */
 
-- 
2.22.0


--=-=-=
Content-Type: text/x-diff
Content-Disposition: attachment;
 filename=v05-0005-Remove-the-old-implemenations-of-XLogRead.patch



^ permalink  raw  reply  [nested|flat] 8+ messages in thread

* [PATCH 3/4] Use only xlogreader.c:XLogRead()
@ 2019-09-09 09:53  Antonin Houska <ah@cybertec.at>
  0 siblings, 0 replies; 8+ messages in thread

From: Antonin Houska @ 2019-09-09 09:53 UTC (permalink / raw)

The implementations in xlogutils.c and walsender.c are just renamed now, to be
removed by the following diff.
---
 src/backend/access/transam/xlogreader.c | 128 ++++++++++++++++++++++++
 src/backend/access/transam/xlogutils.c  |  40 ++++++--
 src/backend/replication/walsender.c     | 125 ++++++++++++++++++++++-
 src/bin/pg_waldump/pg_waldump.c         |  59 ++++++++++-
 src/include/access/xlogreader.h         |  19 ++++
 5 files changed, 359 insertions(+), 12 deletions(-)

diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c
index 7b4ec81493..2a80bc5823 100644
--- a/src/backend/access/transam/xlogreader.c
+++ b/src/backend/access/transam/xlogreader.c
@@ -17,6 +17,8 @@
  */
 #include "postgres.h"
 
+#include <unistd.h>
+
 #include "access/transam.h"
 #include "access/xlogrecord.h"
 #include "access/xlog_internal.h"
@@ -27,6 +29,7 @@
 
 #ifndef FRONTEND
 #include "miscadmin.h"
+#include "pgstat.h"
 #include "utils/memutils.h"
 #endif
 
@@ -1011,7 +1014,132 @@ XLogSegmentInit(XLogSegment *seg, int size)
 	seg->tli = 0;
 	seg->dir = NULL;
 	seg->size = size;
+	seg->last_req = 0;
+}
+
+/*
+ * Read 'count' bytes from WAL into 'buf', starting at location 'startptr'. If
+ * tli is passed, get the data from timeline *tli. 'pos' is the current
+ * position in the XLOG file and openSegment is a callback that opens the next
+ * segment for reading.
+ *
+ * Returns true if the call succeeded, false if it failed. Caller should check
+ * errno in the case of failure. seg->last_req might also be useful for error
+ * messages.
+ *
+ * XXX probably this should be improved to suck data directly from the
+ * WAL buffers when possible.
+ */
+bool
+XLogRead(char *buf, XLogRecPtr startptr, Size count,
+		 TimeLineID *tli, XLogSegment *seg, XLogOpenSegment openSegment)
+{
+	char	   *p;
+	XLogRecPtr	recptr;
+	Size		nbytes;
+
+	p = buf;
+	recptr = startptr;
+	nbytes = count;
+
+	while (nbytes > 0)
+	{
+		int			readbytes;
+
+		seg->off = XLogSegmentOffset(recptr, seg->size);
+
+		if (seg->file < 0 ||
+			!XLByteInSeg(recptr, seg->num, seg->size) ||
+			(tli != NULL && *tli != seg->tli))
+		{
+			XLogSegNo	nextSegNo;
+
+			/* Switch to another logfile segment */
+			if (seg->file >= 0)
+				close(seg->file);
+
+			XLByteToSeg(recptr, nextSegNo, seg->size);
+
+			/* Open the next segment in the caller's way. */
+			openSegment(nextSegNo, tli, seg);
+
+			/*
+			 * If the function is called by the XLOG reader, the reader will
+			 * eventually set both "num" and "off". However we need to care
+			 * about them too because the function can also be used directly,
+			 * see walsender.c.
+			 */
+			seg->num = nextSegNo;
+			seg->off = 0;
+		}
+
+		/* How many bytes are within this segment? */
+		if (nbytes > (seg->size - seg->off))
+			seg->last_req = seg->size - seg->off;
+		else
+			seg->last_req = nbytes;
+
+#ifndef FRONTEND
+		pgstat_report_wait_start(WAIT_EVENT_WAL_READ);
+#endif
+
+		/*
+		 * Failure to read the data does not necessarily imply non-zero errno.
+		 * Set it to zero so that caller can distinguish the failure that does
+		 * not affect errno.
+		 */
+		errno = 0;
+
+		readbytes = pg_pread(seg->file, p, seg->last_req, seg->off);
+
+#ifndef FRONTEND
+		pgstat_report_wait_end();
+#endif
+
+		if (readbytes <= 0)
+			return false;
+
+		/* Update state for read */
+		recptr += readbytes;
+		nbytes -= readbytes;
+		p += readbytes;
+
+		/*
+		 * If the function is called by the XLOG reader, the reader will
+		 * eventually set this field. However we need to care about it too
+		 * because the function can also be used directly (see walsender.c).
+		 */
+		seg->off += readbytes;
+	}
+
+	return true;
+}
+
+#ifndef FRONTEND
+/*
+ * Backend-specific code to handle errors encountered by XLogRead().
+ */
+void
+XLogReadProcessError(XLogSegment *seg)
+{
+	if (errno != 0)
+	{
+		ereport(ERROR,
+				(errcode_for_file_access(),
+				 errmsg("could not read from log segment %s, offset %u, length %zu: %m",
+						XLogFileNameP(seg->tli, seg->num), seg->off,
+						(Size) seg->last_req)));
+	}
+	else
+	{
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_CORRUPTED),
+				 errmsg("could not read from log segment %s, offset %u: length %zu",
+						XLogFileNameP(seg->tli, seg->num), seg->off,
+						(Size) seg->last_req)));
+	}
 }
+#endif
 
 /* ----------------------------------------
  * Functions for decoding the data and block references in a record.
diff --git a/src/backend/access/transam/xlogutils.c b/src/backend/access/transam/xlogutils.c
index 424bb06919..83b014e04c 100644
--- a/src/backend/access/transam/xlogutils.c
+++ b/src/backend/access/transam/xlogutils.c
@@ -653,8 +653,8 @@ XLogTruncateRelation(RelFileNode rnode, ForkNumber forkNum,
  * frontend).  Probably these should be merged at some point.
  */
 static void
-XLogRead(char *buf, int segsize, TimeLineID tli, XLogRecPtr startptr,
-		 Size count)
+XLogReadOld(char *buf, int segsize, TimeLineID tli, XLogRecPtr startptr,
+			Size count)
 {
 	char	   *p;
 	XLogRecPtr	recptr;
@@ -896,6 +896,35 @@ XLogReadDetermineTimeline(XLogReaderState *state, XLogRecPtr wantPage, uint32 wa
 	}
 }
 
+/*
+ * Callback for XLogRead() to open the next segment.
+ */
+static void
+read_local_xlog_page_open_segment(XLogSegNo nextSegNo, TimeLineID *tli,
+								  XLogSegment *seg)
+{
+	char		path[MAXPGPATH];
+
+	XLogFilePath(path, *tli, nextSegNo, seg->size);
+	seg->file = BasicOpenFile(path, O_RDONLY | PG_BINARY);
+
+	if (seg->file < 0)
+	{
+		if (errno == ENOENT)
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("requested WAL segment %s has already been removed",
+							path)));
+		else
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("could not open file \"%s\": %m",
+							path)));
+	}
+
+	seg->tli = *tli;
+}
+
 /*
  * read_page callback for reading local xlog files
  *
@@ -1022,10 +1051,9 @@ read_local_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr,
 	 * as 'count', read the whole page anyway. It's guaranteed to be
 	 * zero-padded up to the page boundary if it's incomplete.
 	 */
-	XLogRead(cur_page, state->seg.size, state->seg.tli, targetPagePtr,
-			 XLOG_BLCKSZ);
-	state->seg.tli = pageTLI;
-
+	if (!XLogRead(cur_page, targetPagePtr, XLOG_BLCKSZ, &pageTLI,
+				  &state->seg, read_local_xlog_page_open_segment))
+		XLogReadProcessError(&state->seg);
 	/* number of valid bytes in the buffer */
 	return count;
 }
diff --git a/src/backend/replication/walsender.c b/src/backend/replication/walsender.c
index f5630a63cf..0685c320b4 100644
--- a/src/backend/replication/walsender.c
+++ b/src/backend/replication/walsender.c
@@ -247,7 +247,9 @@ static void LagTrackerWrite(XLogRecPtr lsn, TimestampTz local_flush_time);
 static TimeOffset LagTrackerRead(int head, XLogRecPtr lsn, TimestampTz now);
 static bool TransactionIdInRecentPast(TransactionId xid, uint32 epoch);
 
-static void XLogRead(char *buf, XLogRecPtr startptr, Size count);
+static void WalSndOpenSegment(XLogSegNo nextSegNo, TimeLineID *tli,
+							  XLogSegment *seg);
+static void XLogReadOld(char *buf, XLogRecPtr startptr, Size count);
 
 
 /* Initialize walsender process before entering the main command loop */
@@ -782,7 +784,9 @@ logical_read_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr, int req
 		count = flushptr - targetPagePtr;	/* part of the page available */
 
 	/* now actually read the data, we know it's there */
-	XLogRead(cur_page, targetPagePtr, XLOG_BLCKSZ);
+	if (!XLogRead(cur_page, targetPagePtr, XLOG_BLCKSZ, NULL, sendSeg,
+				  WalSndOpenSegment))
+		XLogReadProcessError(sendSeg);
 
 	return count;
 }
@@ -2359,7 +2363,7 @@ WalSndKill(int code, Datum arg)
  * more than one.
  */
 static void
-XLogRead(char *buf, XLogRecPtr startptr, Size count)
+XLogReadOld(char *buf, XLogRecPtr startptr, Size count)
 {
 	char	   *p;
 	XLogRecPtr	recptr;
@@ -2532,6 +2536,76 @@ retry:
 	}
 }
 
+/*
+ * Callback for XLogRead() to open the next segment.
+ */
+void
+WalSndOpenSegment(XLogSegNo nextSegNo, TimeLineID *tli, XLogSegment *seg)
+{
+	char		path[MAXPGPATH];
+
+	/*
+	 * The timeline is determined below, caller should not do anything about
+	 * it.
+	 */
+	Assert(tli == NULL);
+
+	/*-------
+	 * When reading from a historic timeline, and there is a timeline switch
+	 * within this segment, read from the WAL segment belonging to the new
+	 * timeline.
+	 *
+	 * For example, imagine that this server is currently on timeline 5, and
+	 * we're streaming timeline 4. The switch from timeline 4 to 5 happened at
+	 * 0/13002088. In pg_wal, we have these files:
+	 *
+	 * ...
+	 * 000000040000000000000012
+	 * 000000040000000000000013
+	 * 000000050000000000000013
+	 * 000000050000000000000014
+	 * ...
+	 *
+	 * In this situation, when requested to send the WAL from segment 0x13, on
+	 * timeline 4, we read the WAL from file 000000050000000000000013. Archive
+	 * recovery prefers files from newer timelines, so if the segment was
+	 * restored from the archive on this server, the file belonging to the old
+	 * timeline, 000000040000000000000013, might not exist. Their contents are
+	 * equal up to the switchpoint, because at a timeline switch, the used
+	 * portion of the old segment is copied to the new file.  -------
+	 */
+	seg->tli = sendTimeLine;
+	if (sendTimeLineIsHistoric)
+	{
+		XLogSegNo	endSegNo;
+
+		XLByteToSeg(sendTimeLineValidUpto, endSegNo, seg->size);
+		if (seg->num == endSegNo)
+			seg->tli = sendTimeLineNextTLI;
+	}
+
+	XLogFilePath(path, seg->tli, nextSegNo, seg->size);
+	seg->file = BasicOpenFile(path, O_RDONLY | PG_BINARY);
+
+	if (seg->file < 0)
+	{
+		/*
+		 * If the file is not found, assume it's because the standby asked for
+		 * a too old WAL segment that has already been removed or recycled.
+		 */
+		if (errno == ENOENT)
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("requested WAL segment %s has already been removed",
+							XLogFileNameP(seg->tli, seg->num))));
+		else
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("could not open file \"%s\": %m",
+							path)));
+	}
+}
+
 /*
  * Send out the WAL in its normal physical/stored form.
  *
@@ -2549,6 +2623,7 @@ XLogSendPhysical(void)
 	XLogRecPtr	startptr;
 	XLogRecPtr	endptr;
 	Size		nbytes;
+	XLogSegNo	segno;
 
 	/* If requested switch the WAL sender to the stopping state. */
 	if (got_STOPPING)
@@ -2764,7 +2839,49 @@ XLogSendPhysical(void)
 	 * calls.
 	 */
 	enlargeStringInfo(&output_message, nbytes);
-	XLogRead(&output_message.data[output_message.len], startptr, nbytes);
+
+retry:
+	if (!XLogRead(&output_message.data[output_message.len], startptr, nbytes,
+				  NULL,			/* WalSndOpenSegment will determine TLI */
+				  sendSeg,
+				  WalSndOpenSegment))
+		XLogReadProcessError(sendSeg);
+
+	/*
+	 * After reading into the buffer, check that what we read was valid. We do
+	 * this after reading, because even though the segment was present when we
+	 * opened it, it might get recycled or removed while we read it. The
+	 * read() succeeds in that case, but the data we tried to read might
+	 * already have been overwritten with new WAL records.
+	 */
+	XLByteToSeg(startptr, segno, wal_segment_size);
+	CheckXLogRemoved(segno, ThisTimeLineID);
+
+	/*
+	 * During recovery, the currently-open WAL file might be replaced with the
+	 * file of the same name retrieved from archive. So we always need to
+	 * check what we read was valid after reading into the buffer. If it's
+	 * invalid, we try to open and read the file again.
+	 */
+	if (am_cascading_walsender)
+	{
+		WalSnd	   *walsnd = MyWalSnd;
+		bool		reload;
+
+		SpinLockAcquire(&walsnd->mutex);
+		reload = walsnd->needreload;
+		walsnd->needreload = false;
+		SpinLockRelease(&walsnd->mutex);
+
+		if (reload && sendSeg->file >= 0)
+		{
+			close(sendSeg->file);
+			sendSeg->file = -1;
+
+			goto retry;
+		}
+	}
+
 	output_message.len += nbytes;
 	output_message.data[output_message.len] = '\0';
 
diff --git a/src/bin/pg_waldump/pg_waldump.c b/src/bin/pg_waldump/pg_waldump.c
index a16793bb8b..3e09519f88 100644
--- a/src/bin/pg_waldump/pg_waldump.c
+++ b/src/bin/pg_waldump/pg_waldump.c
@@ -296,6 +296,45 @@ identify_target_directory(XLogDumpPrivate *private, char *directory,
 		fatal_error("could not find any WAL file");
 }
 
+static void
+XLogDumpOpenSegment(XLogSegNo nextSegNo, TimeLineID *tli, XLogSegment *seg)
+{
+	char		fname[MAXPGPATH];
+	int			tries;
+
+	XLogFileName(fname, *tli, nextSegNo, seg->size);
+
+	/*
+	 * In follow mode there is a short period of time after the server has
+	 * written the end of the previous file before the new file is available.
+	 * So we loop for 5 seconds looking for the file to appear before giving
+	 * up.
+	 */
+	for (tries = 0; tries < 10; tries++)
+	{
+		seg->file = open_file_in_directory(seg->dir, fname);
+		if (seg->file >= 0)
+			break;
+		if (errno == ENOENT)
+		{
+			int			save_errno = errno;
+
+			/* File not there yet, try again */
+			pg_usleep(500 * 1000);
+
+			errno = save_errno;
+			continue;
+		}
+		/* Any other error, fall through and fail */
+		break;
+	}
+
+	if (seg->file < 0)
+		fatal_error("could not find file \"%s\": %s",
+					fname, strerror(errno));
+	seg->tli = *tli;
+}
+
 /*
  * Read count bytes from a segment file in the specified directory, for the
  * given timeline, containing the specified record pointer; store the data in
@@ -441,8 +480,24 @@ XLogDumpReadPage(XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen,
 		}
 	}
 
-	XLogDumpXLogRead(private->inpath, private->timeline, targetPagePtr,
-					 readBuff, count);
+	if (!XLogRead(readBuff, targetPagePtr, count, &private->timeline,
+				  &state->seg, XLogDumpOpenSegment))
+	{
+		XLogSegment *seg = &state->seg;
+		int			err = errno;
+		char		fname[MAXPGPATH];
+		int			save_errno = errno;
+
+		XLogFileName(fname, seg->tli, seg->num, seg->size);
+		errno = save_errno;
+
+		if (errno != 0)
+			fatal_error("could not read from log file %s, offset %u, length %zu: %s",
+						fname, seg->off, (Size) seg->last_req, strerror(err));
+		else
+			fatal_error("could not read from log file %s, offset %u: length: %zu",
+						fname, seg->off, (Size) seg->last_req);
+	}
 
 	return count;
 }
diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h
index c2724fff74..4731023ccc 100644
--- a/src/include/access/xlogreader.h
+++ b/src/include/access/xlogreader.h
@@ -43,6 +43,7 @@ typedef struct XLogSegment
 
 	char	   *dir;			/* directory (only needed by frontends) */
 	int			size;			/* segment size */
+	int			last_req;		/* the amount of data requested last time */
 } XLogSegment;
 
 typedef struct XLogReaderState XLogReaderState;
@@ -225,7 +226,25 @@ extern bool XLogReaderValidatePageHeader(XLogReaderState *state,
 extern XLogRecPtr XLogFindNextRecord(XLogReaderState *state, XLogRecPtr RecPtr);
 #endif							/* FRONTEND */
 
+/*
+ * Callback to open the specified XLOG segment nextSegNo in timeline *tli for
+ * reading, and assign the descriptor to ->file. BasicOpenFile() is the
+ * preferred way to open the segment file in backend code, whereas open(2)
+ * should be used in frontend.
+ *
+ * If NULL is passed for tli, the callback must determine the timeline
+ * itself. In any case it's supposed to eventually set ->tli.
+ */
+typedef void (*XLogOpenSegment) (XLogSegNo nextSegNo, TimeLineID *tli,
+								 XLogSegment *seg);
+
 extern void XLogSegmentInit(XLogSegment *seg, int size);
+extern bool XLogRead(char *buf, XLogRecPtr startptr, Size count,
+					 TimeLineID *tli, XLogSegment *seg,
+					 XLogOpenSegment openSegment);
+#ifndef FRONTEND
+void		XLogReadProcessError(XLogSegment *seg);
+#endif
 
 /* Functions for decoding an XLogRecord */
 
-- 
2.22.0


--=-=-=
Content-Type: text/x-diff
Content-Disposition: attachment;
 filename=v05-0005-Remove-the-old-implemenations-of-XLogRead.patch



^ permalink  raw  reply  [nested|flat] 8+ messages in thread

* [PATCH 5/6] Use only xlogreader.c:XLogRead()
@ 2019-09-23 05:40  Antonin Houska <ah@cybertec.at>
  0 siblings, 0 replies; 8+ messages in thread

From: Antonin Houska @ 2019-09-23 05:40 UTC (permalink / raw)

The implementations in xlogutils.c and walsender.c are just renamed now, to be
removed by the following diff.
---
 src/backend/access/transam/xlogreader.c | 157 ++++++++++++++++++++++++
 src/backend/access/transam/xlogutils.c  |  46 ++++++-
 src/backend/replication/walsender.c     | 139 ++++++++++++++++++++-
 src/bin/pg_waldump/pg_waldump.c         |  64 +++++++++-
 src/include/access/xlogreader.h         |  42 +++++++
 5 files changed, 436 insertions(+), 12 deletions(-)

diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c
index 4de5530b3e..7f77fa95cb 100644
--- a/src/backend/access/transam/xlogreader.c
+++ b/src/backend/access/transam/xlogreader.c
@@ -17,6 +17,8 @@
  */
 #include "postgres.h"
 
+#include <unistd.h>
+
 #include "access/transam.h"
 #include "access/xlogrecord.h"
 #include "access/xlog_internal.h"
@@ -27,6 +29,7 @@
 
 #ifndef FRONTEND
 #include "miscadmin.h"
+#include "pgstat.h"
 #include "utils/memutils.h"
 #endif
 
@@ -1015,6 +1018,160 @@ WALOpenSegmentInit(WALOpenSegment *seg, int size)
 #endif
 }
 
+/*
+ * Read 'count' bytes from WAL into 'buf', starting at location 'startptr'. If
+ * tli_p is passed, get the data from timeline *tli_p. 'pos' is the current
+ * position in the XLOG file and openSegment is a callback that opens the next
+ * segment for reading.
+ *
+ * Returns error information if the data could not be read or NULL if
+ * succeeded.
+ *
+ * XXX probably this should be improved to suck data directly from the
+ * WAL buffers when possible.
+ */
+XLogReadError *
+XLogRead(char *buf, XLogRecPtr startptr, Size count,
+		 TimeLineID *tli_p, WALOpenSegment *seg, WALSegmentOpen openSegment)
+{
+	char	   *p;
+	XLogRecPtr	recptr;
+	Size		nbytes;
+
+	p = buf;
+	recptr = startptr;
+	nbytes = count;
+
+	while (nbytes > 0)
+	{
+		int			segbytes;
+		int			readbytes;
+
+		seg->off = XLogSegmentOffset(recptr, seg->size);
+
+		if (seg->file < 0 ||
+			!XLByteInSeg(recptr, seg->num, seg->size) ||
+			(tli_p != NULL && *tli_p != seg->tli))
+		{
+			XLogSegNo	nextSegNo;
+			TimeLineID	tli = InvalidTimeLineID;
+			int			file;
+
+			/* Switch to another logfile segment */
+			if (seg->file >= 0)
+				close(seg->file);
+
+			XLByteToSeg(recptr, nextSegNo, seg->size);
+
+			/* If we have the TLI, let's pass it to the callback. */
+			if (tli_p != NULL)
+				tli = *tli_p;
+
+			/* Open the next segment in the caller's way. */
+			openSegment(nextSegNo, &tli, &file, seg);
+
+			/*
+			 * If we passed InvalidTimeLineID, the callback should have
+			 * determined the correct TLI and returned it.
+			 */
+			Assert(tli != InvalidTimeLineID);
+
+			/* Update the open segment info. */
+			seg->tli = tli;
+			seg->file = file;
+
+			/*
+			 * If the function is called by the XLOG reader, the reader will
+			 * eventually set both "num" and "off". However we need to care
+			 * about them too because the function can also be used directly,
+			 * see walsender.c.
+			 */
+			seg->num = nextSegNo;
+			seg->off = 0;
+		}
+
+		/* How many bytes are within this segment? */
+		if (nbytes > (seg->size - seg->off))
+			segbytes = seg->size - seg->off;
+		else
+			segbytes = nbytes;
+
+#ifndef FRONTEND
+		pgstat_report_wait_start(WAIT_EVENT_WAL_READ);
+#endif
+
+		/*
+		 * Failure to read the data does not necessarily imply non-zero errno.
+		 * Set it to zero so that caller can distinguish the failure that does
+		 * not affect errno.
+		 */
+		errno = 0;
+
+		readbytes = pg_pread(seg->file, p, segbytes, seg->off);
+
+#ifndef FRONTEND
+		pgstat_report_wait_end();
+#endif
+
+		if (readbytes <= 0)
+		{
+			XLogReadError *errinfo;
+
+			errinfo = (XLogReadError *) palloc(sizeof(XLogReadError));
+			errinfo->read_errno = errno;
+			errinfo->readbytes = readbytes;
+			errinfo->reqbytes = segbytes;
+			errinfo->seg = seg;
+
+			return errinfo;
+		}
+
+		/* Update state for read */
+		recptr += readbytes;
+		nbytes -= readbytes;
+		p += readbytes;
+
+		/*
+		 * If the function is called by the XLOG reader, the reader will
+		 * eventually set this field. However we need to care about it too
+		 * because the function can also be used directly (see walsender.c).
+		 */
+		seg->off += readbytes;
+	}
+
+	return NULL;
+}
+
+#ifndef FRONTEND
+/*
+ * Backend-specific convenience code to handle read errors encountered by
+ * XLogRead().
+ */
+void
+XLogReadProcessError(XLogReadError *errinfo)
+{
+	WALOpenSegment *seg = errinfo->seg;
+
+	if (errinfo->readbytes < 0)
+	{
+		errno = errinfo->read_errno;
+		ereport(ERROR,
+				(errcode_for_file_access(),
+				 errmsg("could not read from log segment %s, offset %u, length %zu: %m",
+						XLogFileNameP(seg->tli, seg->num), seg->off,
+						(Size) errinfo->reqbytes)));
+	}
+	else
+	{
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_CORRUPTED),
+				 errmsg("could not read from log segment %s, offset %u: length %zu",
+						XLogFileNameP(seg->tli, seg->num), seg->off,
+						(Size) errinfo->reqbytes)));
+	}
+}
+#endif
+
 /* ----------------------------------------
  * Functions for decoding the data and block references in a record.
  * ----------------------------------------
diff --git a/src/backend/access/transam/xlogutils.c b/src/backend/access/transam/xlogutils.c
index 424bb06919..38c3196168 100644
--- a/src/backend/access/transam/xlogutils.c
+++ b/src/backend/access/transam/xlogutils.c
@@ -653,8 +653,8 @@ XLogTruncateRelation(RelFileNode rnode, ForkNumber forkNum,
  * frontend).  Probably these should be merged at some point.
  */
 static void
-XLogRead(char *buf, int segsize, TimeLineID tli, XLogRecPtr startptr,
-		 Size count)
+XLogReadOld(char *buf, int segsize, TimeLineID tli, XLogRecPtr startptr,
+			Size count)
 {
 	char	   *p;
 	XLogRecPtr	recptr;
@@ -896,6 +896,39 @@ XLogReadDetermineTimeline(XLogReaderState *state, XLogRecPtr wantPage, uint32 wa
 	}
 }
 
+/*
+ * Callback for XLogRead() to open the next segment.
+ */
+static void
+read_local_xlog_page_segment_open(XLogSegNo nextSegNo, TimeLineID *tli_p,
+								  int *file_p, WALOpenSegment *seg)
+{
+	TimeLineID	tli = *tli_p;
+	char		path[MAXPGPATH];
+	int			file;
+
+	Assert(tli != InvalidTimeLineID);
+
+	XLogFilePath(path, tli, nextSegNo, seg->size);
+	file = BasicOpenFile(path, O_RDONLY | PG_BINARY);
+
+	if (file < 0)
+	{
+		if (errno == ENOENT)
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("requested WAL segment %s has already been removed",
+							path)));
+		else
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("could not open file \"%s\": %m",
+							path)));
+	}
+
+	*file_p = file;
+}
+
 /*
  * read_page callback for reading local xlog files
  *
@@ -915,6 +948,7 @@ read_local_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr,
 				loc;
 	int			count;
 	TimeLineID	pageTLI;
+	XLogReadError *errinfo;
 
 	loc = targetPagePtr + reqLen;
 
@@ -1022,10 +1056,10 @@ read_local_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr,
 	 * as 'count', read the whole page anyway. It's guaranteed to be
 	 * zero-padded up to the page boundary if it's incomplete.
 	 */
-	XLogRead(cur_page, state->seg.size, state->seg.tli, targetPagePtr,
-			 XLOG_BLCKSZ);
-	state->seg.tli = pageTLI;
-
+	if ((errinfo = XLogRead(cur_page, targetPagePtr, XLOG_BLCKSZ, &pageTLI,
+							&state->seg,
+							read_local_xlog_page_segment_open)) != NULL)
+		XLogReadProcessError(errinfo);
 	/* number of valid bytes in the buffer */
 	return count;
 }
diff --git a/src/backend/replication/walsender.c b/src/backend/replication/walsender.c
index a617e20ab6..a7b3e0ecbe 100644
--- a/src/backend/replication/walsender.c
+++ b/src/backend/replication/walsender.c
@@ -247,9 +247,12 @@ static void LagTrackerWrite(XLogRecPtr lsn, TimestampTz local_flush_time);
 static TimeOffset LagTrackerRead(int head, XLogRecPtr lsn, TimestampTz now);
 static bool TransactionIdInRecentPast(TransactionId xid, uint32 epoch);
 
-static void XLogRead(char *buf, XLogRecPtr startptr, Size count);
+static void WalSndSegmentOpen(XLogSegNo nextSegNo, TimeLineID *tli_p,
+							  int *file_p, WALOpenSegment *seg);
 
 
+static void XLogReadOld(char *buf, XLogRecPtr startptr, Size count);
+
 /* Initialize walsender process before entering the main command loop */
 void
 InitWalSender(void)
@@ -763,6 +766,7 @@ logical_read_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr, int req
 {
 	XLogRecPtr	flushptr;
 	int			count;
+	XLogReadError *errinfo;
 
 	XLogReadDetermineTimeline(state, targetPagePtr, reqLen);
 	sendTimeLineIsHistoric = (state->currTLI != ThisTimeLineID);
@@ -783,7 +787,13 @@ logical_read_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr, int req
 		count = flushptr - targetPagePtr;	/* part of the page available */
 
 	/* now actually read the data, we know it's there */
-	XLogRead(cur_page, targetPagePtr, XLOG_BLCKSZ);
+	if ((errinfo = XLogRead(cur_page,
+							targetPagePtr,
+							XLOG_BLCKSZ,
+							NULL,	/* WalSndSegmentOpen will determine TLI */
+							sendSeg,
+							WalSndSegmentOpen)) != NULL)
+		XLogReadProcessError(errinfo);
 
 	return count;
 }
@@ -2360,7 +2370,7 @@ WalSndKill(int code, Datum arg)
  * more than one.
  */
 static void
-XLogRead(char *buf, XLogRecPtr startptr, Size count)
+XLogReadOld(char *buf, XLogRecPtr startptr, Size count)
 {
 	char	   *p;
 	XLogRecPtr	recptr;
@@ -2533,6 +2543,81 @@ retry:
 	}
 }
 
+/*
+ * Callback for XLogRead() to open the next segment.
+ */
+void
+WalSndSegmentOpen(XLogSegNo nextSegNo, TimeLineID *tli_p, int *file_p,
+				  WALOpenSegment *seg)
+{
+	TimeLineID	tli = *tli_p;
+	char		path[MAXPGPATH];
+	int			file;
+
+	/*
+	 * The timeline is determined below, caller should not pass it.
+	 */
+	Assert(tli == InvalidTimeLineID);
+
+	/*-------
+	 * When reading from a historic timeline, and there is a timeline switch
+	 * within this segment, read from the WAL segment belonging to the new
+	 * timeline.
+	 *
+	 * For example, imagine that this server is currently on timeline 5, and
+	 * we're streaming timeline 4. The switch from timeline 4 to 5 happened at
+	 * 0/13002088. In pg_wal, we have these files:
+	 *
+	 * ...
+	 * 000000040000000000000012
+	 * 000000040000000000000013
+	 * 000000050000000000000013
+	 * 000000050000000000000014
+	 * ...
+	 *
+	 * In this situation, when requested to send the WAL from segment 0x13, on
+	 * timeline 4, we read the WAL from file 000000050000000000000013. Archive
+	 * recovery prefers files from newer timelines, so if the segment was
+	 * restored from the archive on this server, the file belonging to the old
+	 * timeline, 000000040000000000000013, might not exist. Their contents are
+	 * equal up to the switchpoint, because at a timeline switch, the used
+	 * portion of the old segment is copied to the new file.  -------
+	 */
+	tli = sendTimeLine;
+	if (sendTimeLineIsHistoric)
+	{
+		XLogSegNo	endSegNo;
+
+		XLByteToSeg(sendTimeLineValidUpto, endSegNo, seg->size);
+		if (seg->num == endSegNo)
+			tli = sendTimeLineNextTLI;
+	}
+
+	XLogFilePath(path, tli, nextSegNo, seg->size);
+	file = BasicOpenFile(path, O_RDONLY | PG_BINARY);
+
+	if (file < 0)
+	{
+		/*
+		 * If the file is not found, assume it's because the standby asked for
+		 * a too old WAL segment that has already been removed or recycled.
+		 */
+		if (errno == ENOENT)
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("requested WAL segment %s has already been removed",
+							XLogFileNameP(tli, nextSegNo))));
+		else
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("could not open file \"%s\": %m",
+							path)));
+	}
+
+	*file_p = file;
+	*tli_p = tli;
+}
+
 /*
  * Send out the WAL in its normal physical/stored form.
  *
@@ -2550,6 +2635,8 @@ XLogSendPhysical(void)
 	XLogRecPtr	startptr;
 	XLogRecPtr	endptr;
 	Size		nbytes;
+	XLogSegNo	segno;
+	XLogReadError *errinfo;
 
 	/* If requested switch the WAL sender to the stopping state. */
 	if (got_STOPPING)
@@ -2765,7 +2852,51 @@ XLogSendPhysical(void)
 	 * calls.
 	 */
 	enlargeStringInfo(&output_message, nbytes);
-	XLogRead(&output_message.data[output_message.len], startptr, nbytes);
+
+retry:
+	if ((errinfo = XLogRead(&output_message.data[output_message.len],
+							startptr,
+							nbytes,
+							NULL,	/* WalSndSegmentOpen will determine TLI */
+							sendSeg,
+							WalSndSegmentOpen)) != NULL)
+		XLogReadProcessError(errinfo);
+
+	/*
+	 * After reading into the buffer, check that what we read was valid. We do
+	 * this after reading, because even though the segment was present when we
+	 * opened it, it might get recycled or removed while we read it. The
+	 * read() succeeds in that case, but the data we tried to read might
+	 * already have been overwritten with new WAL records.
+	 */
+	XLByteToSeg(startptr, segno, wal_segment_size);
+	CheckXLogRemoved(segno, ThisTimeLineID);
+
+	/*
+	 * During recovery, the currently-open WAL file might be replaced with the
+	 * file of the same name retrieved from archive. So we always need to
+	 * check what we read was valid after reading into the buffer. If it's
+	 * invalid, we try to open and read the file again.
+	 */
+	if (am_cascading_walsender)
+	{
+		WalSnd	   *walsnd = MyWalSnd;
+		bool		reload;
+
+		SpinLockAcquire(&walsnd->mutex);
+		reload = walsnd->needreload;
+		walsnd->needreload = false;
+		SpinLockRelease(&walsnd->mutex);
+
+		if (reload && sendSeg->file >= 0)
+		{
+			close(sendSeg->file);
+			sendSeg->file = -1;
+
+			goto retry;
+		}
+	}
+
 	output_message.len += nbytes;
 	output_message.data[output_message.len] = '\0';
 
diff --git a/src/bin/pg_waldump/pg_waldump.c b/src/bin/pg_waldump/pg_waldump.c
index a16793bb8b..cd5f589f03 100644
--- a/src/bin/pg_waldump/pg_waldump.c
+++ b/src/bin/pg_waldump/pg_waldump.c
@@ -296,6 +296,51 @@ identify_target_directory(XLogDumpPrivate *private, char *directory,
 		fatal_error("could not find any WAL file");
 }
 
+static void
+WALDumpOpenSegment(XLogSegNo nextSegNo, TimeLineID *tli_p, int *file_p,
+				   WALOpenSegment *seg)
+{
+	TimeLineID	tli = *tli_p;
+	char		fname[MAXPGPATH];
+	int			file;
+	int			tries;
+
+	Assert(tli != InvalidTimeLineID);
+
+	XLogFileName(fname, tli, nextSegNo, seg->size);
+
+	/*
+	 * In follow mode there is a short period of time after the server has
+	 * written the end of the previous file before the new file is available.
+	 * So we loop for 5 seconds looking for the file to appear before giving
+	 * up.
+	 */
+	for (tries = 0; tries < 10; tries++)
+	{
+		file = open_file_in_directory(seg->dir, fname);
+		if (file >= 0)
+			break;
+		if (errno == ENOENT)
+		{
+			int			save_errno = errno;
+
+			/* File not there yet, try again */
+			pg_usleep(500 * 1000);
+
+			errno = save_errno;
+			continue;
+		}
+		/* Any other error, fall through and fail */
+		break;
+	}
+
+	if (file < 0)
+		fatal_error("could not find file \"%s\": %s",
+					fname, strerror(errno));
+
+	*file_p = file;
+}
+
 /*
  * Read count bytes from a segment file in the specified directory, for the
  * given timeline, containing the specified record pointer; store the data in
@@ -427,6 +472,7 @@ XLogDumpReadPage(XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen,
 {
 	XLogDumpPrivate *private = state->private_data;
 	int			count = XLOG_BLCKSZ;
+	XLogReadError *errinfo;
 
 	if (private->endptr != InvalidXLogRecPtr)
 	{
@@ -441,8 +487,22 @@ XLogDumpReadPage(XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen,
 		}
 	}
 
-	XLogDumpXLogRead(private->inpath, private->timeline, targetPagePtr,
-					 readBuff, count);
+	if ((errinfo = XLogRead(readBuff, targetPagePtr, count, &private->timeline,
+							&state->seg, WALDumpOpenSegment)) != NULL)
+	{
+		WALOpenSegment *seg = errinfo->seg;
+		char		fname[MAXPGPATH];
+
+		XLogFileName(fname, seg->tli, seg->num, seg->size);
+
+		if (errno != 0)
+			fatal_error("could not read from log file %s, offset %u, length %zu: %s",
+						fname, seg->off, (Size) errinfo->reqbytes,
+						strerror(errinfo->read_errno));
+		else
+			fatal_error("could not read from log file %s, offset %u: length: %zu",
+						fname, seg->off, (Size) errinfo->reqbytes);
+	}
 
 	return count;
 }
diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h
index b9d99d524e..3d9742b81b 100644
--- a/src/include/access/xlogreader.h
+++ b/src/include/access/xlogreader.h
@@ -227,8 +227,50 @@ extern bool XLogReaderValidatePageHeader(XLogReaderState *state,
 extern XLogRecPtr XLogFindNextRecord(XLogReaderState *state, XLogRecPtr RecPtr);
 #endif							/* FRONTEND */
 
+/*
+ * Callback to open the specified WAL segment for reading.
+ *
+ * "nextSegNo" is the number of the segment to be opened.
+ *
+ * "tli_p" is an input/output argument. If *tli_p is valid, it's the timeline
+ * the new segment should be in. If *tli_p==InvalidTimeLineID, the callback
+ * needs to determine the timeline itself and put the result into *tli_p.
+ *
+ * "file_p" points to an address the segment file descriptor should be stored
+ * at.
+ *
+ * "seg" provides information on the currently open segment. The callback is
+ * not supposed to change this info.
+ *
+ * BasicOpenFile() is the preferred way to open the segment file in backend
+ * code, whereas open(2) should be used in frontend.
+ */
+typedef void (*WALSegmentOpen) (XLogSegNo nextSegNo, TimeLineID *tli_p,
+								int *file_p, WALOpenSegment *seg);
+
 extern void WALOpenSegmentInit(WALOpenSegment *seg, int size);
 
+/*
+ * Error information that both backend and frontend caller can process.
+ *
+ * XXX Should the name be WALReadError? If so, we probably need to rename
+ * XLogRead() and XLogReadProcessError() too.
+ */
+typedef struct XLogReadError
+{
+	int			read_errno;		/* errno set by the last read(). */
+	int			readbytes;		/* Bytes read by the last read(). */
+	int			reqbytes;		/* Bytes requested to be read. */
+	WALOpenSegment *seg;		/* Segment we tried to read from. */
+} XLogReadError;
+
+extern XLogReadError *XLogRead(char *buf, XLogRecPtr startptr, Size count,
+							   TimeLineID *tli_p, WALOpenSegment *seg,
+							   WALSegmentOpen openSegment);
+#ifndef FRONTEND
+void		XLogReadProcessError(XLogReadError *errinfo);
+#endif
+
 /* Functions for decoding an XLogRecord */
 
 extern bool DecodeXLogRecord(XLogReaderState *state, XLogRecord *record,
-- 
2.20.1


--=-=-=
Content-Type: text/x-diff
Content-Disposition: attachment;
 filename=v06-0006-Remove-the-old-implemenations-of-XLogRead.patch



^ permalink  raw  reply  [nested|flat] 8+ messages in thread

* [PATCH 5/6] Use only xlogreader.c:XLogRead()
@ 2019-09-23 05:40  Antonin Houska <ah@cybertec.at>
  0 siblings, 0 replies; 8+ messages in thread

From: Antonin Houska @ 2019-09-23 05:40 UTC (permalink / raw)

The implementations in xlogutils.c and walsender.c are just renamed now, to be
removed by the following diff.
---
 src/backend/access/transam/xlogreader.c | 157 ++++++++++++++++++++++++
 src/backend/access/transam/xlogutils.c  |  46 ++++++-
 src/backend/replication/walsender.c     | 139 ++++++++++++++++++++-
 src/bin/pg_waldump/pg_waldump.c         |  64 +++++++++-
 src/include/access/xlogreader.h         |  42 +++++++
 5 files changed, 436 insertions(+), 12 deletions(-)

diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c
index 4de5530b3e..7f77fa95cb 100644
--- a/src/backend/access/transam/xlogreader.c
+++ b/src/backend/access/transam/xlogreader.c
@@ -17,6 +17,8 @@
  */
 #include "postgres.h"
 
+#include <unistd.h>
+
 #include "access/transam.h"
 #include "access/xlogrecord.h"
 #include "access/xlog_internal.h"
@@ -27,6 +29,7 @@
 
 #ifndef FRONTEND
 #include "miscadmin.h"
+#include "pgstat.h"
 #include "utils/memutils.h"
 #endif
 
@@ -1015,6 +1018,160 @@ WALOpenSegmentInit(WALOpenSegment *seg, int size)
 #endif
 }
 
+/*
+ * Read 'count' bytes from WAL into 'buf', starting at location 'startptr'. If
+ * tli_p is passed, get the data from timeline *tli_p. 'pos' is the current
+ * position in the XLOG file and openSegment is a callback that opens the next
+ * segment for reading.
+ *
+ * Returns error information if the data could not be read or NULL if
+ * succeeded.
+ *
+ * XXX probably this should be improved to suck data directly from the
+ * WAL buffers when possible.
+ */
+XLogReadError *
+XLogRead(char *buf, XLogRecPtr startptr, Size count,
+		 TimeLineID *tli_p, WALOpenSegment *seg, WALSegmentOpen openSegment)
+{
+	char	   *p;
+	XLogRecPtr	recptr;
+	Size		nbytes;
+
+	p = buf;
+	recptr = startptr;
+	nbytes = count;
+
+	while (nbytes > 0)
+	{
+		int			segbytes;
+		int			readbytes;
+
+		seg->off = XLogSegmentOffset(recptr, seg->size);
+
+		if (seg->file < 0 ||
+			!XLByteInSeg(recptr, seg->num, seg->size) ||
+			(tli_p != NULL && *tli_p != seg->tli))
+		{
+			XLogSegNo	nextSegNo;
+			TimeLineID	tli = InvalidTimeLineID;
+			int			file;
+
+			/* Switch to another logfile segment */
+			if (seg->file >= 0)
+				close(seg->file);
+
+			XLByteToSeg(recptr, nextSegNo, seg->size);
+
+			/* If we have the TLI, let's pass it to the callback. */
+			if (tli_p != NULL)
+				tli = *tli_p;
+
+			/* Open the next segment in the caller's way. */
+			openSegment(nextSegNo, &tli, &file, seg);
+
+			/*
+			 * If we passed InvalidTimeLineID, the callback should have
+			 * determined the correct TLI and returned it.
+			 */
+			Assert(tli != InvalidTimeLineID);
+
+			/* Update the open segment info. */
+			seg->tli = tli;
+			seg->file = file;
+
+			/*
+			 * If the function is called by the XLOG reader, the reader will
+			 * eventually set both "num" and "off". However we need to care
+			 * about them too because the function can also be used directly,
+			 * see walsender.c.
+			 */
+			seg->num = nextSegNo;
+			seg->off = 0;
+		}
+
+		/* How many bytes are within this segment? */
+		if (nbytes > (seg->size - seg->off))
+			segbytes = seg->size - seg->off;
+		else
+			segbytes = nbytes;
+
+#ifndef FRONTEND
+		pgstat_report_wait_start(WAIT_EVENT_WAL_READ);
+#endif
+
+		/*
+		 * Failure to read the data does not necessarily imply non-zero errno.
+		 * Set it to zero so that caller can distinguish the failure that does
+		 * not affect errno.
+		 */
+		errno = 0;
+
+		readbytes = pg_pread(seg->file, p, segbytes, seg->off);
+
+#ifndef FRONTEND
+		pgstat_report_wait_end();
+#endif
+
+		if (readbytes <= 0)
+		{
+			XLogReadError *errinfo;
+
+			errinfo = (XLogReadError *) palloc(sizeof(XLogReadError));
+			errinfo->read_errno = errno;
+			errinfo->readbytes = readbytes;
+			errinfo->reqbytes = segbytes;
+			errinfo->seg = seg;
+
+			return errinfo;
+		}
+
+		/* Update state for read */
+		recptr += readbytes;
+		nbytes -= readbytes;
+		p += readbytes;
+
+		/*
+		 * If the function is called by the XLOG reader, the reader will
+		 * eventually set this field. However we need to care about it too
+		 * because the function can also be used directly (see walsender.c).
+		 */
+		seg->off += readbytes;
+	}
+
+	return NULL;
+}
+
+#ifndef FRONTEND
+/*
+ * Backend-specific convenience code to handle read errors encountered by
+ * XLogRead().
+ */
+void
+XLogReadProcessError(XLogReadError *errinfo)
+{
+	WALOpenSegment *seg = errinfo->seg;
+
+	if (errinfo->readbytes < 0)
+	{
+		errno = errinfo->read_errno;
+		ereport(ERROR,
+				(errcode_for_file_access(),
+				 errmsg("could not read from log segment %s, offset %u, length %zu: %m",
+						XLogFileNameP(seg->tli, seg->num), seg->off,
+						(Size) errinfo->reqbytes)));
+	}
+	else
+	{
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_CORRUPTED),
+				 errmsg("could not read from log segment %s, offset %u: length %zu",
+						XLogFileNameP(seg->tli, seg->num), seg->off,
+						(Size) errinfo->reqbytes)));
+	}
+}
+#endif
+
 /* ----------------------------------------
  * Functions for decoding the data and block references in a record.
  * ----------------------------------------
diff --git a/src/backend/access/transam/xlogutils.c b/src/backend/access/transam/xlogutils.c
index 424bb06919..38c3196168 100644
--- a/src/backend/access/transam/xlogutils.c
+++ b/src/backend/access/transam/xlogutils.c
@@ -653,8 +653,8 @@ XLogTruncateRelation(RelFileNode rnode, ForkNumber forkNum,
  * frontend).  Probably these should be merged at some point.
  */
 static void
-XLogRead(char *buf, int segsize, TimeLineID tli, XLogRecPtr startptr,
-		 Size count)
+XLogReadOld(char *buf, int segsize, TimeLineID tli, XLogRecPtr startptr,
+			Size count)
 {
 	char	   *p;
 	XLogRecPtr	recptr;
@@ -896,6 +896,39 @@ XLogReadDetermineTimeline(XLogReaderState *state, XLogRecPtr wantPage, uint32 wa
 	}
 }
 
+/*
+ * Callback for XLogRead() to open the next segment.
+ */
+static void
+read_local_xlog_page_segment_open(XLogSegNo nextSegNo, TimeLineID *tli_p,
+								  int *file_p, WALOpenSegment *seg)
+{
+	TimeLineID	tli = *tli_p;
+	char		path[MAXPGPATH];
+	int			file;
+
+	Assert(tli != InvalidTimeLineID);
+
+	XLogFilePath(path, tli, nextSegNo, seg->size);
+	file = BasicOpenFile(path, O_RDONLY | PG_BINARY);
+
+	if (file < 0)
+	{
+		if (errno == ENOENT)
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("requested WAL segment %s has already been removed",
+							path)));
+		else
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("could not open file \"%s\": %m",
+							path)));
+	}
+
+	*file_p = file;
+}
+
 /*
  * read_page callback for reading local xlog files
  *
@@ -915,6 +948,7 @@ read_local_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr,
 				loc;
 	int			count;
 	TimeLineID	pageTLI;
+	XLogReadError *errinfo;
 
 	loc = targetPagePtr + reqLen;
 
@@ -1022,10 +1056,10 @@ read_local_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr,
 	 * as 'count', read the whole page anyway. It's guaranteed to be
 	 * zero-padded up to the page boundary if it's incomplete.
 	 */
-	XLogRead(cur_page, state->seg.size, state->seg.tli, targetPagePtr,
-			 XLOG_BLCKSZ);
-	state->seg.tli = pageTLI;
-
+	if ((errinfo = XLogRead(cur_page, targetPagePtr, XLOG_BLCKSZ, &pageTLI,
+							&state->seg,
+							read_local_xlog_page_segment_open)) != NULL)
+		XLogReadProcessError(errinfo);
 	/* number of valid bytes in the buffer */
 	return count;
 }
diff --git a/src/backend/replication/walsender.c b/src/backend/replication/walsender.c
index a617e20ab6..a7b3e0ecbe 100644
--- a/src/backend/replication/walsender.c
+++ b/src/backend/replication/walsender.c
@@ -247,9 +247,12 @@ static void LagTrackerWrite(XLogRecPtr lsn, TimestampTz local_flush_time);
 static TimeOffset LagTrackerRead(int head, XLogRecPtr lsn, TimestampTz now);
 static bool TransactionIdInRecentPast(TransactionId xid, uint32 epoch);
 
-static void XLogRead(char *buf, XLogRecPtr startptr, Size count);
+static void WalSndSegmentOpen(XLogSegNo nextSegNo, TimeLineID *tli_p,
+							  int *file_p, WALOpenSegment *seg);
 
 
+static void XLogReadOld(char *buf, XLogRecPtr startptr, Size count);
+
 /* Initialize walsender process before entering the main command loop */
 void
 InitWalSender(void)
@@ -763,6 +766,7 @@ logical_read_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr, int req
 {
 	XLogRecPtr	flushptr;
 	int			count;
+	XLogReadError *errinfo;
 
 	XLogReadDetermineTimeline(state, targetPagePtr, reqLen);
 	sendTimeLineIsHistoric = (state->currTLI != ThisTimeLineID);
@@ -783,7 +787,13 @@ logical_read_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr, int req
 		count = flushptr - targetPagePtr;	/* part of the page available */
 
 	/* now actually read the data, we know it's there */
-	XLogRead(cur_page, targetPagePtr, XLOG_BLCKSZ);
+	if ((errinfo = XLogRead(cur_page,
+							targetPagePtr,
+							XLOG_BLCKSZ,
+							NULL,	/* WalSndSegmentOpen will determine TLI */
+							sendSeg,
+							WalSndSegmentOpen)) != NULL)
+		XLogReadProcessError(errinfo);
 
 	return count;
 }
@@ -2360,7 +2370,7 @@ WalSndKill(int code, Datum arg)
  * more than one.
  */
 static void
-XLogRead(char *buf, XLogRecPtr startptr, Size count)
+XLogReadOld(char *buf, XLogRecPtr startptr, Size count)
 {
 	char	   *p;
 	XLogRecPtr	recptr;
@@ -2533,6 +2543,81 @@ retry:
 	}
 }
 
+/*
+ * Callback for XLogRead() to open the next segment.
+ */
+void
+WalSndSegmentOpen(XLogSegNo nextSegNo, TimeLineID *tli_p, int *file_p,
+				  WALOpenSegment *seg)
+{
+	TimeLineID	tli = *tli_p;
+	char		path[MAXPGPATH];
+	int			file;
+
+	/*
+	 * The timeline is determined below, caller should not pass it.
+	 */
+	Assert(tli == InvalidTimeLineID);
+
+	/*-------
+	 * When reading from a historic timeline, and there is a timeline switch
+	 * within this segment, read from the WAL segment belonging to the new
+	 * timeline.
+	 *
+	 * For example, imagine that this server is currently on timeline 5, and
+	 * we're streaming timeline 4. The switch from timeline 4 to 5 happened at
+	 * 0/13002088. In pg_wal, we have these files:
+	 *
+	 * ...
+	 * 000000040000000000000012
+	 * 000000040000000000000013
+	 * 000000050000000000000013
+	 * 000000050000000000000014
+	 * ...
+	 *
+	 * In this situation, when requested to send the WAL from segment 0x13, on
+	 * timeline 4, we read the WAL from file 000000050000000000000013. Archive
+	 * recovery prefers files from newer timelines, so if the segment was
+	 * restored from the archive on this server, the file belonging to the old
+	 * timeline, 000000040000000000000013, might not exist. Their contents are
+	 * equal up to the switchpoint, because at a timeline switch, the used
+	 * portion of the old segment is copied to the new file.  -------
+	 */
+	tli = sendTimeLine;
+	if (sendTimeLineIsHistoric)
+	{
+		XLogSegNo	endSegNo;
+
+		XLByteToSeg(sendTimeLineValidUpto, endSegNo, seg->size);
+		if (seg->num == endSegNo)
+			tli = sendTimeLineNextTLI;
+	}
+
+	XLogFilePath(path, tli, nextSegNo, seg->size);
+	file = BasicOpenFile(path, O_RDONLY | PG_BINARY);
+
+	if (file < 0)
+	{
+		/*
+		 * If the file is not found, assume it's because the standby asked for
+		 * a too old WAL segment that has already been removed or recycled.
+		 */
+		if (errno == ENOENT)
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("requested WAL segment %s has already been removed",
+							XLogFileNameP(tli, nextSegNo))));
+		else
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("could not open file \"%s\": %m",
+							path)));
+	}
+
+	*file_p = file;
+	*tli_p = tli;
+}
+
 /*
  * Send out the WAL in its normal physical/stored form.
  *
@@ -2550,6 +2635,8 @@ XLogSendPhysical(void)
 	XLogRecPtr	startptr;
 	XLogRecPtr	endptr;
 	Size		nbytes;
+	XLogSegNo	segno;
+	XLogReadError *errinfo;
 
 	/* If requested switch the WAL sender to the stopping state. */
 	if (got_STOPPING)
@@ -2765,7 +2852,51 @@ XLogSendPhysical(void)
 	 * calls.
 	 */
 	enlargeStringInfo(&output_message, nbytes);
-	XLogRead(&output_message.data[output_message.len], startptr, nbytes);
+
+retry:
+	if ((errinfo = XLogRead(&output_message.data[output_message.len],
+							startptr,
+							nbytes,
+							NULL,	/* WalSndSegmentOpen will determine TLI */
+							sendSeg,
+							WalSndSegmentOpen)) != NULL)
+		XLogReadProcessError(errinfo);
+
+	/*
+	 * After reading into the buffer, check that what we read was valid. We do
+	 * this after reading, because even though the segment was present when we
+	 * opened it, it might get recycled or removed while we read it. The
+	 * read() succeeds in that case, but the data we tried to read might
+	 * already have been overwritten with new WAL records.
+	 */
+	XLByteToSeg(startptr, segno, wal_segment_size);
+	CheckXLogRemoved(segno, ThisTimeLineID);
+
+	/*
+	 * During recovery, the currently-open WAL file might be replaced with the
+	 * file of the same name retrieved from archive. So we always need to
+	 * check what we read was valid after reading into the buffer. If it's
+	 * invalid, we try to open and read the file again.
+	 */
+	if (am_cascading_walsender)
+	{
+		WalSnd	   *walsnd = MyWalSnd;
+		bool		reload;
+
+		SpinLockAcquire(&walsnd->mutex);
+		reload = walsnd->needreload;
+		walsnd->needreload = false;
+		SpinLockRelease(&walsnd->mutex);
+
+		if (reload && sendSeg->file >= 0)
+		{
+			close(sendSeg->file);
+			sendSeg->file = -1;
+
+			goto retry;
+		}
+	}
+
 	output_message.len += nbytes;
 	output_message.data[output_message.len] = '\0';
 
diff --git a/src/bin/pg_waldump/pg_waldump.c b/src/bin/pg_waldump/pg_waldump.c
index a16793bb8b..cd5f589f03 100644
--- a/src/bin/pg_waldump/pg_waldump.c
+++ b/src/bin/pg_waldump/pg_waldump.c
@@ -296,6 +296,51 @@ identify_target_directory(XLogDumpPrivate *private, char *directory,
 		fatal_error("could not find any WAL file");
 }
 
+static void
+WALDumpOpenSegment(XLogSegNo nextSegNo, TimeLineID *tli_p, int *file_p,
+				   WALOpenSegment *seg)
+{
+	TimeLineID	tli = *tli_p;
+	char		fname[MAXPGPATH];
+	int			file;
+	int			tries;
+
+	Assert(tli != InvalidTimeLineID);
+
+	XLogFileName(fname, tli, nextSegNo, seg->size);
+
+	/*
+	 * In follow mode there is a short period of time after the server has
+	 * written the end of the previous file before the new file is available.
+	 * So we loop for 5 seconds looking for the file to appear before giving
+	 * up.
+	 */
+	for (tries = 0; tries < 10; tries++)
+	{
+		file = open_file_in_directory(seg->dir, fname);
+		if (file >= 0)
+			break;
+		if (errno == ENOENT)
+		{
+			int			save_errno = errno;
+
+			/* File not there yet, try again */
+			pg_usleep(500 * 1000);
+
+			errno = save_errno;
+			continue;
+		}
+		/* Any other error, fall through and fail */
+		break;
+	}
+
+	if (file < 0)
+		fatal_error("could not find file \"%s\": %s",
+					fname, strerror(errno));
+
+	*file_p = file;
+}
+
 /*
  * Read count bytes from a segment file in the specified directory, for the
  * given timeline, containing the specified record pointer; store the data in
@@ -427,6 +472,7 @@ XLogDumpReadPage(XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen,
 {
 	XLogDumpPrivate *private = state->private_data;
 	int			count = XLOG_BLCKSZ;
+	XLogReadError *errinfo;
 
 	if (private->endptr != InvalidXLogRecPtr)
 	{
@@ -441,8 +487,22 @@ XLogDumpReadPage(XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen,
 		}
 	}
 
-	XLogDumpXLogRead(private->inpath, private->timeline, targetPagePtr,
-					 readBuff, count);
+	if ((errinfo = XLogRead(readBuff, targetPagePtr, count, &private->timeline,
+							&state->seg, WALDumpOpenSegment)) != NULL)
+	{
+		WALOpenSegment *seg = errinfo->seg;
+		char		fname[MAXPGPATH];
+
+		XLogFileName(fname, seg->tli, seg->num, seg->size);
+
+		if (errno != 0)
+			fatal_error("could not read from log file %s, offset %u, length %zu: %s",
+						fname, seg->off, (Size) errinfo->reqbytes,
+						strerror(errinfo->read_errno));
+		else
+			fatal_error("could not read from log file %s, offset %u: length: %zu",
+						fname, seg->off, (Size) errinfo->reqbytes);
+	}
 
 	return count;
 }
diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h
index b9d99d524e..3d9742b81b 100644
--- a/src/include/access/xlogreader.h
+++ b/src/include/access/xlogreader.h
@@ -227,8 +227,50 @@ extern bool XLogReaderValidatePageHeader(XLogReaderState *state,
 extern XLogRecPtr XLogFindNextRecord(XLogReaderState *state, XLogRecPtr RecPtr);
 #endif							/* FRONTEND */
 
+/*
+ * Callback to open the specified WAL segment for reading.
+ *
+ * "nextSegNo" is the number of the segment to be opened.
+ *
+ * "tli_p" is an input/output argument. If *tli_p is valid, it's the timeline
+ * the new segment should be in. If *tli_p==InvalidTimeLineID, the callback
+ * needs to determine the timeline itself and put the result into *tli_p.
+ *
+ * "file_p" points to an address the segment file descriptor should be stored
+ * at.
+ *
+ * "seg" provides information on the currently open segment. The callback is
+ * not supposed to change this info.
+ *
+ * BasicOpenFile() is the preferred way to open the segment file in backend
+ * code, whereas open(2) should be used in frontend.
+ */
+typedef void (*WALSegmentOpen) (XLogSegNo nextSegNo, TimeLineID *tli_p,
+								int *file_p, WALOpenSegment *seg);
+
 extern void WALOpenSegmentInit(WALOpenSegment *seg, int size);
 
+/*
+ * Error information that both backend and frontend caller can process.
+ *
+ * XXX Should the name be WALReadError? If so, we probably need to rename
+ * XLogRead() and XLogReadProcessError() too.
+ */
+typedef struct XLogReadError
+{
+	int			read_errno;		/* errno set by the last read(). */
+	int			readbytes;		/* Bytes read by the last read(). */
+	int			reqbytes;		/* Bytes requested to be read. */
+	WALOpenSegment *seg;		/* Segment we tried to read from. */
+} XLogReadError;
+
+extern XLogReadError *XLogRead(char *buf, XLogRecPtr startptr, Size count,
+							   TimeLineID *tli_p, WALOpenSegment *seg,
+							   WALSegmentOpen openSegment);
+#ifndef FRONTEND
+void		XLogReadProcessError(XLogReadError *errinfo);
+#endif
+
 /* Functions for decoding an XLogRecord */
 
 extern bool DecodeXLogRecord(XLogReaderState *state, XLogRecord *record,
-- 
2.20.1


--=-=-=
Content-Type: text/x-diff
Content-Disposition: attachment;
 filename=v06-0006-Remove-the-old-implemenations-of-XLogRead.patch



^ permalink  raw  reply  [nested|flat] 8+ messages in thread

* [PATCH 1/2] Use only xlogreader.c:XLogRead()
@ 2019-09-26 11:51  Antonin Houska <ah@cybertec.at>
  0 siblings, 0 replies; 8+ messages in thread

From: Antonin Houska @ 2019-09-26 11:51 UTC (permalink / raw)

The implementations in xlogutils.c and walsender.c are just renamed now, to be
removed by the following diff.
---
 src/backend/access/transam/xlogreader.c | 153 ++++++++++++++++++++++++
 src/backend/access/transam/xlogutils.c  |  45 ++++++-
 src/backend/replication/walsender.c     | 138 ++++++++++++++++++++-
 src/bin/pg_waldump/pg_waldump.c         |  60 +++++++++-
 src/include/access/xlogreader.h         |  47 ++++++++
 5 files changed, 433 insertions(+), 10 deletions(-)

diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c
index c8b0d2303d..a58c5d1633 100644
--- a/src/backend/access/transam/xlogreader.c
+++ b/src/backend/access/transam/xlogreader.c
@@ -17,6 +17,8 @@
  */
 #include "postgres.h"
 
+#include <unistd.h>
+
 #include "access/transam.h"
 #include "access/xlogrecord.h"
 #include "access/xlog_internal.h"
@@ -27,6 +29,7 @@
 
 #ifndef FRONTEND
 #include "miscadmin.h"
+#include "pgstat.h"
 #include "utils/memutils.h"
 #endif
 
@@ -1016,6 +1019,156 @@ out:
 
 #endif							/* FRONTEND */
 
+/*
+ * Read 'count' bytes from WAL into 'buf', starting at location 'startptr'. If
+ * tli_p is passed, get the data from timeline *tli_p. 'pos' is the current
+ * position in the XLOG file and openSegment is a callback that opens the next
+ * segment for reading.
+ *
+ * Returns error information if the data could not be read or NULL if
+ * succeeded.
+ *
+ * XXX probably this should be improved to suck data directly from the
+ * WAL buffers when possible.
+ */
+XLogReadError *
+XLogRead(char *buf, XLogRecPtr startptr, Size count, TimeLineID *tli_p,
+		 WALOpenSegment *seg, WALSegmentContext *segcxt,
+		 WALSegmentOpen openSegment)
+{
+	char	   *p;
+	XLogRecPtr	recptr;
+	Size		nbytes;
+	static XLogReadError errinfo;
+
+	p = buf;
+	recptr = startptr;
+	nbytes = count;
+
+	while (nbytes > 0)
+	{
+		int			segbytes;
+		int			readbytes;
+
+		seg->ws_off = XLogSegmentOffset(recptr, segcxt->ws_segsize);
+
+		if (seg->ws_file < 0 ||
+			!XLByteInSeg(recptr, seg->ws_segno, segcxt->ws_segsize) ||
+			(tli_p != NULL && *tli_p != seg->ws_tli))
+		{
+			XLogSegNo	nextSegNo;
+			TimeLineID	tli;
+			int			file;
+
+			/* Switch to another logfile segment */
+			if (seg->ws_file >= 0)
+				close(seg->ws_file);
+
+			XLByteToSeg(recptr, nextSegNo, segcxt->ws_segsize);
+
+			/*
+			 * If we have the TLI, let's pass it to the callback. If NULL is
+			 * passed, the callback has to find the TLI itself.
+			 */
+			if (tli_p != NULL)
+				tli = *tli_p;
+
+			/* Open the next segment in the caller's way. */
+			openSegment(nextSegNo, &tli, &file, seg, segcxt);
+
+			/* Update the open segment info. */
+			seg->ws_tli = tli;
+			seg->ws_file = file;
+
+			/*
+			 * If the function is called by the XLOG reader, the reader will
+			 * eventually set both "ws_segno" and "ws_off", however the XLOG
+			 * reader is not necessarily involved. Furthermore, we need to set
+			 * the current values for this function to work.
+			 */
+			seg->ws_segno = nextSegNo;
+			seg->ws_off = 0;
+		}
+
+		/* How many bytes are within this segment? */
+		if (nbytes > (segcxt->ws_segsize - seg->ws_off))
+			segbytes = segcxt->ws_segsize - seg->ws_off;
+		else
+			segbytes = nbytes;
+
+#ifndef FRONTEND
+		pgstat_report_wait_start(WAIT_EVENT_WAL_READ);
+#endif
+
+		/*
+		 * Failure to read the data does not necessarily imply non-zero errno.
+		 * Set it to zero so that caller can distinguish the failure that does
+		 * not affect errno.
+		 */
+		errno = 0;
+
+		readbytes = pg_pread(seg->ws_file, p, segbytes, seg->ws_off);
+
+#ifndef FRONTEND
+		pgstat_report_wait_end();
+#endif
+
+		if (readbytes <= 0)
+		{
+			errinfo.read_errno = errno;
+			errinfo.readbytes = readbytes;
+			errinfo.reqbytes = segbytes;
+			errinfo.seg = seg;
+			return &errinfo;
+		}
+
+		/* Update state for read */
+		recptr += readbytes;
+		nbytes -= readbytes;
+		p += readbytes;
+
+		/*
+		 * If the function is called by the XLOG reader, the reader will
+		 * eventually set this field. However we need to care about it too
+		 * because the function can also be used directly (see walsender.c).
+		 */
+		seg->ws_off += readbytes;
+	}
+
+	return NULL;
+}
+
+#ifndef FRONTEND
+/*
+ * Backend-specific convenience code to handle read errors encountered by
+ * XLogRead().
+ */
+void
+XLogReadProcessError(XLogReadError *errinfo)
+{
+	WALOpenSegment *seg = errinfo->seg;
+
+	if (errinfo->readbytes < 0)
+	{
+		errno = errinfo->read_errno;
+		ereport(ERROR,
+				(errcode_for_file_access(),
+				 errmsg("could not read from log segment %s, offset %u, length %zu: %m",
+						XLogFileNameP(seg->ws_tli, seg->ws_segno),
+						seg->ws_off, (Size) errinfo->reqbytes)));
+	}
+	else
+	{
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_CORRUPTED),
+				 errmsg("could not read from log segment %s, offset %u: length %zu",
+						XLogFileNameP(seg->ws_tli, seg->ws_segno),
+						seg->ws_off,
+						(Size) errinfo->reqbytes)));
+	}
+}
+#endif
+
 /* ----------------------------------------
  * Functions for decoding the data and block references in a record.
  * ----------------------------------------
diff --git a/src/backend/access/transam/xlogutils.c b/src/backend/access/transam/xlogutils.c
index 5f1e5ba75d..09d42d3112 100644
--- a/src/backend/access/transam/xlogutils.c
+++ b/src/backend/access/transam/xlogutils.c
@@ -653,8 +653,8 @@ XLogTruncateRelation(RelFileNode rnode, ForkNumber forkNum,
  * frontend).  Probably these should be merged at some point.
  */
 static void
-XLogRead(char *buf, int segsize, TimeLineID tli, XLogRecPtr startptr,
-		 Size count)
+XLogReadOld(char *buf, int segsize, TimeLineID tli, XLogRecPtr startptr,
+			Size count)
 {
 	char	   *p;
 	XLogRecPtr	recptr;
@@ -896,6 +896,38 @@ XLogReadDetermineTimeline(XLogReaderState *state, XLogRecPtr wantPage, uint32 wa
 	}
 }
 
+/*
+ * Callback for XLogRead() to open the next segment.
+ */
+static void
+read_local_xlog_page_segment_open(XLogSegNo nextSegNo, TimeLineID *tli_p,
+								  int *file_p, WALOpenSegment *seg,
+								  WALSegmentContext *segcxt)
+{
+	TimeLineID	tli = *tli_p;
+	char		path[MAXPGPATH];
+	int			file;
+
+	XLogFilePath(path, tli, nextSegNo, segcxt->ws_segsize);
+	file = BasicOpenFile(path, O_RDONLY | PG_BINARY);
+
+	if (file < 0)
+	{
+		if (errno == ENOENT)
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("requested WAL segment %s has already been removed",
+							path)));
+		else
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("could not open file \"%s\": %m",
+							path)));
+	}
+
+	*file_p = file;
+}
+
 /*
  * read_page callback for reading local xlog files
  *
@@ -914,6 +946,7 @@ read_local_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr,
 	XLogRecPtr	read_upto,
 				loc;
 	int			count;
+	XLogReadError *errinfo;
 
 	loc = targetPagePtr + reqLen;
 
@@ -1020,8 +1053,12 @@ read_local_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr,
 	 * as 'count', read the whole page anyway. It's guaranteed to be
 	 * zero-padded up to the page boundary if it's incomplete.
 	 */
-	XLogRead(cur_page, state->segcxt.ws_segsize, state->seg.ws_tli, targetPagePtr,
-			 XLOG_BLCKSZ);
+	if ((errinfo = XLogRead(cur_page, targetPagePtr, XLOG_BLCKSZ,
+							&state->seg.ws_tli,
+							&state->seg,
+							&state->segcxt,
+							read_local_xlog_page_segment_open)) != NULL)
+		XLogReadProcessError(errinfo);
 
 	/* number of valid bytes in the buffer */
 	return count;
diff --git a/src/backend/replication/walsender.c b/src/backend/replication/walsender.c
index eb4a98cc91..dcb84693fc 100644
--- a/src/backend/replication/walsender.c
+++ b/src/backend/replication/walsender.c
@@ -248,9 +248,14 @@ static void LagTrackerWrite(XLogRecPtr lsn, TimestampTz local_flush_time);
 static TimeOffset LagTrackerRead(int head, XLogRecPtr lsn, TimestampTz now);
 static bool TransactionIdInRecentPast(TransactionId xid, uint32 epoch);
 
-static void XLogRead(WALSegmentContext *segcxt, char *buf, XLogRecPtr startptr, Size count);
+static void WalSndSegmentOpen(XLogSegNo nextSegNo, TimeLineID *tli_p,
+							  int *file_p, WALOpenSegment *seg,
+							  WALSegmentContext *segcxt);
 
 
+static void XLogReadOld(WALSegmentContext *segcxt, char *buf,
+						XLogRecPtr startptr, Size count);
+
 /* Initialize walsender process before entering the main command loop */
 void
 InitWalSender(void)
@@ -766,6 +771,8 @@ logical_read_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr, int req
 {
 	XLogRecPtr	flushptr;
 	int			count;
+	XLogReadError *errinfo;
+	XLogSegNo	segno;
 
 	XLogReadDetermineTimeline(state, targetPagePtr, reqLen);
 	sendTimeLineIsHistoric = (state->currTLI != ThisTimeLineID);
@@ -786,7 +793,24 @@ logical_read_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr, int req
 		count = flushptr - targetPagePtr;	/* part of the page available */
 
 	/* now actually read the data, we know it's there */
-	XLogRead(sendCxt, cur_page, targetPagePtr, XLOG_BLCKSZ);
+	if ((errinfo = XLogRead(cur_page,
+							targetPagePtr,
+							XLOG_BLCKSZ,
+							NULL,	/* WalSndSegmentOpen will determine TLI */
+							sendSeg,
+							sendCxt,
+							WalSndSegmentOpen)) != NULL)
+		XLogReadProcessError(errinfo);
+
+	/*
+	 * After reading into the buffer, check that what we read was valid. We do
+	 * this after reading, because even though the segment was present when we
+	 * opened it, it might get recycled or removed while we read it. The
+	 * read() succeeds in that case, but the data we tried to read might
+	 * already have been overwritten with new WAL records.
+	 */
+	XLByteToSeg(targetPagePtr, segno, sendCxt->ws_segsize);
+	CheckXLogRemoved(segno, sendSeg->ws_tli);
 
 	return count;
 }
@@ -2363,7 +2387,7 @@ WalSndKill(int code, Datum arg)
  * more than one.
  */
 static void
-XLogRead(WALSegmentContext *segcxt, char *buf, XLogRecPtr startptr, Size count)
+XLogReadOld(WALSegmentContext *segcxt, char *buf, XLogRecPtr startptr, Size count)
 {
 	char	   *p;
 	XLogRecPtr	recptr;
@@ -2536,6 +2560,71 @@ retry:
 	}
 }
 
+/*
+ * Callback for XLogRead() to open the next segment.
+ */
+void
+WalSndSegmentOpen(XLogSegNo nextSegNo, TimeLineID *tli_p, int *file_p,
+				  WALOpenSegment *seg, WALSegmentContext *segcxt)
+{
+	char		path[MAXPGPATH];
+
+	/*-------
+	 * When reading from a historic timeline, and there is a timeline switch
+	 * within this segment, read from the WAL segment belonging to the new
+	 * timeline.
+	 *
+	 * For example, imagine that this server is currently on timeline 5, and
+	 * we're streaming timeline 4. The switch from timeline 4 to 5 happened at
+	 * 0/13002088. In pg_wal, we have these files:
+	 *
+	 * ...
+	 * 000000040000000000000012
+	 * 000000040000000000000013
+	 * 000000050000000000000013
+	 * 000000050000000000000014
+	 * ...
+	 *
+	 * In this situation, when requested to send the WAL from segment 0x13, on
+	 * timeline 4, we read the WAL from file 000000050000000000000013. Archive
+	 * recovery prefers files from newer timelines, so if the segment was
+	 * restored from the archive on this server, the file belonging to the old
+	 * timeline, 000000040000000000000013, might not exist. Their contents are
+	 * equal up to the switchpoint, because at a timeline switch, the used
+	 * portion of the old segment is copied to the new file.  -------
+	 */
+	*tli_p = sendTimeLine;
+	if (sendTimeLineIsHistoric)
+	{
+		XLogSegNo	endSegNo;
+
+		XLByteToSeg(sendTimeLineValidUpto, endSegNo, segcxt->ws_segsize);
+		if (seg->ws_segno == endSegNo)
+			*tli_p = sendTimeLineNextTLI;
+	}
+
+	XLogFilePath(path, *tli_p, nextSegNo, segcxt->ws_segsize);
+	*file_p = BasicOpenFile(path, O_RDONLY | PG_BINARY);
+
+	if (*file_p < 0)
+	{
+		/*
+		 * If the file is not found, assume it's because the standby asked for
+		 * a too old WAL segment that has already been removed or recycled.
+		 */
+		if (errno == ENOENT)
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("requested WAL segment %s has already been removed",
+							XLogFileNameP(*tli_p, nextSegNo))));
+		else
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("could not open file \"%s\": %m",
+							path)));
+	}
+}
+
 /*
  * Send out the WAL in its normal physical/stored form.
  *
@@ -2553,6 +2642,8 @@ XLogSendPhysical(void)
 	XLogRecPtr	startptr;
 	XLogRecPtr	endptr;
 	Size		nbytes;
+	XLogSegNo	segno;
+	XLogReadError *errinfo;
 
 	/* If requested switch the WAL sender to the stopping state. */
 	if (got_STOPPING)
@@ -2768,7 +2859,46 @@ XLogSendPhysical(void)
 	 * calls.
 	 */
 	enlargeStringInfo(&output_message, nbytes);
-	XLogRead(sendCxt, &output_message.data[output_message.len], startptr, nbytes);
+
+retry:
+	if ((errinfo = XLogRead(&output_message.data[output_message.len],
+							startptr,
+							nbytes,
+							NULL,	/* WalSndSegmentOpen will determine TLI */
+							sendSeg,
+							sendCxt,
+							WalSndSegmentOpen)) != NULL)
+		XLogReadProcessError(errinfo);
+
+	/* See logical_read_xlog_page(). */
+	XLByteToSeg(startptr, segno, sendCxt->ws_segsize);
+	CheckXLogRemoved(segno, sendSeg->ws_tli);
+
+	/*
+	 * During recovery, the currently-open WAL file might be replaced with the
+	 * file of the same name retrieved from archive. So we always need to
+	 * check what we read was valid after reading into the buffer. If it's
+	 * invalid, we try to open and read the file again.
+	 */
+	if (am_cascading_walsender)
+	{
+		WalSnd	   *walsnd = MyWalSnd;
+		bool		reload;
+
+		SpinLockAcquire(&walsnd->mutex);
+		reload = walsnd->needreload;
+		walsnd->needreload = false;
+		SpinLockRelease(&walsnd->mutex);
+
+		if (reload && sendSeg->ws_file >= 0)
+		{
+			close(sendSeg->ws_file);
+			sendSeg->ws_file = -1;
+
+			goto retry;
+		}
+	}
+
 	output_message.len += nbytes;
 	output_message.data[output_message.len] = '\0';
 
diff --git a/src/bin/pg_waldump/pg_waldump.c b/src/bin/pg_waldump/pg_waldump.c
index b79208cd73..c97376101c 100644
--- a/src/bin/pg_waldump/pg_waldump.c
+++ b/src/bin/pg_waldump/pg_waldump.c
@@ -281,6 +281,46 @@ identify_target_directory(char *directory, char *fname)
 	return NULL;				/* not reached */
 }
 
+static void
+WALDumpOpenSegment(XLogSegNo nextSegNo, TimeLineID *tli_p, int *file_p,
+				   WALOpenSegment *seg, WALSegmentContext *segcxt)
+{
+	TimeLineID	tli = *tli_p;
+	char		fname[MAXPGPATH];
+	int			tries;
+
+	XLogFileName(fname, tli, nextSegNo, segcxt->ws_segsize);
+
+	/*
+	 * In follow mode there is a short period of time after the server has
+	 * written the end of the previous file before the new file is available.
+	 * So we loop for 5 seconds looking for the file to appear before giving
+	 * up.
+	 */
+	for (tries = 0; tries < 10; tries++)
+	{
+		*file_p = open_file_in_directory(segcxt->ws_dir, fname);
+		if (*file_p >= 0)
+			break;
+		if (errno == ENOENT)
+		{
+			int			save_errno = errno;
+
+			/* File not there yet, try again */
+			pg_usleep(500 * 1000);
+
+			errno = save_errno;
+			continue;
+		}
+		/* Any other error, fall through and fail */
+		break;
+	}
+
+	if (*file_p < 0)
+		fatal_error("could not find file \"%s\": %s",
+					fname, strerror(errno));
+}
+
 /*
  * Read count bytes from a segment file in the specified directory, for the
  * given timeline, containing the specified record pointer; store the data in
@@ -412,6 +452,7 @@ XLogDumpReadPage(XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen,
 {
 	XLogDumpPrivate *private = state->private_data;
 	int			count = XLOG_BLCKSZ;
+	XLogReadError *errinfo;
 
 	if (private->endptr != InvalidXLogRecPtr)
 	{
@@ -426,8 +467,23 @@ XLogDumpReadPage(XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen,
 		}
 	}
 
-	XLogDumpXLogRead(state->segcxt.ws_dir, private->timeline, targetPagePtr,
-					 readBuff, count);
+	if ((errinfo = XLogRead(readBuff, targetPagePtr, count, &private->timeline,
+							&state->seg, &state->segcxt, WALDumpOpenSegment)) != NULL)
+	{
+		WALOpenSegment *seg = errinfo->seg;
+		char		fname[MAXPGPATH];
+
+		XLogFileName(fname, seg->ws_tli, seg->ws_segno,
+					 state->segcxt.ws_segsize);
+
+		if (errno != 0)
+			fatal_error("could not read from log file %s, offset %u, length %zu: %s",
+						fname, seg->ws_off, (Size) errinfo->reqbytes,
+						strerror(errinfo->read_errno));
+		else
+			fatal_error("could not read from log file %s, offset %u: length: %zu",
+						fname, seg->ws_off, (Size) errinfo->reqbytes);
+	}
 
 	return count;
 }
diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h
index 1bbee386e8..bf4d105de3 100644
--- a/src/include/access/xlogreader.h
+++ b/src/include/access/xlogreader.h
@@ -218,6 +218,31 @@ extern XLogReaderState *XLogReaderAllocate(int wal_segment_size,
 extern void XLogReaderFree(XLogReaderState *state);
 
 /* Initialize supporting structures */
+/*
+ * Callback to open the specified WAL segment for reading.
+ *
+ * "nextSegNo" is the number of the segment to be opened.
+ *
+ * "tli_p" is an input/output argument. If *tli_p is valid, it's the timeline
+ * the new segment should be in. If *tli_p==InvalidTimeLineID, the callback
+ * needs to determine the timeline itself and put the result into *tli_p.
+ *
+ * "file_p" points to an address the segment file descriptor should be stored
+ * at.
+ *
+ * "seg" provides information on the currently open segment. The callback is
+ * not supposed to change this info.
+ *
+ * "segcxt" is additional information about the segment, which logically does
+ * not fit into "seg".
+ *
+ * BasicOpenFile() is the preferred way to open the segment file in backend
+ * code, whereas open(2) should be used in frontend.
+ */
+typedef void (*WALSegmentOpen) (XLogSegNo nextSegNo, TimeLineID *tli_p,
+								int *file_p, WALOpenSegment *seg,
+								WALSegmentContext *segcxt);
+
 extern void WALOpenSegmentInit(WALOpenSegment *seg, WALSegmentContext *segcxt,
 							   int segsize, const char *waldir);
 
@@ -232,6 +257,28 @@ extern bool XLogReaderValidatePageHeader(XLogReaderState *state,
 #ifdef FRONTEND
 extern XLogRecPtr XLogFindNextRecord(XLogReaderState *state, XLogRecPtr RecPtr);
 #endif							/* FRONTEND */
+/*
+ * Error information that both backend and frontend caller can process.
+ *
+ * XXX Should the name be WALReadError? If so, we probably need to rename
+ * XLogRead() and XLogReadProcessError() too.
+ */
+typedef struct XLogReadError
+{
+	int			read_errno;		/* errno set by the last read(). */
+	int			readbytes;		/* Bytes read by the last read(). */
+	int			reqbytes;		/* Bytes requested to be read. */
+	WALOpenSegment *seg;		/* Segment we tried to read from. */
+} XLogReadError;
+
+extern XLogReadError *XLogRead(char *buf, XLogRecPtr startptr,
+							   Size count, TimeLineID *tli_p,
+							   WALOpenSegment *seg, WALSegmentContext *segcxt,
+							   WALSegmentOpen openSegment);
+#ifndef FRONTEND
+void		XLogReadProcessError(XLogReadError *errinfo);
+#endif
+
 /* Functions for decoding an XLogRecord */
 
 extern bool DecodeXLogRecord(XLogReaderState *state, XLogRecord *record,
-- 
2.20.1


--=-=-=
Content-Type: text/x-diff
Content-Disposition: attachment;
 filename=v07-0006-Remove-the-old-implemenations-of-XLogRead.patch



^ permalink  raw  reply  [nested|flat] 8+ messages in thread

* [PATCH 1/2] Use only xlogreader.c:XLogRead()
@ 2019-10-04 10:07  Antonin Houska <ah@cybertec.at>
  0 siblings, 0 replies; 8+ messages in thread

From: Antonin Houska @ 2019-10-04 10:07 UTC (permalink / raw)

The implementations in xlogutils.c and walsender.c are just renamed now, to be
removed by the following diff.
---
 src/backend/access/transam/xlogreader.c | 141 ++++++++++++++++++++++-
 src/backend/access/transam/xlogutils.c  |  61 ++++++++--
 src/backend/replication/walsender.c     | 143 +++++++++++++++++++++++-
 src/bin/pg_waldump/pg_waldump.c         |  60 +++++++++-
 src/include/access/xlogreader.h         |  42 +++++++
 5 files changed, 428 insertions(+), 19 deletions(-)

diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c
index c8b0d2303d..3e2167ca5a 100644
--- a/src/backend/access/transam/xlogreader.c
+++ b/src/backend/access/transam/xlogreader.c
@@ -17,6 +17,8 @@
  */
 #include "postgres.h"
 
+#include <unistd.h>
+
 #include "access/transam.h"
 #include "access/xlogrecord.h"
 #include "access/xlog_internal.h"
@@ -27,6 +29,7 @@
 
 #ifndef FRONTEND
 #include "miscadmin.h"
+#include "pgstat.h"
 #include "utils/memutils.h"
 #endif
 
@@ -626,7 +629,13 @@ ReadPageInternal(XLogReaderState *state, XLogRecPtr pageptr, int reqLen)
 	if (!XLogReaderValidatePageHeader(state, pageptr, (char *) hdr))
 		goto err;
 
-	/* update read state information */
+	/*
+	 * Update read state information.
+	 *
+	 * Note that XLogRead(), if used, should have updated the "seg" too for
+	 * its own reasons, however we cannot rely on ->read_page() to call
+	 * XLogRead().
+	 */
 	state->seg.ws_segno = targetSegNo;
 	state->seg.ws_off = targetPageOff;
 	state->readLen = readLen;
@@ -1016,6 +1025,136 @@ out:
 
 #endif							/* FRONTEND */
 
+/*
+ * Read 'count' bytes from WAL fetched from timeline 'tli' into 'buf',
+ * starting at location 'startptr'. 'seg' is the last segment used,
+ * 'openSegment' is a callback to opens the next segment if needed and
+ * 'segcxt' is additional segment info that does not fit into 'seg'.
+ *
+ * 'errinfo' should point to XLogReadError structure which will receive error
+ * details in case the read fails.
+ *
+ * Returns true if succeeded, false if failed.
+ *
+ * XXX probably this should be improved to suck data directly from the
+ * WAL buffers when possible.
+ */
+bool
+XLogRead(char *buf, XLogRecPtr startptr, Size count, TimeLineID tli,
+		 WALOpenSegment *seg, WALSegmentContext *segcxt,
+		 WALSegmentOpen openSegment, XLogReadError *errinfo)
+{
+	char	   *p;
+	XLogRecPtr	recptr;
+	Size		nbytes;
+
+	p = buf;
+	recptr = startptr;
+	nbytes = count;
+
+	while (nbytes > 0)
+	{
+		int			segbytes;
+		int			readbytes;
+
+		seg->ws_off = XLogSegmentOffset(recptr, segcxt->ws_segsize);
+
+		if (seg->ws_file < 0 ||
+			!XLByteInSeg(recptr, seg->ws_segno, segcxt->ws_segsize) ||
+			tli != seg->ws_tli)
+		{
+			XLogSegNo	nextSegNo;
+
+			/* Switch to another logfile segment */
+			if (seg->ws_file >= 0)
+				close(seg->ws_file);
+
+			XLByteToSeg(recptr, nextSegNo, segcxt->ws_segsize);
+
+			/* Open the next segment in the caller's way. */
+			openSegment(nextSegNo, segcxt, &tli, &seg->ws_file);
+
+			/* Update the current segment info. */
+			seg->ws_tli = tli;
+			seg->ws_segno = nextSegNo;
+			seg->ws_off = 0;
+		}
+
+		/* How many bytes are within this segment? */
+		if (nbytes > (segcxt->ws_segsize - seg->ws_off))
+			segbytes = segcxt->ws_segsize - seg->ws_off;
+		else
+			segbytes = nbytes;
+
+#ifndef FRONTEND
+		pgstat_report_wait_start(WAIT_EVENT_WAL_READ);
+#endif
+
+		/*
+		 * Failure to read the data does not necessarily imply non-zero errno.
+		 * Set it to zero so that caller can distinguish the failure that does
+		 * not affect errno.
+		 */
+		errno = 0;
+
+		readbytes = pg_pread(seg->ws_file, p, segbytes, seg->ws_off);
+
+#ifndef FRONTEND
+		pgstat_report_wait_end();
+#endif
+
+		if (readbytes <= 0)
+		{
+			errinfo->read_errno = errno;
+			errinfo->readbytes = readbytes;
+			errinfo->reqbytes = segbytes;
+			errinfo->seg = seg;
+			return false;
+		}
+
+		/* Update state for read */
+		recptr += readbytes;
+		nbytes -= readbytes;
+		p += readbytes;
+
+		/* Update the current segment info. */
+		seg->ws_off += readbytes;
+	}
+
+	return true;
+}
+
+#ifndef FRONTEND
+/*
+ * Backend-specific convenience code to handle read errors encountered by
+ * XLogRead().
+ */
+void
+XLogReadProcessError(XLogReadError *errinfo)
+{
+	WALOpenSegment *seg = errinfo->seg;
+
+	if (errinfo->readbytes < 0)
+	{
+		errno = errinfo->read_errno;
+		ereport(ERROR,
+				(errcode_for_file_access(),
+				 errmsg("could not read from log segment %s, offset %u, length %zu: %m",
+						XLogFileNameP(seg->ws_tli, seg->ws_segno),
+						seg->ws_off, (Size) errinfo->reqbytes)));
+	}
+	else
+	{
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_CORRUPTED),
+				 errmsg("could not read from log segment %s, offset %u: length %zu",
+						XLogFileNameP(seg->ws_tli, seg->ws_segno),
+						seg->ws_off,
+						(Size) errinfo->reqbytes)));
+	}
+}
+#endif
+
 /* ----------------------------------------
  * Functions for decoding the data and block references in a record.
  * ----------------------------------------
diff --git a/src/backend/access/transam/xlogutils.c b/src/backend/access/transam/xlogutils.c
index 5f1e5ba75d..007974ea99 100644
--- a/src/backend/access/transam/xlogutils.c
+++ b/src/backend/access/transam/xlogutils.c
@@ -653,8 +653,8 @@ XLogTruncateRelation(RelFileNode rnode, ForkNumber forkNum,
  * frontend).  Probably these should be merged at some point.
  */
 static void
-XLogRead(char *buf, int segsize, TimeLineID tli, XLogRecPtr startptr,
-		 Size count)
+XLogReadOld(char *buf, int segsize, TimeLineID tli, XLogRecPtr startptr,
+			Size count)
 {
 	char	   *p;
 	XLogRecPtr	recptr;
@@ -896,6 +896,39 @@ XLogReadDetermineTimeline(XLogReaderState *state, XLogRecPtr wantPage, uint32 wa
 	}
 }
 
+/*
+ * Callback for XLogRead() to open the next segment.
+ */
+static void
+read_local_xlog_page_segment_open(XLogSegNo nextSegNo,
+								  WALSegmentContext *segcxt,
+								  TimeLineID *tli_p,
+								  int *file_p)
+{
+	TimeLineID	tli = *tli_p;
+	char		path[MAXPGPATH];
+	int			file;
+
+	XLogFilePath(path, tli, nextSegNo, segcxt->ws_segsize);
+	file = BasicOpenFile(path, O_RDONLY | PG_BINARY);
+
+	if (file < 0)
+	{
+		if (errno == ENOENT)
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("requested WAL segment %s has already been removed",
+							path)));
+		else
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("could not open file \"%s\": %m",
+							path)));
+	}
+
+	*file_p = file;
+}
+
 /*
  * read_page callback for reading local xlog files
  *
@@ -913,7 +946,9 @@ read_local_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr,
 {
 	XLogRecPtr	read_upto,
 				loc;
+	TimeLineID	tli;
 	int			count;
+	XLogReadError errinfo;
 
 	loc = targetPagePtr + reqLen;
 
@@ -932,7 +967,7 @@ read_local_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr,
 			read_upto = GetFlushRecPtr();
 		else
 			read_upto = GetXLogReplayRecPtr(&ThisTimeLineID);
-		state->seg.ws_tli = ThisTimeLineID;
+		tli = ThisTimeLineID;
 
 		/*
 		 * Check which timeline to get the record from.
@@ -982,14 +1017,14 @@ read_local_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr,
 			read_upto = state->currTLIValidUntil;
 
 			/*
-			 * Setting ws_tli to our wanted record's TLI is slightly wrong;
-			 * the page might begin on an older timeline if it contains a
-			 * timeline switch, since its xlog segment will have been copied
-			 * from the prior timeline. This is pretty harmless though, as
-			 * nothing cares so long as the timeline doesn't go backwards.  We
-			 * should read the page header instead; FIXME someday.
+			 * Setting tli to our wanted record's TLI is slightly wrong; the
+			 * page might begin on an older timeline if it contains a timeline
+			 * switch, since its xlog segment will have been copied from the
+			 * prior timeline. This is pretty harmless though, as nothing
+			 * cares so long as the timeline doesn't go backwards.  We should
+			 * read the page header instead; FIXME someday.
 			 */
-			state->seg.ws_tli = state->currTLI;
+			tli = state->currTLI;
 
 			/* No need to wait on a historical timeline */
 			break;
@@ -1020,8 +1055,10 @@ read_local_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr,
 	 * as 'count', read the whole page anyway. It's guaranteed to be
 	 * zero-padded up to the page boundary if it's incomplete.
 	 */
-	XLogRead(cur_page, state->segcxt.ws_segsize, state->seg.ws_tli, targetPagePtr,
-			 XLOG_BLCKSZ);
+	if (!XLogRead(cur_page, targetPagePtr, XLOG_BLCKSZ, tli, &state->seg,
+				  &state->segcxt, read_local_xlog_page_segment_open,
+				  &errinfo))
+		XLogReadProcessError(&errinfo);
 
 	/* number of valid bytes in the buffer */
 	return count;
diff --git a/src/backend/replication/walsender.c b/src/backend/replication/walsender.c
index eb4a98cc91..b21b143f99 100644
--- a/src/backend/replication/walsender.c
+++ b/src/backend/replication/walsender.c
@@ -248,9 +248,13 @@ static void LagTrackerWrite(XLogRecPtr lsn, TimestampTz local_flush_time);
 static TimeOffset LagTrackerRead(int head, XLogRecPtr lsn, TimestampTz now);
 static bool TransactionIdInRecentPast(TransactionId xid, uint32 epoch);
 
-static void XLogRead(WALSegmentContext *segcxt, char *buf, XLogRecPtr startptr, Size count);
+static void WalSndSegmentOpen(XLogSegNo nextSegNo, WALSegmentContext *segcxt,
+							  TimeLineID *tli_p, int *file_p);
 
 
+static void XLogReadOld(WALSegmentContext *segcxt, char *buf,
+						XLogRecPtr startptr, Size count);
+
 /* Initialize walsender process before entering the main command loop */
 void
 InitWalSender(void)
@@ -766,6 +770,8 @@ logical_read_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr, int req
 {
 	XLogRecPtr	flushptr;
 	int			count;
+	XLogReadError errinfo;
+	XLogSegNo	segno;
 
 	XLogReadDetermineTimeline(state, targetPagePtr, reqLen);
 	sendTimeLineIsHistoric = (state->currTLI != ThisTimeLineID);
@@ -786,7 +792,27 @@ logical_read_xlog_page(XLogReaderState *state, XLogRecPtr targetPagePtr, int req
 		count = flushptr - targetPagePtr;	/* part of the page available */
 
 	/* now actually read the data, we know it's there */
-	XLogRead(sendCxt, cur_page, targetPagePtr, XLOG_BLCKSZ);
+	if (!XLogRead(cur_page,
+				  targetPagePtr,
+				  XLOG_BLCKSZ,
+				  sendSeg->ws_tli,	/* Pass the current TLI because only
+									 * WalSndSegmentOpen controls whether new
+									 * TLI is needed. */
+				  sendSeg,
+				  sendCxt,
+				  WalSndSegmentOpen,
+				  &errinfo))
+		XLogReadProcessError(&errinfo);
+
+	/*
+	 * After reading into the buffer, check that what we read was valid. We do
+	 * this after reading, because even though the segment was present when we
+	 * opened it, it might get recycled or removed while we read it. The
+	 * read() succeeds in that case, but the data we tried to read might
+	 * already have been overwritten with new WAL records.
+	 */
+	XLByteToSeg(targetPagePtr, segno, sendCxt->ws_segsize);
+	CheckXLogRemoved(segno, sendSeg->ws_tli);
 
 	return count;
 }
@@ -2363,7 +2389,7 @@ WalSndKill(int code, Datum arg)
  * more than one.
  */
 static void
-XLogRead(WALSegmentContext *segcxt, char *buf, XLogRecPtr startptr, Size count)
+XLogReadOld(WALSegmentContext *segcxt, char *buf, XLogRecPtr startptr, Size count)
 {
 	char	   *p;
 	XLogRecPtr	recptr;
@@ -2536,6 +2562,71 @@ retry:
 	}
 }
 
+/*
+ * Callback for XLogRead() to open the next segment.
+ */
+void
+WalSndSegmentOpen(XLogSegNo nextSegNo, WALSegmentContext *segcxt,
+				  TimeLineID *tli_p, int *file_p)
+{
+	char		path[MAXPGPATH];
+
+	/*-------
+	 * When reading from a historic timeline, and there is a timeline switch
+	 * within this segment, read from the WAL segment belonging to the new
+	 * timeline.
+	 *
+	 * For example, imagine that this server is currently on timeline 5, and
+	 * we're streaming timeline 4. The switch from timeline 4 to 5 happened at
+	 * 0/13002088. In pg_wal, we have these files:
+	 *
+	 * ...
+	 * 000000040000000000000012
+	 * 000000040000000000000013
+	 * 000000050000000000000013
+	 * 000000050000000000000014
+	 * ...
+	 *
+	 * In this situation, when requested to send the WAL from segment 0x13, on
+	 * timeline 4, we read the WAL from file 000000050000000000000013. Archive
+	 * recovery prefers files from newer timelines, so if the segment was
+	 * restored from the archive on this server, the file belonging to the old
+	 * timeline, 000000040000000000000013, might not exist. Their contents are
+	 * equal up to the switchpoint, because at a timeline switch, the used
+	 * portion of the old segment is copied to the new file.  -------
+	 */
+	*tli_p = sendTimeLine;
+	if (sendTimeLineIsHistoric)
+	{
+		XLogSegNo	endSegNo;
+
+		XLByteToSeg(sendTimeLineValidUpto, endSegNo, segcxt->ws_segsize);
+		if (sendSeg->ws_segno == endSegNo)
+			*tli_p = sendTimeLineNextTLI;
+	}
+
+	XLogFilePath(path, *tli_p, nextSegNo, segcxt->ws_segsize);
+	*file_p = BasicOpenFile(path, O_RDONLY | PG_BINARY);
+
+	if (*file_p < 0)
+	{
+		/*
+		 * If the file is not found, assume it's because the standby asked for
+		 * a too old WAL segment that has already been removed or recycled.
+		 */
+		if (errno == ENOENT)
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("requested WAL segment %s has already been removed",
+							XLogFileNameP(*tli_p, nextSegNo))));
+		else
+			ereport(ERROR,
+					(errcode_for_file_access(),
+					 errmsg("could not open file \"%s\": %m",
+							path)));
+	}
+}
+
 /*
  * Send out the WAL in its normal physical/stored form.
  *
@@ -2553,6 +2644,8 @@ XLogSendPhysical(void)
 	XLogRecPtr	startptr;
 	XLogRecPtr	endptr;
 	Size		nbytes;
+	XLogSegNo	segno;
+	XLogReadError errinfo;
 
 	/* If requested switch the WAL sender to the stopping state. */
 	if (got_STOPPING)
@@ -2768,7 +2861,49 @@ XLogSendPhysical(void)
 	 * calls.
 	 */
 	enlargeStringInfo(&output_message, nbytes);
-	XLogRead(sendCxt, &output_message.data[output_message.len], startptr, nbytes);
+
+retry:
+	if (!XLogRead(&output_message.data[output_message.len],
+				  startptr,
+				  nbytes,
+				  sendSeg->ws_tli,	/* Pass the current TLI because only
+									 * WalSndSegmentOpen controls whether new
+									 * TLI is needed. */
+				  sendSeg,
+				  sendCxt,
+				  WalSndSegmentOpen,
+				  &errinfo))
+		XLogReadProcessError(&errinfo);
+
+	/* See logical_read_xlog_page(). */
+	XLByteToSeg(startptr, segno, sendCxt->ws_segsize);
+	CheckXLogRemoved(segno, sendSeg->ws_tli);
+
+	/*
+	 * During recovery, the currently-open WAL file might be replaced with the
+	 * file of the same name retrieved from archive. So we always need to
+	 * check what we read was valid after reading into the buffer. If it's
+	 * invalid, we try to open and read the file again.
+	 */
+	if (am_cascading_walsender)
+	{
+		WalSnd	   *walsnd = MyWalSnd;
+		bool		reload;
+
+		SpinLockAcquire(&walsnd->mutex);
+		reload = walsnd->needreload;
+		walsnd->needreload = false;
+		SpinLockRelease(&walsnd->mutex);
+
+		if (reload && sendSeg->ws_file >= 0)
+		{
+			close(sendSeg->ws_file);
+			sendSeg->ws_file = -1;
+
+			goto retry;
+		}
+	}
+
 	output_message.len += nbytes;
 	output_message.data[output_message.len] = '\0';
 
diff --git a/src/bin/pg_waldump/pg_waldump.c b/src/bin/pg_waldump/pg_waldump.c
index b79208cd73..4c49d1acf4 100644
--- a/src/bin/pg_waldump/pg_waldump.c
+++ b/src/bin/pg_waldump/pg_waldump.c
@@ -281,6 +281,46 @@ identify_target_directory(char *directory, char *fname)
 	return NULL;				/* not reached */
 }
 
+static void
+WALDumpOpenSegment(XLogSegNo nextSegNo, WALSegmentContext *segcxt,
+				   TimeLineID *tli_p, int *file_p)
+{
+	TimeLineID	tli = *tli_p;
+	char		fname[MAXPGPATH];
+	int			tries;
+
+	XLogFileName(fname, tli, nextSegNo, segcxt->ws_segsize);
+
+	/*
+	 * In follow mode there is a short period of time after the server has
+	 * written the end of the previous file before the new file is available.
+	 * So we loop for 5 seconds looking for the file to appear before giving
+	 * up.
+	 */
+	for (tries = 0; tries < 10; tries++)
+	{
+		*file_p = open_file_in_directory(segcxt->ws_dir, fname);
+		if (*file_p >= 0)
+			break;
+		if (errno == ENOENT)
+		{
+			int			save_errno = errno;
+
+			/* File not there yet, try again */
+			pg_usleep(500 * 1000);
+
+			errno = save_errno;
+			continue;
+		}
+		/* Any other error, fall through and fail */
+		break;
+	}
+
+	if (*file_p < 0)
+		fatal_error("could not find file \"%s\": %s",
+					fname, strerror(errno));
+}
+
 /*
  * Read count bytes from a segment file in the specified directory, for the
  * given timeline, containing the specified record pointer; store the data in
@@ -412,6 +452,7 @@ XLogDumpReadPage(XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen,
 {
 	XLogDumpPrivate *private = state->private_data;
 	int			count = XLOG_BLCKSZ;
+	XLogReadError errinfo;
 
 	if (private->endptr != InvalidXLogRecPtr)
 	{
@@ -426,8 +467,23 @@ XLogDumpReadPage(XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen,
 		}
 	}
 
-	XLogDumpXLogRead(state->segcxt.ws_dir, private->timeline, targetPagePtr,
-					 readBuff, count);
+	if (!XLogRead(readBuff, targetPagePtr, count, private->timeline,
+				  &state->seg, &state->segcxt, WALDumpOpenSegment, &errinfo))
+	{
+		WALOpenSegment *seg = errinfo.seg;
+		char		fname[MAXPGPATH];
+
+		XLogFileName(fname, seg->ws_tli, seg->ws_segno,
+					 state->segcxt.ws_segsize);
+
+		if (errno != 0)
+			fatal_error("could not read from log file %s, offset %u, length %zu: %s",
+						fname, seg->ws_off, (Size) errinfo.reqbytes,
+						strerror(errinfo.read_errno));
+		else
+			fatal_error("could not read from log file %s, offset %u: length: %zu",
+						fname, seg->ws_off, (Size) errinfo.reqbytes);
+	}
 
 	return count;
 }
diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h
index 1bbee386e8..f066b6255d 100644
--- a/src/include/access/xlogreader.h
+++ b/src/include/access/xlogreader.h
@@ -218,6 +218,26 @@ extern XLogReaderState *XLogReaderAllocate(int wal_segment_size,
 extern void XLogReaderFree(XLogReaderState *state);
 
 /* Initialize supporting structures */
+/*
+ * Callback to open the specified WAL segment for reading.
+ *
+ * "nextSegNo" is the number of the segment to be opened.
+ *
+ * "segcxt" is additional information about the segment.
+ *
+ * "tli_p" is an input/output argument. XLogRead() uses it to pass the
+ * timeline in which the new segment should be found, but the callback can use
+ * it to return the TLI that it actually opened.
+ *
+ * "file_p" points to an address the segment file descriptor should be stored
+ * at.
+ *
+ * BasicOpenFile() is the preferred way to open the segment file in backend
+ * code, whereas open(2) should be used in frontend.
+ */
+typedef void (*WALSegmentOpen) (XLogSegNo nextSegNo, WALSegmentContext *segcxt,
+								TimeLineID *tli_p, int *file_p);
+
 extern void WALOpenSegmentInit(WALOpenSegment *seg, WALSegmentContext *segcxt,
 							   int segsize, const char *waldir);
 
@@ -232,6 +252,28 @@ extern bool XLogReaderValidatePageHeader(XLogReaderState *state,
 #ifdef FRONTEND
 extern XLogRecPtr XLogFindNextRecord(XLogReaderState *state, XLogRecPtr RecPtr);
 #endif							/* FRONTEND */
+/*
+ * Error information that both backend and frontend caller can process.
+ *
+ * XXX Should the name be WALReadError? If so, we probably need to rename
+ * XLogRead() and XLogReadProcessError() too.
+ */
+typedef struct XLogReadError
+{
+	int			read_errno;		/* errno set by the last read(). */
+	int			readbytes;		/* Bytes read by the last read(). */
+	int			reqbytes;		/* Bytes requested to be read. */
+	WALOpenSegment *seg;		/* Segment we tried to read from. */
+} XLogReadError;
+
+extern bool XLogRead(char *buf, XLogRecPtr startptr, Size count,
+					 TimeLineID tli, WALOpenSegment *seg,
+					 WALSegmentContext *segcxt, WALSegmentOpen openSegment,
+					 XLogReadError *errinfo);
+#ifndef FRONTEND
+void		XLogReadProcessError(XLogReadError *errinfo);
+#endif
+
 /* Functions for decoding an XLogRecord */
 
 extern bool DecodeXLogRecord(XLogReaderState *state, XLogRecord *record,
-- 
2.20.1


--=-=-=
Content-Type: text/x-diff
Content-Disposition: attachment;
 filename=v08-0006-Remove-the-old-implemenations-of-XLogRead.patch



^ permalink  raw  reply  [nested|flat] 8+ messages in thread

* [PATCH 2/5] Refactor index_concurrently_create_copy() for use with REPACK (CONCURRENTLY).
@ 2026-02-27 18:01  Antonin Houska <ah@cybertec.at>
  0 siblings, 0 replies; 8+ messages in thread

From: Antonin Houska @ 2026-02-27 18:01 UTC (permalink / raw)

This patch moves the code to index_create_copy() and adds a "concurrently"
parameter so it can be used by REPACK (CONCURRENTLY).

With the CONCURRENTLY option, REPACK cannot simply swap the heap file and
rebuild its indexes. Instead, it needs to build a separate set of indexes
(including system catalog entries) *before* the actual swap, to reduce the
time AccessExclusiveLock needs to be held for.
---
 src/backend/catalog/index.c      | 54 +++++++++++++++++++++++---------
 src/backend/commands/indexcmds.c |  6 ++--
 src/backend/nodes/makefuncs.c    |  9 +++---
 src/include/catalog/index.h      |  3 ++
 src/include/nodes/makefuncs.h    |  4 ++-
 5 files changed, 54 insertions(+), 22 deletions(-)

diff --git a/src/backend/catalog/index.c b/src/backend/catalog/index.c
index 5ee6389d39c..f8e6c3d804e 100644
--- a/src/backend/catalog/index.c
+++ b/src/backend/catalog/index.c
@@ -1288,15 +1288,32 @@ index_create(Relation heapRelation,
 /*
  * index_concurrently_create_copy
  *
- * Create concurrently an index based on the definition of the one provided by
- * caller.  The index is inserted into catalogs and needs to be built later
- * on.  This is called during concurrent reindex processing.
- *
- * "tablespaceOid" is the tablespace to use for this index.
+ * Variant of index_create_copy(), called during concurrent reindex
+ * processing.
  */
 Oid
 index_concurrently_create_copy(Relation heapRelation, Oid oldIndexId,
 							   Oid tablespaceOid, const char *newName)
+{
+	return index_create_copy(heapRelation, oldIndexId, tablespaceOid, newName,
+							 true);
+}
+
+/*
+ * index_create_copy
+ *
+ * Create an index based on the definition of the one provided by caller.  The
+ * index is inserted into catalogs. If 'concurrently' is TRUE, it needs to be
+ * built later on, otherwise it's built immediately.
+ *
+ * "tablespaceOid" is the tablespace to use for this index.
+ *
+ * The actual implementation of index_concurrently_create_copy(), reusable for
+ * other purposes.
+ */
+Oid
+index_create_copy(Relation heapRelation, Oid oldIndexId, Oid tablespaceOid,
+				  const char *newName, bool concurrently)
 {
 	Relation	indexRelation;
 	IndexInfo  *oldInfo,
@@ -1315,6 +1332,7 @@ index_concurrently_create_copy(Relation heapRelation, Oid oldIndexId,
 	List	   *indexColNames = NIL;
 	List	   *indexExprs = NIL;
 	List	   *indexPreds = NIL;
+	int			flags = 0;
 
 	indexRelation = index_open(oldIndexId, RowExclusiveLock);
 
@@ -1325,7 +1343,7 @@ index_concurrently_create_copy(Relation heapRelation, Oid oldIndexId,
 	 * Concurrent build of an index with exclusion constraints is not
 	 * supported.
 	 */
-	if (oldInfo->ii_ExclusionOps != NULL)
+	if (oldInfo->ii_ExclusionOps != NULL && concurrently)
 		ereport(ERROR,
 				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
 				 errmsg("concurrent index creation for exclusion constraints is not supported")));
@@ -1381,9 +1399,7 @@ index_concurrently_create_copy(Relation heapRelation, Oid oldIndexId,
 	}
 
 	/*
-	 * Build the index information for the new index.  Note that rebuild of
-	 * indexes with exclusion constraints is not supported, hence there is no
-	 * need to fill all the ii_Exclusion* fields.
+	 * Build the index information for the new index.
 	 */
 	newInfo = makeIndexInfo(oldInfo->ii_NumIndexAttrs,
 							oldInfo->ii_NumIndexKeyAttrs,
@@ -1392,10 +1408,13 @@ index_concurrently_create_copy(Relation heapRelation, Oid oldIndexId,
 							indexPreds,
 							oldInfo->ii_Unique,
 							oldInfo->ii_NullsNotDistinct,
-							false,	/* not ready for inserts */
-							true,
+							!concurrently,	/* isready */
+							concurrently,	/* concurrent */
 							indexRelation->rd_indam->amsummarizing,
-							oldInfo->ii_WithoutOverlaps);
+							oldInfo->ii_WithoutOverlaps,
+							oldInfo->ii_ExclusionOps,
+							oldInfo->ii_ExclusionProcs,
+							oldInfo->ii_ExclusionStrats);
 
 	/*
 	 * Extract the list of column names and the column numbers for the new
@@ -1433,6 +1452,9 @@ index_concurrently_create_copy(Relation heapRelation, Oid oldIndexId,
 		stattargets[i].isnull = isnull;
 	}
 
+	if (concurrently)
+		flags = INDEX_CREATE_SKIP_BUILD | INDEX_CREATE_CONCURRENT;
+
 	/*
 	 * Now create the new index.
 	 *
@@ -1456,7 +1478,7 @@ index_concurrently_create_copy(Relation heapRelation, Oid oldIndexId,
 							  indcoloptions->values,
 							  stattargets,
 							  reloptionsDatum,
-							  INDEX_CREATE_SKIP_BUILD | INDEX_CREATE_CONCURRENT,
+							  flags,
 							  0,
 							  true, /* allow table to be a system catalog? */
 							  false,	/* is_internal? */
@@ -2450,7 +2472,8 @@ BuildIndexInfo(Relation index)
 					   indexStruct->indisready,
 					   false,
 					   index->rd_indam->amsummarizing,
-					   indexStruct->indisexclusion && indexStruct->indisunique);
+					   indexStruct->indisexclusion && indexStruct->indisunique,
+					   NULL, NULL, NULL);
 
 	/* fill in attribute numbers */
 	for (i = 0; i < numAtts; i++)
@@ -2510,7 +2533,8 @@ BuildDummyIndexInfo(Relation index)
 					   indexStruct->indisready,
 					   false,
 					   index->rd_indam->amsummarizing,
-					   indexStruct->indisexclusion && indexStruct->indisunique);
+					   indexStruct->indisexclusion && indexStruct->indisunique,
+					   NULL, NULL, NULL);
 
 	/* fill in attribute numbers */
 	for (i = 0; i < numAtts; i++)
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 635679cc1f2..34209bd1393 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -243,7 +243,8 @@ CheckIndexCompatible(Oid oldId,
 	 */
 	indexInfo = makeIndexInfo(numberOfAttributes, numberOfAttributes,
 							  accessMethodId, NIL, NIL, false, false,
-							  false, false, amsummarizing, isWithoutOverlaps);
+							  false, false, amsummarizing, isWithoutOverlaps,
+							  NULL, NULL, NULL);
 	typeIds = palloc_array(Oid, numberOfAttributes);
 	collationIds = palloc_array(Oid, numberOfAttributes);
 	opclassIds = palloc_array(Oid, numberOfAttributes);
@@ -930,7 +931,8 @@ DefineIndex(ParseState *pstate,
 							  !concurrent,
 							  concurrent,
 							  amissummarizing,
-							  stmt->iswithoutoverlaps);
+							  stmt->iswithoutoverlaps,
+							  NULL, NULL, NULL);
 
 	typeIds = palloc_array(Oid, numberOfAttributes);
 	collationIds = palloc_array(Oid, numberOfAttributes);
diff --git a/src/backend/nodes/makefuncs.c b/src/backend/nodes/makefuncs.c
index 2caec621d73..ca7e21e8349 100644
--- a/src/backend/nodes/makefuncs.c
+++ b/src/backend/nodes/makefuncs.c
@@ -834,7 +834,8 @@ IndexInfo *
 makeIndexInfo(int numattrs, int numkeyattrs, Oid amoid, List *expressions,
 			  List *predicates, bool unique, bool nulls_not_distinct,
 			  bool isready, bool concurrent, bool summarizing,
-			  bool withoutoverlaps)
+			  bool withoutoverlaps, Oid *exclusion_ops, Oid *exclusion_procs,
+			  uint16 *exclusion_strats)
 {
 	IndexInfo  *n = makeNode(IndexInfo);
 
@@ -863,9 +864,9 @@ makeIndexInfo(int numattrs, int numkeyattrs, Oid amoid, List *expressions,
 	n->ii_PredicateState = NULL;
 
 	/* exclusion constraints */
-	n->ii_ExclusionOps = NULL;
-	n->ii_ExclusionProcs = NULL;
-	n->ii_ExclusionStrats = NULL;
+	n->ii_ExclusionOps = exclusion_ops;
+	n->ii_ExclusionProcs = exclusion_procs;
+	n->ii_ExclusionStrats = exclusion_strats;
 
 	/* speculative inserts */
 	n->ii_UniqueOps = NULL;
diff --git a/src/include/catalog/index.h b/src/include/catalog/index.h
index b259c4141ed..3426087b445 100644
--- a/src/include/catalog/index.h
+++ b/src/include/catalog/index.h
@@ -99,6 +99,9 @@ extern Oid	index_concurrently_create_copy(Relation heapRelation,
 										   Oid oldIndexId,
 										   Oid tablespaceOid,
 										   const char *newName);
+extern Oid	index_create_copy(Relation heapRelation, Oid oldIndexId,
+							  Oid tablespaceOid, const char *newName,
+							  bool concurrently);
 
 extern void index_concurrently_build(Oid heapRelationId,
 									 Oid indexRelationId);
diff --git a/src/include/nodes/makefuncs.h b/src/include/nodes/makefuncs.h
index 982ec25ae14..dcea148ae1a 100644
--- a/src/include/nodes/makefuncs.h
+++ b/src/include/nodes/makefuncs.h
@@ -99,7 +99,9 @@ extern IndexInfo *makeIndexInfo(int numattrs, int numkeyattrs, Oid amoid,
 								List *expressions, List *predicates,
 								bool unique, bool nulls_not_distinct,
 								bool isready, bool concurrent,
-								bool summarizing, bool withoutoverlaps);
+								bool summarizing, bool withoutoverlaps,
+								Oid *exclusion_ops, Oid *exclusion_procs,
+								uint16 *exclusion_strats);
 
 extern Node *makeStringConst(char *str, int location);
 extern DefElem *makeDefElem(char *name, Node *arg, int location);
-- 
2.47.3


--=-=-=
Content-Type: text/x-diff
Content-Disposition: attachment;
 filename=v36-0003-Move-conversion-of-a-historic-to-MVCC-snapshot-to-a-.patch



^ permalink  raw  reply  [nested|flat] 8+ messages in thread

end of thread, other threads:[~2026-02-27 18:01 UTC | newest]

Thread overview: 8+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2019-07-09 09:54 [PATCH 4/5] Use only xlogreader.c:XLogRead() Antonin Houska <ah@cybertec.at>
2019-09-09 09:53 [PATCH 3/4] Use only xlogreader.c:XLogRead() Antonin Houska <ah@cybertec.at>
2019-09-09 09:53 [PATCH 3/4] Use only xlogreader.c:XLogRead() Antonin Houska <ah@cybertec.at>
2019-09-23 05:40 [PATCH 5/6] Use only xlogreader.c:XLogRead() Antonin Houska <ah@cybertec.at>
2019-09-23 05:40 [PATCH 5/6] Use only xlogreader.c:XLogRead() Antonin Houska <ah@cybertec.at>
2019-09-26 11:51 [PATCH 1/2] Use only xlogreader.c:XLogRead() Antonin Houska <ah@cybertec.at>
2019-10-04 10:07 [PATCH 1/2] Use only xlogreader.c:XLogRead() Antonin Houska <ah@cybertec.at>
2026-02-27 18:01 [PATCH 2/5] Refactor index_concurrently_create_copy() for use with REPACK (CONCURRENTLY). Antonin Houska <ah@cybertec.at>

This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox