Hello,
Thank you for reviewing this.
At Mon, 28 Aug 2017 20:14:54 +0900, Michael Paquier <michael.paqu...@gmail.com>
wrote in <cab7npqt03+uahxun3ft4ljwndviktgwszdsxiqyndtccfeq...@mail.gmail.com>
> On Mon, Aug 28, 2017 at 8:02 PM, Kyotaro HORIGUCHI
> <horiguchi.kyot...@lab.ntt.co.jp> wrote:
> > The first patch (0001-) fixes this problem, preventing the
> > problematic state of WAL segments by retarding restart LSN of a
> > physical replication slot in a certain condition.
>
> FWIW, I have this patch marked on my list of things to look at, so you
> can count me as a reviewer. There are also some approaches that I
> would like to test because I rely on replication slots for some
> infrastructure. Still...
This test patch modifies the code for easiness. The window for
this bug to occur is from receiving the first record of a segment
to in most cases receiving the second record or after receiving
several records. Intentionally emitting a record spanning two or
more segments would work?
> + if (oldFlushPtr != InvalidXLogRecPtr &&
> + (restartLSN == InvalidXLogRecPtr ?
> + oldFlushPtr / XLOG_SEG_SIZE != flushPtr / XLOG_SEG_SIZE :
> + restartLSN / XLOG_BLCKSZ != flushPtr / XLOG_BLCKSZ))
> I find such code patterns not readable.
Yeah, I agree. I rewrote there and the result in the attached
patch is far cleaner than the blob.
regards,
--
Kyotaro Horiguchi
NTT Open Source Software Center
>From c15473998467a17ac6070ed68405a14ca0f98ae6 Mon Sep 17 00:00:00 2001
From: Kyotaro Horiguchi <horiguchi.kyot...@lab.ntt.co.jp>
Date: Wed, 1 Feb 2017 16:07:22 +0900
Subject: [PATCH 1/2] Retard restart LSN of a slot when a segment starts with a
contrecord.
A physical-replication standby can stop just at the boundary of WAL
segments. restart_lsn of a slot on the master can be assumed to be the
same location. The last segment on the master will be removed after
some checkpoints for the case. If the last record of the last
replicated segment continues to the next segment, the continuation
record is only on the master. The standby cannot start in the case
because the split record is not available from only one source.
This patch detains restart_lsn in the last sgement when the first page
of the next segment is a continuation record.
---
src/backend/replication/walsender.c | 123 ++++++++++++++++++++++++++++++++++--
1 file changed, 116 insertions(+), 7 deletions(-)
diff --git a/src/backend/replication/walsender.c b/src/backend/replication/walsender.c
index db346e6..30e222c 100644
--- a/src/backend/replication/walsender.c
+++ b/src/backend/replication/walsender.c
@@ -217,6 +217,13 @@ static struct
WalTimeSample last_read[NUM_SYNC_REP_WAIT_MODE];
} LagTracker;
+/*
+ * This variable corresponds to restart_lsn in pg_replication_slots for a
+ * physical slot. This has a valid value only when it differs from the current
+ * flush pointer.
+ */
+static XLogRecPtr restartLSN = InvalidXLogRecPtr;
+
/* Signal handlers */
static void WalSndLastCycleHandler(SIGNAL_ARGS);
@@ -251,7 +258,7 @@ 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 bool XLogRead(char *buf, XLogRecPtr startptr, Size count, bool noutfoundok);
/* Initialize walsender process before entering the main command loop */
@@ -546,6 +553,9 @@ StartReplication(StartReplicationCmd *cmd)
ereport(ERROR,
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
(errmsg("cannot use a logical replication slot for physical replication"))));
+
+ /* Restore restartLSN from replication slot */
+ restartLSN = MyReplicationSlot->data.restart_lsn;
}
/*
@@ -561,6 +571,10 @@ StartReplication(StartReplicationCmd *cmd)
else
FlushPtr = GetFlushRecPtr();
+ /* Set InvalidXLogRecPtr if catching up */
+ if (restartLSN == FlushPtr)
+ restartLSN = InvalidXLogRecPtr;
+
if (cmd->timeline != 0)
{
XLogRecPtr switchpoint;
@@ -770,7 +784,7 @@ 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);
+ XLogRead(cur_page, targetPagePtr, XLOG_BLCKSZ, false);
return count;
}
@@ -1738,7 +1752,7 @@ static void
ProcessStandbyReplyMessage(void)
{
XLogRecPtr writePtr,
- flushPtr,
+ flushPtr, oldFlushPtr,
applyPtr;
bool replyRequested;
TimeOffset writeLag,
@@ -1798,6 +1812,7 @@ ProcessStandbyReplyMessage(void)
WalSnd *walsnd = MyWalSnd;
SpinLockAcquire(&walsnd->mutex);
+ oldFlushPtr = walsnd->flush;
walsnd->write = writePtr;
walsnd->flush = flushPtr;
walsnd->apply = applyPtr;
@@ -1821,7 +1836,93 @@ ProcessStandbyReplyMessage(void)
if (SlotIsLogical(MyReplicationSlot))
LogicalConfirmReceivedLocation(flushPtr);
else
- PhysicalConfirmReceivedLocation(flushPtr);
+ {
+ /*
+ * Recovery on standby requires that a continuation record is
+ * available from single WAL source. For the reason, physical
+ * replication slot should stay in the first segment of the
+ * multiple segments that a continued record is spanning
+ * over. Since we look pages and don't look into individual record
+ * here, restartLSN may stay a bit too behind but it doesn't
+ * matter.
+ *
+ * Since the objective is avoiding to remove required segments,
+ * checking at the beginning of every segment is enough. But once
+ * restartLSN goes behind, check every page for quick restoration.
+ *
+ * restartLSN has a valid value only when it is behind flushPtr.
+ */
+ bool check_contrecords = false;
+
+ if (restartLSN != InvalidXLogRecPtr)
+ {
+ /*
+ * We're retarding restartLSN, check continuation records
+ * at every page boundary for quick restoration.
+ */
+ if (restartLSN / XLOG_BLCKSZ != flushPtr / XLOG_BLCKSZ)
+ check_contrecords = true;
+ }
+ else if (oldFlushPtr != InvalidXLogRecPtr)
+ {
+ /*
+ * We're not retarding restartLSN , check continuation records
+ * only at segment boundaries. No need to check if
+ */
+ if (oldFlushPtr / XLOG_SEG_SIZE != flushPtr / XLOG_SEG_SIZE)
+ check_contrecords = true;
+ }
+
+ if (check_contrecords)
+ {
+ XLogRecPtr rp;
+
+ if (restartLSN == InvalidXLogRecPtr)
+ restartLSN = oldFlushPtr;
+
+ rp = restartLSN - (restartLSN % XLOG_BLCKSZ);
+
+ /*
+ * We may have let the record at flushPtr be sent, so it's
+ * worth looking
+ */
+ while (rp <= flushPtr)
+ {
+ XLogPageHeaderData header;
+
+ /*
+ * If the page header is not available for now, don't move
+ * restartLSN forward. We can read it by the next chance.
+ */
+ if(sentPtr - rp >= sizeof(XLogPageHeaderData))
+ {
+ bool found;
+ /*
+ * Fetch the page header of the next page. Move
+ * restartLSN forward only if it is not a continuation
+ * page.
+ */
+ found = XLogRead((char *)&header, rp,
+ sizeof(XLogPageHeaderData), true);
+ if (found &&
+ (header.xlp_info & XLP_FIRST_IS_CONTRECORD) == 0)
+ restartLSN = rp;
+ }
+ rp += XLOG_BLCKSZ;
+ }
+
+ /*
+ * If restartLSN is on the same page with flushPtr, it means
+ * that we are catching up.
+ */
+ if (restartLSN / XLOG_BLCKSZ == flushPtr / XLOG_BLCKSZ)
+ restartLSN = InvalidXLogRecPtr;
+ }
+
+ /* restartLSN == InvalidXLogRecPtr means catching up */
+ PhysicalConfirmReceivedLocation(restartLSN != InvalidXLogRecPtr ?
+ restartLSN : flushPtr);
+ }
}
}
@@ -2288,6 +2389,7 @@ WalSndKill(int code, Datum arg)
/*
* Read 'count' bytes from WAL into 'buf', starting at location 'startptr'
+ * Returns false if the segment file is not found when notfoundok is true.
*
* XXX probably this should be improved to suck data directly from the
* WAL buffers when possible.
@@ -2297,8 +2399,8 @@ WalSndKill(int code, Datum arg)
* always be one descriptor left open until the process ends, but never
* more than one.
*/
-static void
-XLogRead(char *buf, XLogRecPtr startptr, Size count)
+static bool
+XLogRead(char *buf, XLogRecPtr startptr, Size count, bool notfoundok)
{
char *p;
XLogRecPtr recptr;
@@ -2375,10 +2477,15 @@ retry:
* removed or recycled.
*/
if (errno == ENOENT)
+ {
+ if (notfoundok)
+ return false;
+
ereport(ERROR,
(errcode_for_file_access(),
errmsg("requested WAL segment %s has already been removed",
XLogFileNameP(curFileTimeLine, sendSegNo))));
+ }
else
ereport(ERROR,
(errcode_for_file_access(),
@@ -2460,6 +2567,8 @@ retry:
goto retry;
}
}
+
+ return true;
}
/*
@@ -2694,7 +2803,7 @@ XLogSendPhysical(void)
* calls.
*/
enlargeStringInfo(&output_message, nbytes);
- XLogRead(&output_message.data[output_message.len], startptr, nbytes);
+ XLogRead(&output_message.data[output_message.len], startptr, nbytes, false);
output_message.len += nbytes;
output_message.data[output_message.len] = '\0';
--
2.9.2
--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers
