Hi,

There's a bug in ProcArrayApplyRecoveryInfo, introduced by 8431e296ea, which may cause failures when starting a replica, making it unusable. The commit message for 8431e296ea is not very clear about what exactly is being done and why, but the root cause is that at while processing RUNNING_XACTS, the XIDs are sorted like this:

    /*
     * Sort the array so that we can add them safely into
     * KnownAssignedXids.
     */
    qsort(xids, nxids, sizeof(TransactionId), xidComparator);

where "safely" likely means "not violating the ordering expected by KnownAssignedXidsAdd". Unfortunately, xidComparator compares the values as plain uint32 values, while KnownAssignedXidsAdd actually calls TransactionIdFollowsOrEquals() and compares the logical XIDs :-(

Triggering this is pretty simple - all you need is two transactions with XIDs before/after the 4B limit, and then (re)start a replica. The replica refuses to start with a message like this:

    LOG:  9 KnownAssignedXids (num=4 tail=0 head=4) [0]=32705 [1]=32706
          [2]=32707 [3]=32708
    CONTEXT:  WAL redo at 0/6000120 for Standby/RUNNING_XACTS: nextXid
              32715 latestCompletedXid 32714 oldestRunningXid
              4294967001; 8 xacts: 32708 32707 32706 32705 4294967009
              4294967008 4294967007 4294967006
    FATAL:  out-of-order XID insertion in KnownAssignedXids

Clearly, we add the 4 "younger" XIDs first (because that's what the XID comparator does), but then KnownAssignedXidsAdd thinks there's some sort of corruption because logically 4294967006 is older.

This does not affect replicas in STANDBY_SNAPSHOT_READY state, because in that case ProcArrayApplyRecoveryInfo ignores RUNNING_XACTS messages.


The probability of hitting this in practice is proportional to how long you leave transactions running. The system where we observed this leaves transactions with XIDs open for days, and the age may be ~40M. Intuitivelly, that's ~40M/4B (=1%) probability that at any given time there are transactions with contradicting ordering. So most restarts worked fine, until one that happened at just the "right" time.

This likely explains why we never got any reports about this - most systems probably don't leave transactions running for this long, so the probability is much lower. And replica restarts are generally not that common events either.

Attached patch is fixing this by just sorting the XIDs logically. The xidComparator is meant for places that can't do logical ordering. But these XIDs come from RUNNING_XACTS, so they actually come from the same wraparound epoch (so sorting logically seems perfectly fine).


regards

--
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
From c96f5d64f977bf252451657cfe80d603869a2c1e Mon Sep 17 00:00:00 2001
From: Tomas Vondra <tomas.von...@postgresql.org>
Date: Sun, 23 Jan 2022 01:00:29 +0100
Subject: [PATCH] Fix ordering of XIDs in ProcArrayApplyRecoveryInfo

Commit 8431e296ea reworked ProcArrayApplyRecoveryInfo to sort XIDs
before adding them to KnownAssignedXids, but it used xidComparator for
that purpose, which simply compares the values as uint32 values. This
may easily confuse KnownAssignedXidsAdd() which enforces the logical
ordering by calling TransactionIdFollowsOrEquals() etc.

Hitting this issue is fairly easy - you just need two transactons, one
started before the 4B limit (e.g. XID 4294967290), the other sometime
after it (e.g. XID 1000). Logically (4294967290 <= 1000) but when
compared using xidComparator we try to add them in the opposite order.
Which makes KnownAssignedXidsAdd() fail with an error like this:

  ERROR: out-of-order XID insertion in KnownAssignedXids

This however only happens during replica initialization - so you have to
restart (or create) a replica while there are such transactions with
contradictory XID ordering. Long-running transactions naturally increase
the likelihood of hitting this issue. Once the replica gets into this
state, it can't be started (even if the old transactions are terminated
in some way).

Fixed by sorting the XIDs logically - this is fine because we're dealing
with normal XIDs (because it's XIDs assigned to backends) and from the
same wraparound epoch (otherwise the backends could not be running at
the same time on the primary node). So there are no problems with the
triangle inequality, which is why xidComparator compares raw values.

Patch by me. Investigation and root cause analysis by Abhijit Menon-Sen.

This issue is present in all releases since 9.4, however releases up to
9.6 are EOL already so backpatch to 10 only.

Reviewed-by: Abhijit Menon-Sen
Reviewed-by: Alvaro Herrera
Backpatch-through: 10
---
 src/backend/storage/ipc/procarray.c |  7 ++++++-
 src/backend/utils/adt/xid.c         | 26 ++++++++++++++++++++++++++
 src/include/utils/builtins.h        |  1 +
 3 files changed, 33 insertions(+), 1 deletion(-)

diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index 3be60402890..9d3efb7d80e 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -1164,8 +1164,13 @@ ProcArrayApplyRecoveryInfo(RunningTransactions running)
 		/*
 		 * Sort the array so that we can add them safely into
 		 * KnownAssignedXids.
+		 *
+		 * We have to sort them logically, because in KnownAssignedXidsAdd we
+		 * call TransactionIdFollowsOrEquals and so on. But we know these XIDs
+		 * come from RUNNING_XACTS, which means there are only normal XIDs from
+		 * the same epoch, so this is safe.
 		 */
-		qsort(xids, nxids, sizeof(TransactionId), xidComparator);
+		qsort(xids, nxids, sizeof(TransactionId), xidLogicalComparator);
 
 		/*
 		 * Add the sorted snapshot into KnownAssignedXids.  The running-xacts
diff --git a/src/backend/utils/adt/xid.c b/src/backend/utils/adt/xid.c
index e6633e9cffe..9b4ceaea47f 100644
--- a/src/backend/utils/adt/xid.c
+++ b/src/backend/utils/adt/xid.c
@@ -145,6 +145,32 @@ xidComparator(const void *arg1, const void *arg2)
 	return 0;
 }
 
+/*
+ * xidLogicalComparator
+ *		qsort comparison function for XIDs
+ *
+ * This is used to compare only XIDs from the same epoch (e.g. for backends
+ * running at the same time). So there must be only normal XIDs, so there's
+ * no issue with triangle inequality.
+ */
+int
+xidLogicalComparator(const void *arg1, const void *arg2)
+{
+	TransactionId xid1 = *(const TransactionId *) arg1;
+	TransactionId xid2 = *(const TransactionId *) arg2;
+
+	Assert(TransactionIdIsNormal(xid1));
+	Assert(TransactionIdIsNormal(xid2));
+
+	if (TransactionIdPrecedes(xid1, xid2))
+		return -1;
+
+	if (TransactionIdPrecedes(xid2, xid1))
+		return 1;
+
+	return 0;
+}
+
 Datum
 xid8toxid(PG_FUNCTION_ARGS)
 {
diff --git a/src/include/utils/builtins.h b/src/include/utils/builtins.h
index 7ac4780e3fc..d8f05a7df3a 100644
--- a/src/include/utils/builtins.h
+++ b/src/include/utils/builtins.h
@@ -87,6 +87,7 @@ extern void text_to_cstring_buffer(const text *src, char *dst, size_t dst_len);
 
 /* xid.c */
 extern int	xidComparator(const void *arg1, const void *arg2);
+extern int	xidLogicalComparator(const void *arg1, const void *arg2);
 
 /* inet_cidr_ntop.c */
 extern char *pg_inet_cidr_ntop(int af, const void *src, int bits,
-- 
2.31.1

Reply via email to