rdelaage
/
postgresql
mirror of https://git.postgresql.org/git/postgresql.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Fix catalog lookup with the wrong snapshot during logical decoding. 

Previously, we relied on HEAP2_NEW_CID records and XACT_INVALIDATION
records to know if the transaction has modified the catalog, and that
information is not serialized to snapshot. Therefore, after the restart,
if the logical decoding decodes only the commit record of the transaction
that has actually modified a catalog, we will miss adding its XID to the
snapshot. Thus, we will end up looking at catalogs with the wrong
snapshot.

To fix this problem, this changes the snapshot builder so that it
remembers the last-running-xacts list of the decoded RUNNING_XACTS record
after restoring the previously serialized snapshot. Then, we mark the
transaction as containing catalog changes if it's in the list of initial
running transactions and its commit record has XACT_XINFO_HAS_INVALS. To
avoid ABI breakage, we store the array of the initial running transactions
in the static variables InitialRunningXacts and NInitialRunningXacts,
instead of storing those in SnapBuild or ReorderBuffer.

This approach has a false positive; we could end up adding the transaction
that didn't change catalog to the snapshot since we cannot distinguish
whether the transaction has catalog changes only by checking the COMMIT
record. It doesn't have the information on which (sub) transaction has
catalog changes, and XACT_XINFO_HAS_INVALS doesn't necessarily indicate
that the transaction has catalog change. But that won't be a problem since
we use snapshot built during decoding only to read system catalogs.

On the master branch, we took a more future-proof approach by writing
catalog modifying transactions to the serialized snapshot which avoids the
above false positive. But we cannot backpatch it because of a change in
the SnapBuild.

Reported-by: Mike Oh
Author: Masahiko Sawada
Reviewed-by: Amit Kapila, Shi yu, Takamichi Osumi, Kyotaro Horiguchi, Bertrand Drouvot, Ahsan Hadi
Backpatch-through: 10
Discussion: https://postgr.es/m/81D0D8B0-E7C4-4999-B616-1E5004DBDCD2%40amazon.com
This commit is contained in:
Amit Kapila 2022-08-11 09:45:04 +05:30
parent 95bfadd4e8
commit 68dcce247f
6 changed files with 232 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
contrib/test_decoding/Makefile
View File

@ -8,7 +8,7 @@ REGRESS = ddl xact rewrite toast permissions decoding_in_xact \
spill slot truncate stream stats twophase twophase_stream
ISOLATION = mxact delayed_startup ondisk_startup concurrent_ddl_dml \
oldest_xmin snapshot_transfer subxact_without_top concurrent_stream \
twophase_snapshot
twophase_snapshot catalog_change_snapshot
REGRESS_OPTS = --temp-config $(top_srcdir)/contrib/test_decoding/logical.conf
ISOLATION_OPTS = --temp-config $(top_srcdir)/contrib/test_decoding/logical.conf

44
contrib/test_decoding/expected/catalog_change_snapshot.out Normal file
View File

@ -0,0 +1,44 @@
Parsed test spec with 2 sessions
starting permutation: s0_init s0_begin s0_savepoint s0_truncate s1_checkpoint s1_get_changes s0_commit s0_begin s0_insert s1_checkpoint s1_get_changes s0_commit s1_get_changes
step s0_init: SELECT 'init' FROM pg_create_logical_replication_slot('isolation_slot', 'test_decoding');
?column?
--------
init
(1 row)
step s0_begin: BEGIN;
step s0_savepoint: SAVEPOINT sp1;
step s0_truncate: TRUNCATE tbl1;
step s1_checkpoint: CHECKPOINT;
step s1_get_changes: SELECT data FROM pg_logical_slot_get_changes('isolation_slot', NULL, NULL, 'skip-empty-xacts', '1', 'include-xids', '0');
data
----
(0 rows)
step s0_commit: COMMIT;
step s0_begin: BEGIN;
step s0_insert: INSERT INTO tbl1 VALUES (1);
step s1_checkpoint: CHECKPOINT;
step s1_get_changes: SELECT data FROM pg_logical_slot_get_changes('isolation_slot', NULL, NULL, 'skip-empty-xacts', '1', 'include-xids', '0');
data
---------------------------------------
BEGIN
table public.tbl1: TRUNCATE: (no-flags)
COMMIT
(3 rows)
step s0_commit: COMMIT;
step s1_get_changes: SELECT data FROM pg_logical_slot_get_changes('isolation_slot', NULL, NULL, 'skip-empty-xacts', '1', 'include-xids', '0');
data
-------------------------------------------------------------
BEGIN
table public.tbl1: INSERT: val1[integer]:1 val2[integer]:null
COMMIT
(3 rows)
?column?
--------
stop
(1 row)

39
contrib/test_decoding/specs/catalog_change_snapshot.spec Normal file
View File

@ -0,0 +1,39 @@
# Test decoding only the commit record of the transaction that have
# modified catalogs.
setup
{
DROP TABLE IF EXISTS tbl1;
CREATE TABLE tbl1 (val1 integer, val2 integer);
}
teardown
{
DROP TABLE tbl1;
SELECT 'stop' FROM pg_drop_replication_slot('isolation_slot');
}
session "s0"
setup { SET synchronous_commit=on; }
step "s0_init" { SELECT 'init' FROM pg_create_logical_replication_slot('isolation_slot', 'test_decoding'); }
step "s0_begin" { BEGIN; }
step "s0_savepoint" { SAVEPOINT sp1; }
step "s0_truncate" { TRUNCATE tbl1; }
step "s0_insert" { INSERT INTO tbl1 VALUES (1); }
step "s0_commit" { COMMIT; }
session "s1"
setup { SET synchronous_commit=on; }
step "s1_checkpoint" { CHECKPOINT; }
step "s1_get_changes" { SELECT data FROM pg_logical_slot_get_changes('isolation_slot', NULL, NULL, 'skip-empty-xacts', '1', 'include-xids', '0'); }
# For the transaction that TRUNCATEd the table tbl1, the last decoding decodes
# only its COMMIT record, because it starts from the RUNNING_XACTS record emitted
# during the first checkpoint execution. This transaction must be marked as
# containing catalog changes while decoding the COMMIT record and the decoding
# of the INSERT record must read the pg_class with the correct historic snapshot.
#
# Note that in a case where bgwriter wrote the RUNNING_XACTS record between "s0_commit"
# and "s0_begin", this doesn't happen as the decoding starts from the RUNNING_XACTS
# record written by bgwriter. One might think we can either stop the bgwriter or
# increase LOG_SNAPSHOT_INTERVAL_MS but it's not practical via tests.
permutation "s0_init" "s0_begin" "s0_savepoint" "s0_truncate" "s1_checkpoint" "s1_get_changes" "s0_commit" "s0_begin" "s0_insert" "s1_checkpoint" "s1_get_changes" "s0_commit" "s1_get_changes"

15
src/backend/replication/logical/decode.c
View File

@ -691,6 +691,21 @@ DecodeCommit(LogicalDecodingContext *ctx, XLogRecordBuffer *buf,
commit_time = parsed->origin_timestamp;
}
/*
* If the COMMIT record has invalidation messages, it could have catalog
* changes. It is possible that we didn't mark this transaction as
* containing catalog changes when the decoding starts from a commit
* record without decoding the transaction's other changes. So, we ensure
* to mark such transactions as containing catalog change.
*
* This must be done before SnapBuildCommitTxn() so that we can include
* these transactions in the historic snapshot.
*/
if (parsed->xinfo & XACT_XINFO_HAS_INVALS)
SnapBuildXidSetCatalogChanges(ctx->snapshot_builder, xid,
parsed->nsubxacts, parsed->subxacts,
buf->origptr);
SnapBuildCommitTxn(ctx->snapshot_builder, buf->origptr, xid,
parsed->nsubxacts, parsed->subxacts);

137
src/backend/replication/logical/snapbuild.c
View File

@ -250,8 +250,38 @@ struct SnapBuild
static ResourceOwner SavedResourceOwnerDuringExport = NULL;
static bool ExportInProgress = false;
/* ->committed manipulation */
static void SnapBuildPurgeCommittedTxn(SnapBuild *builder);
/*
* Array of transactions and subtransactions that were running when
* the xl_running_xacts record that we decoded was written. The array is
* sorted in xidComparator order. We remove xids from this array when
* they become old enough to matter, and then it eventually becomes empty.
* This array is allocated in builder->context so its lifetime is the same
* as the snapshot builder.
*
* We normally rely on some WAL record types such as HEAP2_NEW_CID to know
* if the transaction has changed the catalog. But it could happen that the
* logical decoding decodes only the commit record of the transaction after
* restoring the previously serialized snapshot in which case we will miss
* adding the xid to the snapshot and end up looking at the catalogs with the
* wrong snapshot.
*
* Now to avoid the above problem, if the COMMIT record of the xid listed in
* InitialRunningXacts has XACT_XINFO_HAS_INVALS flag, we mark both the top
* transaction and its substransactions as containing catalog changes.
*
* We could end up adding the transaction that didn't change catalog
* to the snapshot since we cannot distinguish whether the transaction
* has catalog changes only by checking the COMMIT record. It doesn't
* have the information on which (sub) transaction has catalog changes,
* and XACT_XINFO_HAS_INVALS doesn't necessarily indicate that the
* transaction has catalog change. But that won't be a problem since we
* use snapshot built during decoding only for reading system catalogs.
*/
static TransactionId *InitialRunningXacts = NULL;
static int NInitialRunningXacts = 0;
/* ->committed and InitailRunningXacts manipulation */
static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
/* snapshot building/manipulation/distribution functions */
static Snapshot SnapBuildBuildSnapshot(SnapBuild *builder);
@ -879,12 +909,17 @@ SnapBuildAddCommittedTxn(SnapBuild *builder, TransactionId xid)
}
/*
* Remove knowledge about transactions we treat as committed that are smaller
* than ->xmin. Those won't ever get checked via the ->committed array but via
* the clog machinery, so we don't need to waste memory on them.
* Remove knowledge about transactions we treat as committed and the initial
* running transactions that are smaller than ->xmin. Those won't ever get
* checked via the ->committed or InitialRunningXacts array, respectively.
* The committed xids will get checked via the clog machinery.
*
* We can ideally remove the transaction from InitialRunningXacts array
* once it is finished (committed/aborted) but that could be costly as we need
* to maintain the xids order in the array.
*/
static void
SnapBuildPurgeCommittedTxn(SnapBuild *builder)
SnapBuildPurgeOlderTxn(SnapBuild *builder)
{
int off;
TransactionId *workspace;
@ -919,6 +954,49 @@ SnapBuildPurgeCommittedTxn(SnapBuild *builder)
builder->committed.xcnt = surviving_xids;
pfree(workspace);
/* Quick exit if there is no initial running transactions */
if (NInitialRunningXacts == 0)
return;
/* bound check if there is at least one transaction to remove */
if (!NormalTransactionIdPrecedes(InitialRunningXacts[0],
builder->xmin))
return;
/*
* purge xids in InitialRunningXacts as well. The purged array must also
* be sorted in xidComparator order.
*/
workspace =
MemoryContextAlloc(builder->context,
NInitialRunningXacts * sizeof(TransactionId));
surviving_xids = 0;
for (off = 0; off < NInitialRunningXacts; off++)
{
if (NormalTransactionIdPrecedes(InitialRunningXacts[off],
builder->xmin))
; /* remove */
else
workspace[surviving_xids++] = InitialRunningXacts[off];
}
if (surviving_xids > 0)
memcpy(InitialRunningXacts, workspace,
sizeof(TransactionId) * surviving_xids);
else
{
pfree(InitialRunningXacts);
InitialRunningXacts = NULL;
}
elog(DEBUG3, "purged initial running transactions from %u to %u, oldest running xid %u",
(uint32) NInitialRunningXacts,
(uint32) surviving_xids,
builder->xmin);
NInitialRunningXacts = surviving_xids;
pfree(workspace);
}
/*
@ -1126,7 +1204,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
builder->xmin = running->oldestRunningXid;
/* Remove transactions we don't need to keep track off anymore */
SnapBuildPurgeCommittedTxn(builder);
SnapBuildPurgeOlderTxn(builder);
/*
* Advance the xmin limit for the current replication slot, to allow
@ -1277,6 +1355,20 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn
else if (!builder->building_full_snapshot &&
SnapBuildRestore(builder, lsn))
{
int nxacts = running->subxcnt + running->xcnt;
Size sz = sizeof(TransactionId) * nxacts;
/*
* Remember the transactions and subtransactions that were running
* when xl_running_xacts record that we decoded was written. We use
* this later to identify the transactions have performed catalog
* changes. See SnapBuildXidSetCatalogChanges.
*/
NInitialRunningXacts = nxacts;
InitialRunningXacts = MemoryContextAlloc(builder->context, sz);
memcpy(InitialRunningXacts, running->xids, sz);
qsort(InitialRunningXacts, nxacts, sizeof(TransactionId), xidComparator);
/* there won't be any state to cleanup */
return false;
}
@ -1993,3 +2085,34 @@ CheckPointSnapBuild(void)
}
FreeDir(snap_dir);
}
/*
* If the given xid is in the list of the initial running xacts, we mark the
* transaction and its subtransactions as containing catalog changes. See
* comments for NInitialRunningXacts and InitialRunningXacts for additional
* info.
*/
void
SnapBuildXidSetCatalogChanges(SnapBuild *builder, TransactionId xid, int subxcnt,
TransactionId *subxacts, XLogRecPtr lsn)
{
/*
* Skip if there is no initial running xacts information or the
* transaction is already marked as containing catalog changes.
*/
if (NInitialRunningXacts == 0 ||
ReorderBufferXidHasCatalogChanges(builder->reorder, xid))
return;
if (bsearch(&xid, InitialRunningXacts, NInitialRunningXacts,
sizeof(TransactionId), xidComparator) != NULL)
{
ReorderBufferXidSetCatalogChanges(builder->reorder, xid, lsn);
for (int i = 0; i < subxcnt; i++)
{
ReorderBufferAssignChild(builder->reorder, xid, subxacts[i], lsn);
ReorderBufferXidSetCatalogChanges(builder->reorder, subxacts[i], lsn);
}
}
}

3
src/include/replication/snapbuild.h
View File

@ -90,4 +90,7 @@ extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
struct xl_running_xacts *running);
extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
extern void SnapBuildXidSetCatalogChanges(SnapBuild *builder, TransactionId xid,
int subxcnt, TransactionId *subxacts,
XLogRecPtr lsn);
#endif /* SNAPBUILD_H */