Skip to content

Commit 85600b7

Browse files
alvherrealvherre
committed
Avoid spurious deadlocks when upgrading a tuple lock
When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for X Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for X T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once X releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after X finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com
1 parent 07accce commit 85600b7

File tree

5 files changed

+281
-21
lines changed

5 files changed

+281
-21
lines changed

‎src/backend/access/heap/README.tuplock

Lines changed: 10 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -36,6 +36,16 @@ do LockTuple as well, if there is any conflict, to ensure that they don't
3636
starve out waiting exclusive-lockers. However, if there is not any active
3737
conflict for a tuple, we don't incur any extra overhead.
3838

39+
We make an exception to the above rule for those lockers that already hold
40+
some lock on a tuple and attempt to acquire a stronger one on it. In that
41+
case, we skip the LockTuple() call even when there are conflicts, provided
42+
that the target tuple is being locked, updated or deleted by multiple sessions
43+
concurrently. Failing to skip the lock would risk a deadlock, e.g., between a
44+
session that was first to record its weaker lock in the tuple header and would
45+
be waiting on the LockTuple() call to upgrade to the stronger lock level, and
46+
another session that has already done LockTuple() and is waiting for the first
47+
session transaction to release its tuple header-level lock.
48+
3949
We provide four levels of tuple locking strength: SELECT FOR UPDATE obtains an
4050
exclusive lock which prevents any kind of modification of the tuple. This is
4151
the lock level that is implicitly taken by DELETE operations, and also by

‎src/backend/access/heap/heapam.c

Lines changed: 63 additions & 21 deletions
Original file line numberDiff line numberDiff line change
@@ -120,7 +120,7 @@ static void GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
120120
static TransactionId MultiXactIdGetUpdateXid(TransactionId xmax,
121121
uint16 t_infomask);
122122
static bool DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
123-
LockTupleMode lockmode);
123+
LockTupleMode lockmode, bool *current_is_member);
124124
static void MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
125125
Relation rel, ItemPointer ctid, XLTW_Oper oper,
126126
int *remaining);
@@ -3161,15 +3161,20 @@ heap_delete(Relation relation, ItemPointer tid,
31613161
*/
31623162
if (infomask & HEAP_XMAX_IS_MULTI)
31633163
{
3164-
/* wait for multixact */
3164+
bool current_is_member = false;
3165+
31653166
if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
3166-
LockTupleExclusive))
3167+
LockTupleExclusive, &current_is_member))
31673168
{
31683169
LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
31693170

3170-
/* acquire tuple lock, if necessary */
3171-
heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
3172-
LockWaitBlock, &have_tuple_lock);
3171+
/*
3172+
* Acquire the lock, if necessary (but skip it when we're
3173+
* requesting a lock and already have one; avoids deadlock).
3174+
*/
3175+
if (!current_is_member)
3176+
heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
3177+
LockWaitBlock, &have_tuple_lock);
31733178

31743179
/* wait for multixact */
31753180
MultiXactIdWait((MultiXactId) xwait, MultiXactStatusUpdate, infomask,
@@ -3768,15 +3773,20 @@ heap_update(Relation relation, ItemPointer otid, HeapTuple newtup,
37683773
{
37693774
TransactionId update_xact;
37703775
int remain;
3776+
bool current_is_member = false;
37713777

37723778
if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
3773-
*lockmode))
3779+
*lockmode, &current_is_member))
37743780
{
37753781
LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
37763782

3777-
/* acquire tuple lock, if necessary */
3778-
heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3779-
LockWaitBlock, &have_tuple_lock);
3783+
/*
3784+
* Acquire the lock, if necessary (but skip it when we're
3785+
* requesting a lock and already have one; avoids deadlock).
3786+
*/
3787+
if (!current_is_member)
3788+
heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3789+
LockWaitBlock, &have_tuple_lock);
37803790

37813791
/* wait for multixact */
37823792
MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
@@ -4746,6 +4756,7 @@ heap_lock_tuple(Relation relation, HeapTuple tuple,
47464756
uint16 infomask;
47474757
uint16 infomask2;
47484758
bool require_sleep;
4759+
bool skip_tuple_lock;
47494760
ItemPointerData t_ctid;
47504761

47514762
/* must copy state data before unlocking buffer */
@@ -4771,6 +4782,7 @@ heap_lock_tuple(Relation relation, HeapTuple tuple,
47714782
if (first_time)
47724783
{
47734784
first_time = false;
4785+
skip_tuple_lock = false;
47744786

47754787
if (infomask & HEAP_XMAX_IS_MULTI)
47764788
{
@@ -4799,6 +4811,21 @@ heap_lock_tuple(Relation relation, HeapTuple tuple,
47994811
result = HeapTupleMayBeUpdated;
48004812
goto out_unlocked;
48014813
}
4814+
else
4815+
{
4816+
/*
4817+
* Disable acquisition of the heavyweight tuple lock.
4818+
* Otherwise, when promoting a weaker lock, we might
4819+
* deadlock with another locker that has acquired the
4820+
* heavyweight tuple lock and is waiting for our
4821+
* transaction to finish.
4822+
*
4823+
* Note that in this case we still need to wait for
4824+
* the multixact if required, to avoid acquiring
4825+
* conflicting locks.
4826+
*/
4827+
skip_tuple_lock = true;
4828+
}
48024829
}
48034830

48044831
if (members)
@@ -4953,7 +4980,7 @@ heap_lock_tuple(Relation relation, HeapTuple tuple,
49534980
if (infomask & HEAP_XMAX_IS_MULTI)
49544981
{
49554982
if (!DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
4956-
mode))
4983+
mode, NULL))
49574984
{
49584985
/*
49594986
* No conflict, but if the xmax changed under us in the
@@ -5030,13 +5057,15 @@ heap_lock_tuple(Relation relation, HeapTuple tuple,
50305057
/*
50315058
* Acquire tuple lock to establish our priority for the tuple, or
50325059
* die trying. LockTuple will release us when we are next-in-line
5033-
* for the tuple. We must do this even if we are share-locking.
5060+
* for the tuple. We must do this even if we are share-locking,
5061+
* but not if we already have a weaker lock on the tuple.
50345062
*
50355063
* If we are forced to "start over" below, we keep the tuple lock;
50365064
* this arranges that we stay at the head of the line while
50375065
* rechecking tuple state.
50385066
*/
5039-
if (!heap_acquire_tuplock(relation, tid, mode, wait_policy,
5067+
if (!skip_tuple_lock &&
5068+
!heap_acquire_tuplock(relation, tid, mode, wait_policy,
50405069
&have_tuple_lock))
50415070
{
50425071
/*
@@ -7214,10 +7243,13 @@ HeapTupleGetUpdateXid(HeapTupleHeader tuple)
72147243
* tuple lock of the given strength?
72157244
*
72167245
* The passed infomask pairs up with the given multixact in the tuple header.
7246+
*
7247+
* If current_is_member is not NULL, it is set to 'true' if the current
7248+
* transaction is a member of the given multixact.
72177249
*/
72187250
static bool
72197251
DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
7220-
LockTupleMode lockmode)
7252+
LockTupleMode lockmode, bool *current_is_member)
72217253
{
72227254
int nmembers;
72237255
MultiXactMember *members;
@@ -7238,15 +7270,24 @@ DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
72387270
TransactionId memxid;
72397271
LOCKMODE memlockmode;
72407272

7241-
memlockmode = LOCKMODE_from_mxstatus(members[i].status);
7273+
if (result && (current_is_member == NULL || *current_is_member))
7274+
break;
72427275

7243-
/* ignore members that don't conflict with the lock we want */
7244-
if (!DoLockModesConflict(memlockmode, wanted))
7245-
continue;
7276+
memlockmode = LOCKMODE_from_mxstatus(members[i].status);
72467277

7247-
/* ignore members from current xact */
7278+
/* ignore members from current xact (but track their presence) */
72487279
memxid = members[i].xid;
72497280
if (TransactionIdIsCurrentTransactionId(memxid))
7281+
{
7282+
if (current_is_member != NULL)
7283+
*current_is_member = true;
7284+
continue;
7285+
}
7286+
else if (result)
7287+
continue;
7288+
7289+
/* ignore members that don't conflict with the lock we want */
7290+
if (!DoLockModesConflict(memlockmode, wanted))
72507291
continue;
72517292

72527293
if (ISUPDATE_from_mxstatus(members[i].status))
@@ -7265,10 +7306,11 @@ DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
72657306
/*
72667307
* Whatever remains are either live lockers that conflict with our
72677308
* wanted lock, and updaters that are not aborted. Those conflict
7268-
* with what we want, so return true.
7309+
* with what we want. Set up to return true, but keep going to
7310+
* look for the current transaction among the multixact members,
7311+
* if needed.
72697312
*/
72707313
result = true;
7271-
break;
72727314
}
72737315
pfree(members);
72747316
}

‎src/test/isolation/expected/tuplelock-upgrade-no-deadlock.out

Lines changed: 150 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,150 @@
1+
Parsed test spec with 3 sessions
2+
3+
starting permutation: s1_share s2_for_update s3_share s3_for_update s1_rollback s3_rollback s2_rollback
4+
step s1_share: select id from tlu_job where id = 1 for share;
5+
id
6+
7+
1
8+
step s2_for_update: select id from tlu_job where id = 1 for update; <waiting ...>
9+
step s3_share: select id from tlu_job where id = 1 for share;
10+
id
11+
12+
1
13+
step s3_for_update: select id from tlu_job where id = 1 for update; <waiting ...>
14+
step s1_rollback: rollback;
15+
step s3_for_update: <... completed>
16+
id
17+
18+
1
19+
step s3_rollback: rollback;
20+
step s2_for_update: <... completed>
21+
id
22+
23+
1
24+
step s2_rollback: rollback;
25+
26+
starting permutation: s1_keyshare s2_for_update s3_keyshare s1_update s3_update s1_rollback s3_rollback s2_rollback
27+
step s1_keyshare: select id from tlu_job where id = 1 for key share;
28+
id
29+
30+
1
31+
step s2_for_update: select id from tlu_job where id = 1 for update; <waiting ...>
32+
step s3_keyshare: select id from tlu_job where id = 1 for key share;
33+
id
34+
35+
1
36+
step s1_update: update tlu_job set name = 'b' where id = 1;
37+
step s3_update: update tlu_job set name = 'c' where id = 1; <waiting ...>
38+
step s1_rollback: rollback;
39+
step s3_update: <... completed>
40+
step s3_rollback: rollback;
41+
step s2_for_update: <... completed>
42+
id
43+
44+
1
45+
step s2_rollback: rollback;
46+
47+
starting permutation: s1_keyshare s2_for_update s3_keyshare s1_update s3_update s1_commit s3_rollback s2_rollback
48+
step s1_keyshare: select id from tlu_job where id = 1 for key share;
49+
id
50+
51+
1
52+
step s2_for_update: select id from tlu_job where id = 1 for update; <waiting ...>
53+
step s3_keyshare: select id from tlu_job where id = 1 for key share;
54+
id
55+
56+
1
57+
step s1_update: update tlu_job set name = 'b' where id = 1;
58+
step s3_update: update tlu_job set name = 'c' where id = 1; <waiting ...>
59+
step s1_commit: commit;
60+
step s3_update: <... completed>
61+
step s3_rollback: rollback;
62+
step s2_for_update: <... completed>
63+
id
64+
65+
1
66+
step s2_rollback: rollback;
67+
68+
starting permutation: s1_keyshare s2_for_update s3_keyshare s3_delete s1_rollback s3_rollback s2_rollback
69+
step s1_keyshare: select id from tlu_job where id = 1 for key share;
70+
id
71+
72+
1
73+
step s2_for_update: select id from tlu_job where id = 1 for update; <waiting ...>
74+
step s3_keyshare: select id from tlu_job where id = 1 for key share;
75+
id
76+
77+
1
78+
step s3_delete: delete from tlu_job where id = 1; <waiting ...>
79+
step s1_rollback: rollback;
80+
step s3_delete: <... completed>
81+
step s3_rollback: rollback;
82+
step s2_for_update: <... completed>
83+
id
84+
85+
1
86+
step s2_rollback: rollback;
87+
88+
starting permutation: s1_keyshare s2_for_update s3_keyshare s3_delete s1_rollback s3_commit s2_rollback
89+
step s1_keyshare: select id from tlu_job where id = 1 for key share;
90+
id
91+
92+
1
93+
step s2_for_update: select id from tlu_job where id = 1 for update; <waiting ...>
94+
step s3_keyshare: select id from tlu_job where id = 1 for key share;
95+
id
96+
97+
1
98+
step s3_delete: delete from tlu_job where id = 1; <waiting ...>
99+
step s1_rollback: rollback;
100+
step s3_delete: <... completed>
101+
step s3_commit: commit;
102+
step s2_for_update: <... completed>
103+
id
104+
105+
step s2_rollback: rollback;
106+
107+
starting permutation: s1_share s2_for_update s3_for_update s1_rollback s2_rollback s3_rollback
108+
step s1_share: select id from tlu_job where id = 1 for share;
109+
id
110+
111+
1
112+
step s2_for_update: select id from tlu_job where id = 1 for update; <waiting ...>
113+
step s3_for_update: select id from tlu_job where id = 1 for update; <waiting ...>
114+
step s1_rollback: rollback;
115+
step s2_for_update: <... completed>
116+
id
117+
118+
1
119+
step s2_rollback: rollback;
120+
step s3_for_update: <... completed>
121+
id
122+
123+
1
124+
step s3_rollback: rollback;
125+
126+
starting permutation: s1_share s2_update s3_update s1_rollback s2_rollback s3_rollback
127+
step s1_share: select id from tlu_job where id = 1 for share;
128+
id
129+
130+
1
131+
step s2_update: update tlu_job set name = 'b' where id = 1; <waiting ...>
132+
step s3_update: update tlu_job set name = 'c' where id = 1; <waiting ...>
133+
step s1_rollback: rollback;
134+
step s2_update: <... completed>
135+
step s2_rollback: rollback;
136+
step s3_update: <... completed>
137+
step s3_rollback: rollback;
138+
139+
starting permutation: s1_share s2_delete s3_delete s1_rollback s2_rollback s3_rollback
140+
step s1_share: select id from tlu_job where id = 1 for share;
141+
id
142+
143+
1
144+
step s2_delete: delete from tlu_job where id = 1; <waiting ...>
145+
step s3_delete: delete from tlu_job where id = 1; <waiting ...>
146+
step s1_rollback: rollback;
147+
step s2_delete: <... completed>
148+
step s2_rollback: rollback;
149+
step s3_delete: <... completed>
150+
step s3_rollback: rollback;

‎src/test/isolation/isolation_schedule

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -46,6 +46,7 @@ test: update-locked-tuple
4646
test: propagate-lock-delete
4747
test: tuplelock-conflict
4848
test: tuplelock-update
49+
test: tuplelock-upgrade-no-deadlock
4950
test: freeze-the-dead
5051
test: nowait
5152
test: nowait-2

0 commit comments

Comments
 (0)