Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Skip to content

Commit 89962bf

Browse files
committed
postgres_fdw: re-issue cancel requests a few times if necessary.
Despite the best efforts of commit 0e5c823, we're still seeing occasional failures of postgres_fdw's query_cancel test in the buildfarm. Investigation suggests that its 100ms timeout is still not enough to reliably ensure that the remote side starts the query before receiving the cancel request --- and if it hasn't, it will just discard the request because it's idle. We discussed allowing a cancel request to kill the next-received query, but that would have wide and perhaps unpleasant side-effects. What seems safer is to make postgres_fdw do what a human user would likely do, which is issue another cancel request if the first one didn't seem to do anything. We'll keep the same overall 30 second grace period before concluding things are broken, but issue additional cancel requests after 1 second, then 2 more seconds, then 4, then 8. (The next one in series is 16 seconds, but we'll hit the 30 second timeout before that.) Having done that, revert the timeout in query_cancel.sql to 10 ms. That will still be enough on most machines, most of the time, for the remote query to start; but now we're intentionally risking the race condition occurring sometimes in the buildfarm, so that the repeat-cancel code path will get some testing. As before, back-patch to v17. We might eventually contemplate back-patching this further, and/or adding similar logic to dblink. But given the lack of field complaints to date, this feels like mostly an exercise in test case stabilization, so v17 is enough. Discussion: https://postgr.es/m/colnv3lzzmc53iu5qoawynr6qq7etn47lmggqr65ddtpjliq5d@glkveb4m6nop
1 parent bbe68c1 commit 89962bf

File tree

3 files changed

+84
-24
lines changed

3 files changed

+84
-24
lines changed

contrib/postgres_fdw/connection.c

+76-20
Original file line numberDiff line numberDiff line change
@@ -95,6 +95,13 @@ static uint32 pgfdw_we_get_result = 0;
9595
*/
9696
#define CONNECTION_CLEANUP_TIMEOUT 30000
9797

98+
/*
99+
* Milliseconds to wait before issuing another cancel request. This covers
100+
* the race condition where the remote session ignored our cancel request
101+
* because it arrived while idle.
102+
*/
103+
#define RETRY_CANCEL_TIMEOUT 1000
104+
98105
/* Macro for constructing abort command to be sent */
99106
#define CONSTRUCT_ABORT_COMMAND(sql, entry, toplevel) \
100107
do { \
@@ -135,6 +142,7 @@ static void pgfdw_reset_xact_state(ConnCacheEntry *entry, bool toplevel);
135142
static bool pgfdw_cancel_query(PGconn *conn);
136143
static bool pgfdw_cancel_query_begin(PGconn *conn, TimestampTz endtime);
137144
static bool pgfdw_cancel_query_end(PGconn *conn, TimestampTz endtime,
145+
TimestampTz retrycanceltime,
138146
bool consume_input);
139147
static bool pgfdw_exec_cleanup_query(PGconn *conn, const char *query,
140148
bool ignore_errors);
@@ -144,6 +152,7 @@ static bool pgfdw_exec_cleanup_query_end(PGconn *conn, const char *query,
144152
bool consume_input,
145153
bool ignore_errors);
146154
static bool pgfdw_get_cleanup_result(PGconn *conn, TimestampTz endtime,
155+
TimestampTz retrycanceltime,
147156
PGresult **result, bool *timed_out);
148157
static void pgfdw_abort_cleanup(ConnCacheEntry *entry, bool toplevel);
149158
static bool pgfdw_abort_cleanup_begin(ConnCacheEntry *entry, bool toplevel,
@@ -1308,18 +1317,25 @@ pgfdw_reset_xact_state(ConnCacheEntry *entry, bool toplevel)
13081317
static bool
13091318
pgfdw_cancel_query(PGconn *conn)
13101319
{
1320+
TimestampTz now = GetCurrentTimestamp();
13111321
TimestampTz endtime;
1322+
TimestampTz retrycanceltime;
13121323

13131324
/*
13141325
* If it takes too long to cancel the query and discard the result, assume
13151326
* the connection is dead.
13161327
*/
1317-
endtime = TimestampTzPlusMilliseconds(GetCurrentTimestamp(),
1318-
CONNECTION_CLEANUP_TIMEOUT);
1328+
endtime = TimestampTzPlusMilliseconds(now, CONNECTION_CLEANUP_TIMEOUT);
1329+
1330+
/*
1331+
* Also, lose patience and re-issue the cancel request after a little bit.
1332+
* (This serves to close some race conditions.)
1333+
*/
1334+
retrycanceltime = TimestampTzPlusMilliseconds(now, RETRY_CANCEL_TIMEOUT);
13191335

13201336
if (!pgfdw_cancel_query_begin(conn, endtime))
13211337
return false;
1322-
return pgfdw_cancel_query_end(conn, endtime, false);
1338+
return pgfdw_cancel_query_end(conn, endtime, retrycanceltime, false);
13231339
}
13241340

13251341
/*
@@ -1345,9 +1361,10 @@ pgfdw_cancel_query_begin(PGconn *conn, TimestampTz endtime)
13451361
}
13461362

13471363
static bool
1348-
pgfdw_cancel_query_end(PGconn *conn, TimestampTz endtime, bool consume_input)
1364+
pgfdw_cancel_query_end(PGconn *conn, TimestampTz endtime,
1365+
TimestampTz retrycanceltime, bool consume_input)
13491366
{
1350-
PGresult *result = NULL;
1367+
PGresult *result;
13511368
bool timed_out;
13521369

13531370
/*
@@ -1366,7 +1383,8 @@ pgfdw_cancel_query_end(PGconn *conn, TimestampTz endtime, bool consume_input)
13661383
}
13671384

13681385
/* Get and discard the result of the query. */
1369-
if (pgfdw_get_cleanup_result(conn, endtime, &result, &timed_out))
1386+
if (pgfdw_get_cleanup_result(conn, endtime, retrycanceltime,
1387+
&result, &timed_out))
13701388
{
13711389
if (timed_out)
13721390
ereport(WARNING,
@@ -1439,7 +1457,7 @@ pgfdw_exec_cleanup_query_end(PGconn *conn, const char *query,
14391457
TimestampTz endtime, bool consume_input,
14401458
bool ignore_errors)
14411459
{
1442-
PGresult *result = NULL;
1460+
PGresult *result;
14431461
bool timed_out;
14441462

14451463
Assert(query != NULL);
@@ -1457,7 +1475,7 @@ pgfdw_exec_cleanup_query_end(PGconn *conn, const char *query,
14571475
}
14581476

14591477
/* Get the result of the query. */
1460-
if (pgfdw_get_cleanup_result(conn, endtime, &result, &timed_out))
1478+
if (pgfdw_get_cleanup_result(conn, endtime, endtime, &result, &timed_out))
14611479
{
14621480
if (timed_out)
14631481
ereport(WARNING,
@@ -1481,28 +1499,36 @@ pgfdw_exec_cleanup_query_end(PGconn *conn, const char *query,
14811499
}
14821500

14831501
/*
1484-
* Get, during abort cleanup, the result of a query that is in progress. This
1485-
* might be a query that is being interrupted by transaction abort, or it might
1486-
* be a query that was initiated as part of transaction abort to get the remote
1487-
* side back to the appropriate state.
1502+
* Get, during abort cleanup, the result of a query that is in progress.
1503+
* This might be a query that is being interrupted by a cancel request or by
1504+
* transaction abort, or it might be a query that was initiated as part of
1505+
* transaction abort to get the remote side back to the appropriate state.
1506+
*
1507+
* endtime is the time at which we should give up and assume the remote side
1508+
* is dead. retrycanceltime is the time at which we should issue a fresh
1509+
* cancel request (pass the same value as endtime if this is not wanted).
14881510
*
1489-
* endtime is the time at which we should give up and assume the remote
1490-
* side is dead. Returns true if the timeout expired or connection trouble
1491-
* occurred, false otherwise. Sets *result except in case of a timeout.
1492-
* Sets timed_out to true only when the timeout expired.
1511+
* Returns true if the timeout expired or connection trouble occurred,
1512+
* false otherwise. Sets *result except in case of a true result.
1513+
* Sets *timed_out to true only when the timeout expired.
14931514
*/
14941515
static bool
1495-
pgfdw_get_cleanup_result(PGconn *conn, TimestampTz endtime, PGresult **result,
1516+
pgfdw_get_cleanup_result(PGconn *conn, TimestampTz endtime,
1517+
TimestampTz retrycanceltime,
1518+
PGresult **result,
14961519
bool *timed_out)
14971520
{
14981521
volatile bool failed = false;
14991522
PGresult *volatile last_res = NULL;
15001523

1524+
*result = NULL;
15011525
*timed_out = false;
15021526

15031527
/* In what follows, do not leak any PGresults on an error. */
15041528
PG_TRY();
15051529
{
1530+
int canceldelta = RETRY_CANCEL_TIMEOUT * 2;
1531+
15061532
for (;;)
15071533
{
15081534
PGresult *res;
@@ -1513,8 +1539,33 @@ pgfdw_get_cleanup_result(PGconn *conn, TimestampTz endtime, PGresult **result,
15131539
TimestampTz now = GetCurrentTimestamp();
15141540
long cur_timeout;
15151541

1542+
/* If timeout has expired, give up. */
1543+
if (now >= endtime)
1544+
{
1545+
*timed_out = true;
1546+
failed = true;
1547+
goto exit;
1548+
}
1549+
1550+
/* If we need to re-issue the cancel request, do that. */
1551+
if (now >= retrycanceltime)
1552+
{
1553+
/* We ignore failure to issue the repeated request. */
1554+
(void) libpqsrv_cancel(conn, endtime);
1555+
1556+
/* Recompute "now" in case that took measurable time. */
1557+
now = GetCurrentTimestamp();
1558+
1559+
/* Adjust re-cancel timeout in increasing steps. */
1560+
retrycanceltime = TimestampTzPlusMilliseconds(now,
1561+
canceldelta);
1562+
canceldelta += canceldelta;
1563+
}
1564+
15161565
/* If timeout has expired, give up, else get sleep time. */
1517-
cur_timeout = TimestampDifferenceMilliseconds(now, endtime);
1566+
cur_timeout = TimestampDifferenceMilliseconds(now,
1567+
Min(endtime,
1568+
retrycanceltime));
15181569
if (cur_timeout <= 0)
15191570
{
15201571
*timed_out = true;
@@ -1835,7 +1886,9 @@ pgfdw_finish_abort_cleanup(List *pending_entries, List *cancel_requested,
18351886
foreach(lc, cancel_requested)
18361887
{
18371888
ConnCacheEntry *entry = (ConnCacheEntry *) lfirst(lc);
1889+
TimestampTz now = GetCurrentTimestamp();
18381890
TimestampTz endtime;
1891+
TimestampTz retrycanceltime;
18391892
char sql[100];
18401893

18411894
Assert(entry->changing_xact_state);
@@ -1849,10 +1902,13 @@ pgfdw_finish_abort_cleanup(List *pending_entries, List *cancel_requested,
18491902
* remaining entries in the list, leading to slamming that entry's
18501903
* connection shut.
18511904
*/
1852-
endtime = TimestampTzPlusMilliseconds(GetCurrentTimestamp(),
1905+
endtime = TimestampTzPlusMilliseconds(now,
18531906
CONNECTION_CLEANUP_TIMEOUT);
1907+
retrycanceltime = TimestampTzPlusMilliseconds(now,
1908+
RETRY_CANCEL_TIMEOUT);
18541909

1855-
if (!pgfdw_cancel_query_end(entry->conn, endtime, true))
1910+
if (!pgfdw_cancel_query_end(entry->conn, endtime,
1911+
retrycanceltime, true))
18561912
{
18571913
/* Unable to cancel running query */
18581914
pgfdw_reset_xact_state(entry, toplevel);

contrib/postgres_fdw/expected/query_cancel.out

+4-2
Original file line numberDiff line numberDiff line change
@@ -24,8 +24,10 @@ SELECT count(*) FROM ft1 a;
2424
822
2525
(1 row)
2626

27-
-- Timeout needs to be long enough to be sure that we've sent the slow query.
28-
SET LOCAL statement_timeout = '100ms';
27+
-- On most machines, 10ms will be enough to be sure that we've sent the slow
28+
-- query. We may sometimes exercise the race condition where we send cancel
29+
-- before the remote side starts the query, but that's fine too.
30+
SET LOCAL statement_timeout = '10ms';
2931
-- This would take very long if not canceled:
3032
SELECT count(*) FROM ft1 a CROSS JOIN ft1 b CROSS JOIN ft1 c CROSS JOIN ft1 d;
3133
ERROR: canceling statement due to statement timeout

contrib/postgres_fdw/sql/query_cancel.sql

+4-2
Original file line numberDiff line numberDiff line change
@@ -13,8 +13,10 @@ SELECT count(*) FROM ft1 a CROSS JOIN ft1 b CROSS JOIN ft1 c CROSS JOIN ft1 d;
1313
BEGIN;
1414
-- Make sure that connection is open and set up.
1515
SELECT count(*) FROM ft1 a;
16-
-- Timeout needs to be long enough to be sure that we've sent the slow query.
17-
SET LOCAL statement_timeout = '100ms';
16+
-- On most machines, 10ms will be enough to be sure that we've sent the slow
17+
-- query. We may sometimes exercise the race condition where we send cancel
18+
-- before the remote side starts the query, but that's fine too.
19+
SET LOCAL statement_timeout = '10ms';
1820
-- This would take very long if not canceled:
1921
SELECT count(*) FROM ft1 a CROSS JOIN ft1 b CROSS JOIN ft1 c CROSS JOIN ft1 d;
2022
COMMIT;

0 commit comments

Comments
 (0)