agora inbox for pgsql-hackers@postgresql.orghelp / color / mirror / Atom feed
[PATCH 7/8] A couple more places for incremental sort 6+ messages / 3 participants [nested] [flat]
* [PATCH 7/8] A couple more places for incremental sort @ 2019-07-28 14:03 Tomas Vondra <tomas@2ndquadrant.com> 0 siblings, 0 replies; 6+ messages in thread From: Tomas Vondra @ 2019-07-28 14:03 UTC (permalink / raw) --- src/backend/optimizer/geqo/geqo_eval.c | 2 +- src/backend/optimizer/plan/planner.c | 220 ++++++++++++++++++++++++- 2 files changed, 217 insertions(+), 5 deletions(-) diff --git a/src/backend/optimizer/geqo/geqo_eval.c b/src/backend/optimizer/geqo/geqo_eval.c index 6d897936d7..ff33acc7b6 100644 --- a/src/backend/optimizer/geqo/geqo_eval.c +++ b/src/backend/optimizer/geqo/geqo_eval.c @@ -274,7 +274,7 @@ merge_clump(PlannerInfo *root, List *clumps, Clump *new_clump, int num_gene, * grouping_planner). */ if (old_clump->size + new_clump->size < num_gene) - generate_gather_paths(root, joinrel, false); + generate_useful_gather_paths(root, joinrel, false); /* Find and save the cheapest paths for this joinrel */ set_cheapest(joinrel); diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c index 46dc355af3..2880fcabe8 100644 --- a/src/backend/optimizer/plan/planner.c +++ b/src/backend/optimizer/plan/planner.c @@ -5080,6 +5080,67 @@ create_ordered_paths(PlannerInfo *root, add_path(ordered_rel, path); } + + /* + * Consider incremental sort with a gather merge on partial paths. + * + * XXX This is probably duplicate with the paths we already generate + * in generate_useful_gather_paths in apply_scanjoin_target_to_paths. + */ + if (enable_incrementalsort) + { + ListCell *lc; + + foreach(lc, input_rel->partial_pathlist) + { + Path *input_path = (Path *) lfirst(lc); + Path *sorted_path = input_path; + bool is_sorted; + int presorted_keys; + double total_groups; + + /* + * We don't care if this is the cheapest partial path - we + * can't simply skip it, because it may be partially sorted in + * which case we want to consider incremental sort on top of + * it (instead of full sort, which is what happens above). + */ + + is_sorted = pathkeys_common_contained_in(root->sort_pathkeys, + input_path->pathkeys, + &presorted_keys); + + /* Ignore already sorted paths */ + if (is_sorted) + continue; + + if (presorted_keys == 0) + continue; + + /* Since we have presorted keys, consider incremental sort. */ + sorted_path = (Path *) create_incremental_sort_path(root, + ordered_rel, + input_path, + root->sort_pathkeys, + presorted_keys, + limit_tuples); + total_groups = input_path->rows * + input_path->parallel_workers; + sorted_path = (Path *) + create_gather_merge_path(root, ordered_rel, + sorted_path, + sorted_path->pathtarget, + root->sort_pathkeys, NULL, + &total_groups); + + /* Add projection step if needed */ + if (sorted_path->pathtarget != target) + sorted_path = apply_projection_to_path(root, ordered_rel, + sorted_path, target); + + add_path(ordered_rel, sorted_path); + } + } } /* @@ -6580,12 +6641,18 @@ add_paths_to_grouping_rel(PlannerInfo *root, RelOptInfo *input_rel, foreach(lc, partially_grouped_rel->pathlist) { Path *path = (Path *) lfirst(lc); + Path *path_original = path; + bool is_sorted; + int presorted_keys; + + is_sorted = pathkeys_contained_in(root->group_pathkeys, + path->pathkeys); /* * Insert a Sort node, if required. But there's no point in * sorting anything but the cheapest path. */ - if (!pathkeys_contained_in(root->group_pathkeys, path->pathkeys)) + if (!is_sorted) { if (path != partially_grouped_rel->cheapest_total_path) continue; @@ -6616,6 +6683,56 @@ add_paths_to_grouping_rel(PlannerInfo *root, RelOptInfo *input_rel, parse->groupClause, havingQual, dNumGroups)); + + /* + * Now we may consider incremental sort on this path, but only + * when the path is not already sorted and when incremental + * sort is enabled. + */ + if (is_sorted || !enable_incrementalsort) + continue; + + /* Restore the input path (we might have added Sort on top). */ + path = path_original; + + is_sorted = pathkeys_common_contained_in(root->group_pathkeys, + path->pathkeys, + &presorted_keys); + + /* We've already skipped fully sorted paths above. */ + Assert(!is_sorted); + + /* no shared prefix, not point in building incremental sort */ + if (presorted_keys == 0) + continue; + + path = (Path *) create_incremental_sort_path(root, + grouped_rel, + path, + root->group_pathkeys, + presorted_keys, + -1.0); + + if (parse->hasAggs) + add_path(grouped_rel, (Path *) + create_agg_path(root, + grouped_rel, + path, + grouped_rel->reltarget, + parse->groupClause ? AGG_SORTED : AGG_PLAIN, + AGGSPLIT_FINAL_DESERIAL, + parse->groupClause, + havingQual, + agg_final_costs, + dNumGroups)); + else + add_path(grouped_rel, (Path *) + create_group_path(root, + grouped_rel, + path, + parse->groupClause, + havingQual, + dNumGroups)); } } } @@ -6887,6 +7004,60 @@ create_partial_grouping_paths(PlannerInfo *root, dNumPartialGroups)); } } + + /* + * Also consider incremental sort on all partially sorted paths. + */ + if (enable_incrementalsort) + { + foreach(lc, input_rel->pathlist) + { + Path *path = (Path *) lfirst(lc); + bool is_sorted; + int presorted_keys; + + is_sorted = pathkeys_common_contained_in(root->group_pathkeys, + path->pathkeys, + &presorted_keys); + + /* Ignore already sorted paths */ + if (is_sorted) + continue; + + if (presorted_keys == 0) + continue; + + /* Since we have presorted keys, consider incremental sort. */ + path = (Path *) create_incremental_sort_path(root, + partially_grouped_rel, + path, + root->group_pathkeys, + presorted_keys, + -1.0); + + if (parse->hasAggs) + add_path(partially_grouped_rel, (Path *) + create_agg_path(root, + partially_grouped_rel, + path, + partially_grouped_rel->reltarget, + parse->groupClause ? AGG_SORTED : AGG_PLAIN, + AGGSPLIT_INITIAL_SERIAL, + parse->groupClause, + NIL, + agg_partial_costs, + dNumPartialGroups)); + else + add_path(partially_grouped_rel, (Path *) + create_group_path(root, + partially_grouped_rel, + path, + parse->groupClause, + NIL, + dNumPartialGroups)); + } + } + } if (can_sort && cheapest_partial_path != NULL) @@ -6951,10 +7122,10 @@ create_partial_grouping_paths(PlannerInfo *root, /* We've already skipped fully sorted paths above. */ Assert(!is_sorted); - /* no shared prefix, not point in building incremental sort */ if (presorted_keys == 0) continue; + /* Since we have presorted keys, consider incremental sort. */ path = (Path *) create_incremental_sort_path(root, partially_grouped_rel, path, @@ -7079,10 +7250,11 @@ create_partial_grouping_paths(PlannerInfo *root, static void gather_grouping_paths(PlannerInfo *root, RelOptInfo *rel) { + ListCell *lc; Path *cheapest_partial_path; /* Try Gather for unordered paths and Gather Merge for ordered ones. */ - generate_gather_paths(root, rel, true); + generate_useful_gather_paths(root, rel, true); /* Try cheapest partial path + explicit Sort + Gather Merge. */ cheapest_partial_path = linitial(rel->partial_pathlist); @@ -7108,6 +7280,46 @@ gather_grouping_paths(PlannerInfo *root, RelOptInfo *rel) add_path(rel, path); } + + if (!enable_incrementalsort) + return; + + /* also consider incremental sort on partial paths, if enabled */ + foreach(lc, rel->partial_pathlist) + { + Path *path = (Path *) lfirst(lc); + bool is_sorted; + int presorted_keys; + double total_groups; + + is_sorted = pathkeys_common_contained_in(root->group_pathkeys, + path->pathkeys, + &presorted_keys); + + if (is_sorted) + continue; + + if (presorted_keys == 0) + continue; + + path = (Path *) create_incremental_sort_path(root, + rel, + path, + root->group_pathkeys, + presorted_keys, + -1.0); + + path = (Path *) + create_gather_merge_path(root, + rel, + path, + rel->reltarget, + root->group_pathkeys, + NULL, + &total_groups); + + add_path(rel, path); + } } /* @@ -7209,7 +7421,7 @@ apply_scanjoin_target_to_paths(PlannerInfo *root, * paths by doing it after the final scan/join target has been * applied. */ - generate_gather_paths(root, rel, false); + generate_useful_gather_paths(root, rel, false); /* Can't use parallel query above this level. */ rel->partial_pathlist = NIL; -- 2.21.1 --dfcjsgdukgytabqd Content-Type: text/plain; charset=us-ascii Content-Disposition: attachment; filename="v39-0008-fix.patch" ^ permalink raw reply [nested|flat] 6+ messages in thread
* [PATCH 5/5] A couple more places for incremental sort @ 2019-07-28 14:03 Tomas Vondra <tomas@2ndquadrant.com> 0 siblings, 0 replies; 6+ messages in thread From: Tomas Vondra @ 2019-07-28 14:03 UTC (permalink / raw) --- src/backend/optimizer/geqo/geqo_eval.c | 2 +- src/backend/optimizer/plan/planner.c | 218 ++++++++++++++++++++++++- 2 files changed, 216 insertions(+), 4 deletions(-) diff --git a/src/backend/optimizer/geqo/geqo_eval.c b/src/backend/optimizer/geqo/geqo_eval.c index 6d897936d7..ff33acc7b6 100644 --- a/src/backend/optimizer/geqo/geqo_eval.c +++ b/src/backend/optimizer/geqo/geqo_eval.c @@ -274,7 +274,7 @@ merge_clump(PlannerInfo *root, List *clumps, Clump *new_clump, int num_gene, * grouping_planner). */ if (old_clump->size + new_clump->size < num_gene) - generate_gather_paths(root, joinrel, false); + generate_useful_gather_paths(root, joinrel, false); /* Find and save the cheapest paths for this joinrel */ set_cheapest(joinrel); diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c index 84ed69ec5e..15223017c0 100644 --- a/src/backend/optimizer/plan/planner.c +++ b/src/backend/optimizer/plan/planner.c @@ -5070,6 +5070,67 @@ create_ordered_paths(PlannerInfo *root, add_path(ordered_rel, path); } + + /* + * Consider incremental sort with a gather merge on partial paths. + * + * XXX This is probably duplicate with the paths we already generate + * in generate_useful_gather_paths in apply_scanjoin_target_to_paths. + */ + if (enable_incrementalsort) + { + ListCell *lc; + + foreach (lc, input_rel->partial_pathlist) + { + Path *input_path = (Path *) lfirst(lc); + Path *sorted_path = input_path; + bool is_sorted; + int presorted_keys; + double total_groups; + + /* + * We don't care if this is the cheapest partial path - we + * can't simply skip it, because it may be partially sorted + * in which case we want to consider incremental sort on top + * of it (instead of full sort, which is what happens above). + */ + + is_sorted = pathkeys_common_contained_in(root->sort_pathkeys, + input_path->pathkeys, + &presorted_keys); + + /* also ignore already sorted paths */ + if (is_sorted) + continue; + + if (presorted_keys == 0) + continue; + + /* Also consider incremental sort. */ + sorted_path = (Path *) create_incremental_sort_path(root, + ordered_rel, + input_path, + root->sort_pathkeys, + presorted_keys, + limit_tuples); + total_groups = input_path->rows * + input_path->parallel_workers; + sorted_path = (Path *) + create_gather_merge_path(root, ordered_rel, + sorted_path, + sorted_path->pathtarget, + root->sort_pathkeys, NULL, + &total_groups); + + /* Add projection step if needed */ + if (sorted_path->pathtarget != target) + sorted_path = apply_projection_to_path(root, ordered_rel, + sorted_path, target); + + add_path(ordered_rel, sorted_path); + } + } } /* @@ -6570,12 +6631,18 @@ add_paths_to_grouping_rel(PlannerInfo *root, RelOptInfo *input_rel, foreach(lc, partially_grouped_rel->pathlist) { Path *path = (Path *) lfirst(lc); + Path *path_original = path; + bool is_sorted; + int presorted_keys; + + is_sorted = pathkeys_contained_in(root->group_pathkeys, + path->pathkeys); /* * Insert a Sort node, if required. But there's no point in * sorting anything but the cheapest path. */ - if (!pathkeys_contained_in(root->group_pathkeys, path->pathkeys)) + if (!is_sorted) { if (path != partially_grouped_rel->cheapest_total_path) continue; @@ -6606,6 +6673,56 @@ add_paths_to_grouping_rel(PlannerInfo *root, RelOptInfo *input_rel, parse->groupClause, havingQual, dNumGroups)); + + /* + * Now we may consider incremental sort on this path, but only + * when the path is not already sorted and when incremental sort + * is enabled. + */ + if (is_sorted || !enable_incrementalsort) + continue; + + /* Restore the input path (we might have addes Sort on top). */ + path = path_original; + + is_sorted = pathkeys_common_contained_in(root->group_pathkeys, + path->pathkeys, + &presorted_keys); + + /* We've already skipped fully sorted paths above. */ + Assert(!is_sorted); + + /* no shared prefix, not point in building incremental sort */ + if (presorted_keys == 0) + continue; + + path = (Path *) create_incremental_sort_path(root, + grouped_rel, + path, + root->group_pathkeys, + presorted_keys, + -1.0); + + if (parse->hasAggs) + add_path(grouped_rel, (Path *) + create_agg_path(root, + grouped_rel, + path, + grouped_rel->reltarget, + parse->groupClause ? AGG_SORTED : AGG_PLAIN, + AGGSPLIT_FINAL_DESERIAL, + parse->groupClause, + havingQual, + agg_final_costs, + dNumGroups)); + else + add_path(grouped_rel, (Path *) + create_group_path(root, + grouped_rel, + path, + parse->groupClause, + havingQual, + dNumGroups)); } } } @@ -6875,6 +6992,60 @@ create_partial_grouping_paths(PlannerInfo *root, dNumPartialGroups)); } } + + /* + * Also consider incremental sort on all partially sorted paths. + */ + if (enable_incrementalsort) + { + foreach(lc, input_rel->pathlist) + { + Path *path = (Path *) lfirst(lc); + bool is_sorted; + int presorted_keys; + + is_sorted = pathkeys_common_contained_in(root->group_pathkeys, + path->pathkeys, + &presorted_keys); + + /* also ignore already sorted paths */ + if (is_sorted) + continue; + + if (presorted_keys == 0) + continue; + + /* add incremental sort */ + path = (Path *) create_incremental_sort_path(root, + partially_grouped_rel, + path, + root->group_pathkeys, + presorted_keys, + -1.0); + + if (parse->hasAggs) + add_path(partially_grouped_rel, (Path *) + create_agg_path(root, + partially_grouped_rel, + path, + partially_grouped_rel->reltarget, + parse->groupClause ? AGG_SORTED : AGG_PLAIN, + AGGSPLIT_INITIAL_SERIAL, + parse->groupClause, + NIL, + agg_partial_costs, + dNumPartialGroups)); + else + add_path(partially_grouped_rel, (Path *) + create_group_path(root, + partially_grouped_rel, + path, + parse->groupClause, + NIL, + dNumPartialGroups)); + } + } + } if (can_sort && cheapest_partial_path != NULL) @@ -7067,10 +7238,11 @@ create_partial_grouping_paths(PlannerInfo *root, static void gather_grouping_paths(PlannerInfo *root, RelOptInfo *rel) { + ListCell *lc; Path *cheapest_partial_path; /* Try Gather for unordered paths and Gather Merge for ordered ones. */ - generate_gather_paths(root, rel, true); + generate_useful_gather_paths(root, rel, true); /* Try cheapest partial path + explicit Sort + Gather Merge. */ cheapest_partial_path = linitial(rel->partial_pathlist); @@ -7096,6 +7268,46 @@ gather_grouping_paths(PlannerInfo *root, RelOptInfo *rel) add_path(rel, path); } + + if (!enable_incrementalsort) + return; + + /* also consider incremental sort on partial paths, if enabled */ + foreach (lc, rel->partial_pathlist) + { + Path *path = (Path *) lfirst(lc); + bool is_sorted; + int presorted_keys; + double total_groups; + + is_sorted = pathkeys_common_contained_in(root->group_pathkeys, + path->pathkeys, + &presorted_keys); + + if (is_sorted) + continue; + + if (presorted_keys == 0) + continue; + + path = (Path *) create_incremental_sort_path(root, + rel, + path, + root->group_pathkeys, + presorted_keys, + -1.0); + + path = (Path *) + create_gather_merge_path(root, + rel, + path, + rel->reltarget, + root->group_pathkeys, + NULL, + &total_groups); + + add_path(rel, path); + } } /* @@ -7197,7 +7409,7 @@ apply_scanjoin_target_to_paths(PlannerInfo *root, * paths by doing it after the final scan/join target has been * applied. */ - generate_gather_paths(root, rel, false); + generate_useful_gather_paths(root, rel, false); /* Can't use parallel query above this level. */ rel->partial_pathlist = NIL; -- 2.21.1 --xvcuvgto6w2bcqjv-- ^ permalink raw reply [nested|flat] 6+ messages in thread
* [PATCH v4] Allow pgbnech to cancel queries during benchmark @ 2023-07-24 12:53 Yugo Nagata <nagata@sraoss.co.jp> 0 siblings, 0 replies; 6+ messages in thread From: Yugo Nagata @ 2023-07-24 12:53 UTC (permalink / raw) Previously, Ctrl+C during benchmark killed pgbench immediately, but queries running at that time were not cancelled. The commit fixes this so that cancel requests are sent for all connections before pgbench exits. In thread #0, setup_cancel_handler is called before the benchmark so that CancelRequested is set when SIGINT is sent. When SIGINT is sent during the benchmark, on non-Windows, thread #0 will be interrupted, return from I/O wait, and send cancel requests to all connections. After queries are cancelled, other threads also be interrupted and pgbench will exit at the end. On Windows, cancel requests are sent in the callback function specified by setup_cancel_hander. --- src/bin/pgbench/pgbench.c | 89 ++++++++++++++++++++ src/bin/pgbench/t/001_pgbench_with_server.pl | 42 +++++++++ 2 files changed, 131 insertions(+) diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c index 713e8a06bb..5adf099b76 100644 --- a/src/bin/pgbench/pgbench.c +++ b/src/bin/pgbench/pgbench.c @@ -596,6 +596,7 @@ typedef enum typedef struct { PGconn *con; /* connection handle to DB */ + PGcancel *cancel; /* query cancel */ int id; /* client No. */ ConnectionStateEnum state; /* state machine's current state. */ ConditionalStack cstack; /* enclosing conditionals state */ @@ -638,6 +639,8 @@ typedef struct * here */ } CState; +CState *client_states; /* status of all clients */ + /* * Thread state */ @@ -837,6 +840,10 @@ static void add_socket_to_set(socket_set *sa, int fd, int idx); static int wait_on_socket_set(socket_set *sa, int64 usecs); static bool socket_has_input(socket_set *sa, int fd, int idx); +#ifdef WIN32 +static void pgbench_cancel_callback(void); +#endif + /* callback used to build rows for COPY during data loading */ typedef void (*initRowMethod) (PQExpBufferData *sql, int64 curr); @@ -3639,6 +3646,7 @@ advanceConnectionState(TState *thread, CState *st, StatsData *agg) st->state = CSTATE_ABORTED; break; } + st->cancel = PQgetCancel(st->con); /* reset now after connection */ now = pg_time_now(); @@ -4670,6 +4678,18 @@ disconnect_all(CState *state, int length) finishCon(&state[i]); } +/* send cancel requests to all connections */ +static void +cancel_all() +{ + for (int i = 0; i < nclients; i++) + { + char errbuf[1]; + if (client_states[i].cancel != NULL) + (void) PQcancel(client_states[i].cancel, errbuf, sizeof(errbuf)); + } +} + /* * Remove old pgbench tables, if any exist */ @@ -7146,6 +7166,9 @@ main(int argc, char **argv) } } + /* enable threads to access the status of all clients */ + client_states = state; + /* other CState initializations */ for (i = 0; i < nclients; i++) { @@ -7358,6 +7381,37 @@ threadRun(void *arg) StatsData last, aggs; + /* + * Query cancellation is handled only in thread #0. + * + * On Windows, a callback function is set in which query cancel requests + * are sent to all benchmark queries running in the backend. + * + * On non-Windows, any callback function is not set. When SIGINT is + * received, CancelRequested is just set, and only thread #0 is interrupted + * and returns from waiting input from the backend. After that, the thread + * sends cancel requests to all benchmark queries. + */ + if (thread->tid == 0) +#ifdef WIN32 + setup_cancel_handler(pgbench_cancel_callback); +#else + setup_cancel_handler(NULL); +#endif + +#ifndef WIN32 + if (thread->tid > 0) + { + sigset_t sigint_sigset; + sigset_t osigset; + sigemptyset(&sigint_sigset); + sigaddset(&sigint_sigset, SIGINT); + + /* Block SIGINT in all threads except one. */ + pthread_sigmask(SIG_BLOCK, &sigint_sigset, &osigset); + } +#endif + /* open log file if requested */ if (use_log) { @@ -7400,6 +7454,7 @@ threadRun(void *arg) pg_fatal("could not create connection for client %d", state[i].id); } + state[i].cancel = PQgetCancel(state[i].con); } } @@ -7427,6 +7482,26 @@ threadRun(void *arg) pg_time_usec_t min_usec; pg_time_usec_t now = 0; /* set this only if needed */ + /* + * If pgbench is cancelled, send cancel requests to all connections + * and exit the benchmark. + * + * Note that only thread #0 can be interrupted by SIGINT while waiting + * the result from the backend. Other threads will return from waiting + * just after queries they running are cancelled by thread #0. + * + * On Windows, cancel requests are sent in the callback function, so + * do nothing but exit the benchmark. + */ + if (CancelRequested) + { +#ifndef WIN32 + if (thread->tid == 0) + cancel_all(); +#endif + goto done; + } + /* * identify which client sockets should be checked for input, and * compute the nearest time (if any) at which we need to wake up. @@ -7650,6 +7725,8 @@ finishCon(CState *st) { PQfinish(st->con); st->con = NULL; + PQfreeCancel(st->cancel); + st->cancel = NULL; } } @@ -7867,3 +7944,15 @@ socket_has_input(socket_set *sa, int fd, int idx) } #endif /* POLL_USING_SELECT */ + +#ifdef WIN32 +/* + * query cancellation callback for Windows + */ +static void +pgbench_cancel_callback(void) +{ + /* send cancel requests to all connections */ + cancel_all(); +} +#endif diff --git a/src/bin/pgbench/t/001_pgbench_with_server.pl b/src/bin/pgbench/t/001_pgbench_with_server.pl index 96be529d6b..43eadfbe9f 100644 --- a/src/bin/pgbench/t/001_pgbench_with_server.pl +++ b/src/bin/pgbench/t/001_pgbench_with_server.pl @@ -7,6 +7,7 @@ use warnings; use PostgreSQL::Test::Cluster; use PostgreSQL::Test::Utils; use Test::More; +use Time::HiRes qw(usleep); # Check the initial state of the data generated. Tables for tellers and # branches use NULL for their filler attribute. The table accounts uses @@ -1502,6 +1503,47 @@ update counter set i = i+1 returning i \gset # Clean up $node->safe_psql('postgres', 'DROP TABLE counter;'); +# Test query canceling by sending SIGINT to a running pgbench +SKIP: +{ + skip "sending SIGINT on Windows terminates the test itself", 3 + if $windows_os; + + my ($stdin, $stdout, $stderr, @file); + + @file = $node->_pgbench_make_files( + { + '003_pgbench_cancel' => qq{ +select pg_sleep($PostgreSQL::Test::Utils::timeout_default); + }}); + + local %ENV = $node->_get_env(); + + my $h = IPC::Run::start( + [ 'pgbench', '-c', '2', '-j', '2', + '-T', "$PostgreSQL::Test::Utils::timeout_default", @file ], + \$stdin, \$stdout, \$stderr); + + $node->poll_query_until('postgres', + q{SELECT (SELECT count(*) FROM pg_stat_activity WHERE query ~ '^select pg_sleep') = 2;} + ) or die "timed out"; + + # Send cancel request + $h->signal('INT'); + + my $result = finish $h; + + ok(!$result, 'pgbench failed as expected'); + like( + $stderr, + qr/Run was aborted; the above results are incomplete/, + 'pgbench was canceled'); + + is($node->safe_psql('postgres', + q{SELECT count(*) FROM pg_stat_activity WHERE query ~ '^select pg_sleep'}), + '0', 'all queries were canceled'); +} + # done $node->safe_psql('postgres', 'DROP TABLESPACE regress_pgbench_tap_1_ts'); $node->stop; -- 2.25.1 --Multipart=_Tue__19_Sep_2023_17_30_11_+0900_JATr+KlP9+ALhH+Y-- ^ permalink raw reply [nested|flat] 6+ messages in thread
* [PATCH v4] Allow pgbnech to cancel queries during benchmark @ 2023-07-24 12:53 Yugo Nagata <nagata@sraoss.co.jp> 0 siblings, 0 replies; 6+ messages in thread From: Yugo Nagata @ 2023-07-24 12:53 UTC (permalink / raw) Previously, Ctrl+C during benchmark killed pgbench immediately, but queries running at that time were not cancelled. The commit fixes this so that cancel requests are sent for all connections before pgbench exits. In thread #0, setup_cancel_handler is called before the benchmark so that CancelRequested is set when SIGINT is sent. When SIGINT is sent during the benchmark, on non-Windows, thread #0 will be interrupted, return from I/O wait, and send cancel requests to all connections. After queries are cancelled, other threads also be interrupted and pgbench will exit at the end. On Windows, cancel requests are sent in the callback function specified by setup_cancel_hander. --- src/bin/pgbench/pgbench.c | 89 ++++++++++++++++++++ src/bin/pgbench/t/001_pgbench_with_server.pl | 42 +++++++++ 2 files changed, 131 insertions(+) diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c index 713e8a06bb..5adf099b76 100644 --- a/src/bin/pgbench/pgbench.c +++ b/src/bin/pgbench/pgbench.c @@ -596,6 +596,7 @@ typedef enum typedef struct { PGconn *con; /* connection handle to DB */ + PGcancel *cancel; /* query cancel */ int id; /* client No. */ ConnectionStateEnum state; /* state machine's current state. */ ConditionalStack cstack; /* enclosing conditionals state */ @@ -638,6 +639,8 @@ typedef struct * here */ } CState; +CState *client_states; /* status of all clients */ + /* * Thread state */ @@ -837,6 +840,10 @@ static void add_socket_to_set(socket_set *sa, int fd, int idx); static int wait_on_socket_set(socket_set *sa, int64 usecs); static bool socket_has_input(socket_set *sa, int fd, int idx); +#ifdef WIN32 +static void pgbench_cancel_callback(void); +#endif + /* callback used to build rows for COPY during data loading */ typedef void (*initRowMethod) (PQExpBufferData *sql, int64 curr); @@ -3639,6 +3646,7 @@ advanceConnectionState(TState *thread, CState *st, StatsData *agg) st->state = CSTATE_ABORTED; break; } + st->cancel = PQgetCancel(st->con); /* reset now after connection */ now = pg_time_now(); @@ -4670,6 +4678,18 @@ disconnect_all(CState *state, int length) finishCon(&state[i]); } +/* send cancel requests to all connections */ +static void +cancel_all() +{ + for (int i = 0; i < nclients; i++) + { + char errbuf[1]; + if (client_states[i].cancel != NULL) + (void) PQcancel(client_states[i].cancel, errbuf, sizeof(errbuf)); + } +} + /* * Remove old pgbench tables, if any exist */ @@ -7146,6 +7166,9 @@ main(int argc, char **argv) } } + /* enable threads to access the status of all clients */ + client_states = state; + /* other CState initializations */ for (i = 0; i < nclients; i++) { @@ -7358,6 +7381,37 @@ threadRun(void *arg) StatsData last, aggs; + /* + * Query cancellation is handled only in thread #0. + * + * On Windows, a callback function is set in which query cancel requests + * are sent to all benchmark queries running in the backend. + * + * On non-Windows, any callback function is not set. When SIGINT is + * received, CancelRequested is just set, and only thread #0 is interrupted + * and returns from waiting input from the backend. After that, the thread + * sends cancel requests to all benchmark queries. + */ + if (thread->tid == 0) +#ifdef WIN32 + setup_cancel_handler(pgbench_cancel_callback); +#else + setup_cancel_handler(NULL); +#endif + +#ifndef WIN32 + if (thread->tid > 0) + { + sigset_t sigint_sigset; + sigset_t osigset; + sigemptyset(&sigint_sigset); + sigaddset(&sigint_sigset, SIGINT); + + /* Block SIGINT in all threads except one. */ + pthread_sigmask(SIG_BLOCK, &sigint_sigset, &osigset); + } +#endif + /* open log file if requested */ if (use_log) { @@ -7400,6 +7454,7 @@ threadRun(void *arg) pg_fatal("could not create connection for client %d", state[i].id); } + state[i].cancel = PQgetCancel(state[i].con); } } @@ -7427,6 +7482,26 @@ threadRun(void *arg) pg_time_usec_t min_usec; pg_time_usec_t now = 0; /* set this only if needed */ + /* + * If pgbench is cancelled, send cancel requests to all connections + * and exit the benchmark. + * + * Note that only thread #0 can be interrupted by SIGINT while waiting + * the result from the backend. Other threads will return from waiting + * just after queries they running are cancelled by thread #0. + * + * On Windows, cancel requests are sent in the callback function, so + * do nothing but exit the benchmark. + */ + if (CancelRequested) + { +#ifndef WIN32 + if (thread->tid == 0) + cancel_all(); +#endif + goto done; + } + /* * identify which client sockets should be checked for input, and * compute the nearest time (if any) at which we need to wake up. @@ -7650,6 +7725,8 @@ finishCon(CState *st) { PQfinish(st->con); st->con = NULL; + PQfreeCancel(st->cancel); + st->cancel = NULL; } } @@ -7867,3 +7944,15 @@ socket_has_input(socket_set *sa, int fd, int idx) } #endif /* POLL_USING_SELECT */ + +#ifdef WIN32 +/* + * query cancellation callback for Windows + */ +static void +pgbench_cancel_callback(void) +{ + /* send cancel requests to all connections */ + cancel_all(); +} +#endif diff --git a/src/bin/pgbench/t/001_pgbench_with_server.pl b/src/bin/pgbench/t/001_pgbench_with_server.pl index 96be529d6b..43eadfbe9f 100644 --- a/src/bin/pgbench/t/001_pgbench_with_server.pl +++ b/src/bin/pgbench/t/001_pgbench_with_server.pl @@ -7,6 +7,7 @@ use warnings; use PostgreSQL::Test::Cluster; use PostgreSQL::Test::Utils; use Test::More; +use Time::HiRes qw(usleep); # Check the initial state of the data generated. Tables for tellers and # branches use NULL for their filler attribute. The table accounts uses @@ -1502,6 +1503,47 @@ update counter set i = i+1 returning i \gset # Clean up $node->safe_psql('postgres', 'DROP TABLE counter;'); +# Test query canceling by sending SIGINT to a running pgbench +SKIP: +{ + skip "sending SIGINT on Windows terminates the test itself", 3 + if $windows_os; + + my ($stdin, $stdout, $stderr, @file); + + @file = $node->_pgbench_make_files( + { + '003_pgbench_cancel' => qq{ +select pg_sleep($PostgreSQL::Test::Utils::timeout_default); + }}); + + local %ENV = $node->_get_env(); + + my $h = IPC::Run::start( + [ 'pgbench', '-c', '2', '-j', '2', + '-T', "$PostgreSQL::Test::Utils::timeout_default", @file ], + \$stdin, \$stdout, \$stderr); + + $node->poll_query_until('postgres', + q{SELECT (SELECT count(*) FROM pg_stat_activity WHERE query ~ '^select pg_sleep') = 2;} + ) or die "timed out"; + + # Send cancel request + $h->signal('INT'); + + my $result = finish $h; + + ok(!$result, 'pgbench failed as expected'); + like( + $stderr, + qr/Run was aborted; the above results are incomplete/, + 'pgbench was canceled'); + + is($node->safe_psql('postgres', + q{SELECT count(*) FROM pg_stat_activity WHERE query ~ '^select pg_sleep'}), + '0', 'all queries were canceled'); +} + # done $node->safe_psql('postgres', 'DROP TABLESPACE regress_pgbench_tap_1_ts'); $node->stop; -- 2.25.1 --Multipart=_Tue__19_Sep_2023_17_30_11_+0900_JATr+KlP9+ALhH+Y-- ^ permalink raw reply [nested|flat] 6+ messages in thread
* [PATCH v4] Allow pgbnech to cancel queries during benchmark @ 2023-07-24 12:53 Yugo Nagata <nagata@sraoss.co.jp> 0 siblings, 0 replies; 6+ messages in thread From: Yugo Nagata @ 2023-07-24 12:53 UTC (permalink / raw) Previously, Ctrl+C during benchmark killed pgbench immediately, but queries running at that time were not cancelled. The commit fixes this so that cancel requests are sent for all connections before pgbench exits. In thread #0, setup_cancel_handler is called before the benchmark so that CancelRequested is set when SIGINT is sent. When SIGINT is sent during the benchmark, on non-Windows, thread #0 will be interrupted, return from I/O wait, and send cancel requests to all connections. After queries are cancelled, other threads also be interrupted and pgbench will exit at the end. On Windows, cancel requests are sent in the callback function specified by setup_cancel_hander. --- src/bin/pgbench/pgbench.c | 89 ++++++++++++++++++++ src/bin/pgbench/t/001_pgbench_with_server.pl | 42 +++++++++ 2 files changed, 131 insertions(+) diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c index 713e8a06bb..5adf099b76 100644 --- a/src/bin/pgbench/pgbench.c +++ b/src/bin/pgbench/pgbench.c @@ -596,6 +596,7 @@ typedef enum typedef struct { PGconn *con; /* connection handle to DB */ + PGcancel *cancel; /* query cancel */ int id; /* client No. */ ConnectionStateEnum state; /* state machine's current state. */ ConditionalStack cstack; /* enclosing conditionals state */ @@ -638,6 +639,8 @@ typedef struct * here */ } CState; +CState *client_states; /* status of all clients */ + /* * Thread state */ @@ -837,6 +840,10 @@ static void add_socket_to_set(socket_set *sa, int fd, int idx); static int wait_on_socket_set(socket_set *sa, int64 usecs); static bool socket_has_input(socket_set *sa, int fd, int idx); +#ifdef WIN32 +static void pgbench_cancel_callback(void); +#endif + /* callback used to build rows for COPY during data loading */ typedef void (*initRowMethod) (PQExpBufferData *sql, int64 curr); @@ -3639,6 +3646,7 @@ advanceConnectionState(TState *thread, CState *st, StatsData *agg) st->state = CSTATE_ABORTED; break; } + st->cancel = PQgetCancel(st->con); /* reset now after connection */ now = pg_time_now(); @@ -4670,6 +4678,18 @@ disconnect_all(CState *state, int length) finishCon(&state[i]); } +/* send cancel requests to all connections */ +static void +cancel_all() +{ + for (int i = 0; i < nclients; i++) + { + char errbuf[1]; + if (client_states[i].cancel != NULL) + (void) PQcancel(client_states[i].cancel, errbuf, sizeof(errbuf)); + } +} + /* * Remove old pgbench tables, if any exist */ @@ -7146,6 +7166,9 @@ main(int argc, char **argv) } } + /* enable threads to access the status of all clients */ + client_states = state; + /* other CState initializations */ for (i = 0; i < nclients; i++) { @@ -7358,6 +7381,37 @@ threadRun(void *arg) StatsData last, aggs; + /* + * Query cancellation is handled only in thread #0. + * + * On Windows, a callback function is set in which query cancel requests + * are sent to all benchmark queries running in the backend. + * + * On non-Windows, any callback function is not set. When SIGINT is + * received, CancelRequested is just set, and only thread #0 is interrupted + * and returns from waiting input from the backend. After that, the thread + * sends cancel requests to all benchmark queries. + */ + if (thread->tid == 0) +#ifdef WIN32 + setup_cancel_handler(pgbench_cancel_callback); +#else + setup_cancel_handler(NULL); +#endif + +#ifndef WIN32 + if (thread->tid > 0) + { + sigset_t sigint_sigset; + sigset_t osigset; + sigemptyset(&sigint_sigset); + sigaddset(&sigint_sigset, SIGINT); + + /* Block SIGINT in all threads except one. */ + pthread_sigmask(SIG_BLOCK, &sigint_sigset, &osigset); + } +#endif + /* open log file if requested */ if (use_log) { @@ -7400,6 +7454,7 @@ threadRun(void *arg) pg_fatal("could not create connection for client %d", state[i].id); } + state[i].cancel = PQgetCancel(state[i].con); } } @@ -7427,6 +7482,26 @@ threadRun(void *arg) pg_time_usec_t min_usec; pg_time_usec_t now = 0; /* set this only if needed */ + /* + * If pgbench is cancelled, send cancel requests to all connections + * and exit the benchmark. + * + * Note that only thread #0 can be interrupted by SIGINT while waiting + * the result from the backend. Other threads will return from waiting + * just after queries they running are cancelled by thread #0. + * + * On Windows, cancel requests are sent in the callback function, so + * do nothing but exit the benchmark. + */ + if (CancelRequested) + { +#ifndef WIN32 + if (thread->tid == 0) + cancel_all(); +#endif + goto done; + } + /* * identify which client sockets should be checked for input, and * compute the nearest time (if any) at which we need to wake up. @@ -7650,6 +7725,8 @@ finishCon(CState *st) { PQfinish(st->con); st->con = NULL; + PQfreeCancel(st->cancel); + st->cancel = NULL; } } @@ -7867,3 +7944,15 @@ socket_has_input(socket_set *sa, int fd, int idx) } #endif /* POLL_USING_SELECT */ + +#ifdef WIN32 +/* + * query cancellation callback for Windows + */ +static void +pgbench_cancel_callback(void) +{ + /* send cancel requests to all connections */ + cancel_all(); +} +#endif diff --git a/src/bin/pgbench/t/001_pgbench_with_server.pl b/src/bin/pgbench/t/001_pgbench_with_server.pl index 96be529d6b..43eadfbe9f 100644 --- a/src/bin/pgbench/t/001_pgbench_with_server.pl +++ b/src/bin/pgbench/t/001_pgbench_with_server.pl @@ -7,6 +7,7 @@ use warnings; use PostgreSQL::Test::Cluster; use PostgreSQL::Test::Utils; use Test::More; +use Time::HiRes qw(usleep); # Check the initial state of the data generated. Tables for tellers and # branches use NULL for their filler attribute. The table accounts uses @@ -1502,6 +1503,47 @@ update counter set i = i+1 returning i \gset # Clean up $node->safe_psql('postgres', 'DROP TABLE counter;'); +# Test query canceling by sending SIGINT to a running pgbench +SKIP: +{ + skip "sending SIGINT on Windows terminates the test itself", 3 + if $windows_os; + + my ($stdin, $stdout, $stderr, @file); + + @file = $node->_pgbench_make_files( + { + '003_pgbench_cancel' => qq{ +select pg_sleep($PostgreSQL::Test::Utils::timeout_default); + }}); + + local %ENV = $node->_get_env(); + + my $h = IPC::Run::start( + [ 'pgbench', '-c', '2', '-j', '2', + '-T', "$PostgreSQL::Test::Utils::timeout_default", @file ], + \$stdin, \$stdout, \$stderr); + + $node->poll_query_until('postgres', + q{SELECT (SELECT count(*) FROM pg_stat_activity WHERE query ~ '^select pg_sleep') = 2;} + ) or die "timed out"; + + # Send cancel request + $h->signal('INT'); + + my $result = finish $h; + + ok(!$result, 'pgbench failed as expected'); + like( + $stderr, + qr/Run was aborted; the above results are incomplete/, + 'pgbench was canceled'); + + is($node->safe_psql('postgres', + q{SELECT count(*) FROM pg_stat_activity WHERE query ~ '^select pg_sleep'}), + '0', 'all queries were canceled'); +} + # done $node->safe_psql('postgres', 'DROP TABLESPACE regress_pgbench_tap_1_ts'); $node->stop; -- 2.25.1 --Multipart=_Tue__19_Sep_2023_17_30_11_+0900_JATr+KlP9+ALhH+Y-- ^ permalink raw reply [nested|flat] 6+ messages in thread
* proposal: parentid and naturalid for plpgsql nodes @ 2026-01-22 08:26 Pavel Stehule <pavel.stehule@gmail.com> 0 siblings, 0 replies; 6+ messages in thread From: Pavel Stehule @ 2026-01-22 08:26 UTC (permalink / raw) To: PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>; +Cc: Tom Lane <tgl@sss.pgh.pa.us> Hi I propose two enhancing of PLpgSQL_stmt structure 1. parentid This is the id (stmtid) of the near outer statement. Why do I need it? Inside the plpgsql_check profiler I need to handle exceptions. Exceptions are not supported by the plpgsql debug API, so I need to hold a stack of executed statements. When any statements start, I need to check this stack and sometimes (after a handled exception) I need to reduce this stack until I find a statement that has the same parenid as the parentid of the current statement. Currently I hold an array with parentid outside, but it is not practical and increases the complexity of plpgsql_check. 2. naturalid stmtid is assigned by the plpgsql parser. It is unique, but has a little bit of a messy order, and when something like a statement id is displayed in some reports, then it is confusing for users. I propose extra id (that is unique too), but with order based on searching statements tree example BEGIN --> stmtid = 3 PERFORM --> stmtid = 1 PERFORM --> stmtid = 2 BEGIN --> naturalid = 1 PERFORM --> naturalid = 2 PERFORM --> naturalid = 3 Maybe the ordering of stmtid can be changed, and then naturalid can be useless. Proposed change can reduce complexity of plpgsql_check, but I believe it can help with other exceptions that use pl debug api. Comments, notes? Regards Pavel ^ permalink raw reply [nested|flat] 6+ messages in thread
end of thread, other threads:[~2026-01-22 08:26 UTC | newest] Thread overview: 6+ messages (download: mbox mbox.gz follow: Atom feed) -- links below jump to the message on this page -- 2019-07-28 14:03 [PATCH 7/8] A couple more places for incremental sort Tomas Vondra <tomas@2ndquadrant.com> 2019-07-28 14:03 [PATCH 5/5] A couple more places for incremental sort Tomas Vondra <tomas@2ndquadrant.com> 2023-07-24 12:53 [PATCH v4] Allow pgbnech to cancel queries during benchmark Yugo Nagata <nagata@sraoss.co.jp> 2023-07-24 12:53 [PATCH v4] Allow pgbnech to cancel queries during benchmark Yugo Nagata <nagata@sraoss.co.jp> 2023-07-24 12:53 [PATCH v4] Allow pgbnech to cancel queries during benchmark Yugo Nagata <nagata@sraoss.co.jp> 2026-01-22 08:26 proposal: parentid and naturalid for plpgsql nodes Pavel Stehule <pavel.stehule@gmail.com>
This inbox is served by agora; see mirroring instructions for how to clone and mirror all data and code used for this inbox