Update statistics to address slow queries (#16838)
* Run analyze on aclk_alert tables Add analyze option -W sqlite-analyze * Remove empty line * Remove analyze during runtime * Remove health_log_entries_written * Replace index * Remove forced index skip * Change version and run database analyze * Adjust analyze to run on specific tables Fix previous migration v14 -> v15 typo * Fix v15 -> v16 migration message * Fix v15 -> v16 migration message (typo) * Increase analysis limit
Stelios Fragkakis committed
Jan 29, 2024 at 15:36 UTC
0398c6cff4d611eb53d923d7f464e0ce15ceba9e
7 files changed
+67
-64
daemon/main.c
+6
@@ -812,6 +812,7 @@ int help(int exitcode) {
812
" -W unittest Run internal unittests and exit.\n\n"
813
" -W sqlite-meta-recover Run recovery on the metadata database and exit.\n\n"
814
" -W sqlite-compact Reclaim metadata database unused space and exit.\n\n"
815
+ " -W sqlite-analyze Run update statistics and exit.\n\n"
816
#ifdef ENABLE_DBENGINE
817
" -W createdataset=N Create a DB engine dataset of N seconds and exit.\n\n"
818
" -W stresstest=A,B,C,D,E,F,G\n"
@@ -1528,6 +1529,11 @@ int main(int argc, char **argv) {
1529
return 0;
1530
}
1531
1532
+ if(strcmp(optarg, "sqlite-analyze") == 0) {
1533
+ sql_init_database(DB_CHECK_ANALYZE, 0);
1534
+ return 0;
1535
+ }
1536
+
1537
if(strcmp(optarg, "unittest") == 0) {
1538
unittest_running = true;
1539
database/rrd.h
-1
@@ -1093,7 +1093,6 @@ typedef struct health {
1093
time_t health_delay_up_to; // a timestamp to delay alarms processing up to
1094
STRING *health_default_exec; // the full path of the alarms notifications program
1095
STRING *health_default_recipient; // the default recipient for all alarms
1096
- int health_log_entries_written; // the number of alarm events written to the alarms event log
1096
uint32_t health_default_warn_repeat_every; // the default value for the interval between repeating warning notifications
1097
uint32_t health_default_crit_repeat_every; // the default value for the interval between repeating critical notifications
1098
unsigned int health_enabled; // 1 when this host has health enabled
database/sqlite/sqlite_aclk_alert.c
+1
-1
@@ -97,7 +97,7 @@ done:
97
//decide if some events should be sent or not
98
#define SQL_SELECT_ALERT_BY_ID \
99
"SELECT hld.new_status, hl.config_hash_id, hld.unique_id FROM health_log hl, aclk_alert_%s aa, health_log_detail hld " \
100
- "WHERE hl.host_id = @host_id AND +hld.unique_id = aa.filtered_alert_unique_id " \
100
+ "WHERE hl.host_id = @host_id AND hld.unique_id = aa.filtered_alert_unique_id " \
101
"AND hld.alarm_id = @alarm_id AND hl.health_log_id = hld.health_log_id " \
102
"ORDER BY hld.rowid DESC LIMIT 1"
103
database/sqlite/sqlite_db_migration.c
+39
-3
@@ -382,14 +382,49 @@ static int do_migration_v14_v15(sqlite3 *database)
382
}
383
384
BUFFER *wb = buffer_create(128, NULL);
385
- while (sqlite3_step_monitored(res) == SQLITE_ROW)
386
- buffer_sprintf(wb, "DROP INDEX IF EXISTS %s", (char *) sqlite3_column_text(res, 0));
385
+ size_t count = 0;
386
+ while (sqlite3_step_monitored(res) == SQLITE_ROW) {
387
+ buffer_sprintf(wb, "DROP INDEX IF EXISTS %s; ", (char *)sqlite3_column_text(res, 0));
388
+ count++;
389
+ }
390
391
rc = sqlite3_finalize(res);
392
if (unlikely(rc != SQLITE_OK))
393
error_report("Failed to finalize statement when dropping unused indices, rc = %d", rc);
394
392
- (void) db_execute(database, buffer_tostring(wb));
395
+ if (count)
396
+ (void) db_execute(database, buffer_tostring(wb));
397
+
398
+ buffer_free(wb);
399
+ return 0;
400
+}
401
+
402
+static int do_migration_v15_v16(sqlite3 *database)
403
+{
404
+ char sql[256];
405
+
406
+ int rc;
407
+ sqlite3_stmt *res = NULL;
408
+ snprintfz(sql, sizeof(sql) - 1, "SELECT name FROM sqlite_schema WHERE type = \"table\" AND name LIKE \"aclk_alert_%%\"");
409
+ rc = sqlite3_prepare_v2(database, sql, -1, &res, 0);
410
+ if (rc != SQLITE_OK) {
411
+ error_report("Failed to prepare statement to drop unused indices");
412
+ return 1;
413
+ }
414
+
415
+ BUFFER *wb = buffer_create(128, NULL);
416
+ size_t count = 0;
417
+ while (sqlite3_step_monitored(res) == SQLITE_ROW) {
418
+ buffer_sprintf(wb, "ANALYZE %s ; ", (char *)sqlite3_column_text(res, 0));
419
+ count++;
420
+ }
421
+
422
+ rc = sqlite3_finalize(res);
423
+ if (unlikely(rc != SQLITE_OK))
424
+ error_report("Failed to finalize statement when running ANALYZE on aclk_alert_tables, rc = %d", rc);
425
+
426
+ if (count)
427
+ (void) db_execute(database, buffer_tostring(wb));
428
429
buffer_free(wb);
430
return 0;
@@ -491,6 +526,7 @@ DATABASE_FUNC_MIGRATION_LIST migration_action[] = {
526
{.name = "v12 to v13", .func = do_migration_v12_v13},
527
{.name = "v13 to v14", .func = do_migration_v13_v14},
528
{.name = "v14 to v15", .func = do_migration_v14_v15},
529
+ {.name = "v15 to v16", .func = do_migration_v15_v16},
530
// the terminator of this array
531
{.name = NULL, .func = NULL}
532
};
database/sqlite/sqlite_functions.c
+18
-3
@@ -4,7 +4,7 @@
4
#include "sqlite3recover.h"
5
#include "sqlite_db_migration.h"
6
7
-#define DB_METADATA_VERSION 15
7
+#define DB_METADATA_VERSION 16
8
9
const char *database_config[] = {
10
"CREATE TABLE IF NOT EXISTS host(host_id BLOB PRIMARY KEY, hostname TEXT NOT NULL, "
@@ -64,7 +64,7 @@ const char *database_config[] = {
64
65
"CREATE INDEX IF NOT EXISTS health_log_d_ind_2 ON health_log_detail (global_id)",
66
"CREATE INDEX IF NOT EXISTS health_log_d_ind_3 ON health_log_detail (transition_id)",
67
- "CREATE INDEX IF NOT EXISTS health_log_d_ind_5 ON health_log_detail (health_log_id, unique_id DESC)",
67
+ "CREATE INDEX IF NOT EXISTS health_log_d_ind_9 ON health_log_detail (unique_id DESC, health_log_id)",
68
"CREATE INDEX IF NOT EXISTS health_log_d_ind_6 on health_log_detail (health_log_id, when_key)",
69
"CREATE INDEX IF NOT EXISTS health_log_d_ind_7 on health_log_detail (alarm_id)",
70
"CREATE INDEX IF NOT EXISTS health_log_d_ind_8 on health_log_detail (new_status, updated_by_id)",
@@ -84,6 +84,7 @@ const char *database_cleanup[] = {
84
"DROP INDEX IF EXISTS alert_hash_index",
85
"DROP INDEX IF EXISTS health_log_d_ind_4",
86
"DROP INDEX IF EXISTS health_log_d_ind_1",
87
+ "DROP INDEX IF EXISTS health_log_d_ind_5",
88
NULL
89
};
90
@@ -448,6 +449,20 @@ int sql_init_database(db_check_action_type_t rebuild, int memory)
449
return 1;
450
}
451
452
+ if (rebuild & DB_CHECK_ANALYZE) {
453
+ netdata_log_info("Running ANALYZE on %s", sqlite_database);
454
+ rc = sqlite3_exec_monitored(db_meta, "ANALYZE", 0, 0, &err_msg);
455
+ if (rc != SQLITE_OK) {
456
+ error_report("Failed to execute ANALYZE rc = %d (%s)", rc, err_msg);
457
+ sqlite3_free(err_msg);
458
+ }
459
+ else {
460
+ (void) db_execute(db_meta, "select count(*) from sqlite_master limit 0");
461
+ (void) sqlite3_close(db_meta);
462
+ }
463
+ return 1;
464
+ }
465
+
466
netdata_log_info("SQLite database %s initialization", sqlite_database);
467
468
rc = sqlite3_create_function(db_meta, "u2h", 1, SQLITE_ANY | SQLITE_DETERMINISTIC, 0, sqlite_uuid_parse, 0, 0);
@@ -497,7 +512,7 @@ void sql_close_database(void)
512
513
add_stmt_to_list(NULL);
514
500
- (void) db_execute(db_meta, "PRAGMA analysis_limit=1000");
515
+ (void) db_execute(db_meta, "PRAGMA analysis_limit=10000");
516
(void) db_execute(db_meta, "PRAGMA optimize");
517
518
rc = sqlite3_close_v2(db_meta);
database/sqlite/sqlite_functions.h
+3
-2
@@ -21,8 +21,9 @@ struct node_instance_list {
21
typedef enum db_check_action_type {
22
DB_CHECK_NONE = (1 << 0),
23
DB_CHECK_RECLAIM_SPACE = (1 << 1),
24
- DB_CHECK_CONT = (1 << 2),
25
- DB_CHECK_RECOVER = (1 << 3),
24
+ DB_CHECK_ANALYZE = (1 << 2),
25
+ DB_CHECK_CONT = (1 << 3),
26
+ DB_CHECK_RECOVER = (1 << 4),
27
} db_check_action_type_t;
28
29
#define SQL_MAX_RETRY (100)
database/sqlite/sqlite_health.c
-54
@@ -359,7 +359,6 @@ static void sql_health_alarm_log_insert(RRDHOST *host, ALARM_ENTRY *ae) {
359
}
360
361
ae->flags |= HEALTH_ENTRY_FLAG_SAVED;
362
- host->health.health_log_entries_written++;
362
363
failed:
364
if (unlikely(sqlite3_finalize(res) != SQLITE_OK))
@@ -380,48 +379,6 @@ void sql_health_alarm_log_save(RRDHOST *host, ALARM_ENTRY *ae)
379
}
380
}
381
383
-/* Health related SQL queries
384
- Get a count of rows from health log table
385
-*/
386
-#define SQL_COUNT_HEALTH_LOG_DETAIL "SELECT count(1) FROM health_log_detail hld, health_log hl " \
387
- "where hl.host_id = @host_id and hl.health_log_id = hld.health_log_id"
388
-
389
-static int sql_health_alarm_log_count(RRDHOST *host) {
390
- sqlite3_stmt *res = NULL;
391
- int rc;
392
-
393
- if (unlikely(!db_meta)) {
394
- if (default_rrd_memory_mode == RRD_MEMORY_MODE_DBENGINE)
395
- error_report("Database has not been initialized");
396
- return -1;
397
- }
398
-
399
- int entries_in_db = -1;
400
-
401
- rc = sqlite3_prepare_v2(db_meta, SQL_COUNT_HEALTH_LOG_DETAIL, -1, &res, 0);
402
- if (unlikely(rc != SQLITE_OK)) {
403
- error_report("Failed to prepare statement to count health log entries from db");
404
- goto done;
405
- }
406
-
407
- rc = sqlite3_bind_blob(res, 1, &host->host_uuid, sizeof(host->host_uuid), SQLITE_STATIC);
408
- if (unlikely(rc != SQLITE_OK)) {
409
- error_report("Failed to bind host_id for SQL_COUNT_HEALTH_LOG.");
410
- goto done;
411
- }
412
-
413
- rc = sqlite3_step_monitored(res);
414
- if (likely(rc == SQLITE_ROW))
415
- entries_in_db = (int) sqlite3_column_int64(res, 0);
416
-
417
-done:
418
- rc = sqlite3_finalize(res);
419
- if (unlikely(rc != SQLITE_OK))
420
- error_report("Failed to finalize the prepared statement to count health log entries from db");
421
-
422
- return entries_in_db;
423
-}
424
-
382
/*
383
*
384
* Health related SQL queries
@@ -492,10 +449,6 @@ void sql_health_alarm_log_cleanup(RRDHOST *host, bool claimed) {
449
if (unlikely(rc != SQLITE_DONE))
450
error_report("Failed to cleanup health log detail table, rc = %d", rc);
451
495
- int rows = sql_health_alarm_log_count(host);
496
- if (rows >= 0)
497
- host->health.health_log_entries_written = rows;
498
-
452
if (aclk_table_exists)
453
sql_aclk_alert_clean_dead_entries(host);
454
@@ -769,8 +722,6 @@ void sql_health_alarm_log_load(RRDHOST *host)
722
int ret;
723
ssize_t errored = 0, loaded = 0;
724
772
- host->health.health_log_entries_written = 0;
773
-
725
if (unlikely(!db_meta)) {
726
if (default_rrd_memory_mode == RRD_MEMORY_MODE_DBENGINE)
727
error_report("HEALTH [%s]: Database has not been initialized", rrdhost_hostname(host));
@@ -939,11 +890,6 @@ void sql_health_alarm_log_load(RRDHOST *host)
890
ret = sqlite3_finalize(res);
891
if (unlikely(ret != SQLITE_OK))
892
error_report("Failed to finalize the health log read statement");
942
-
943
- int rows = sql_health_alarm_log_count(host);
944
-
945
- if (rows >= 0)
946
- host->health.health_log_entries_written = rows;
893
}
894
895
/*