@cryptotaxi247 / netdata-1 / commits / 2fd260747

Send chart context with alert events to the cloud (#13409)

* add chart context to alert events * migrate health log tables to add chart_context * send it via proto message * add from v3 to v4 * free table * free chart_context

Emmanuel Vasilakis committed Aug 4, 2022 at 10:18 UTC 2fd2607475479bcb1e5d929f109ffd5537303bb7

12 files changed +77 -15

aclk/schema-wrappers/alarm_stream.cc

@@ -118,6 +118,7 @@ void destroy_alarm_log_entry(struct alarm_log_entry *entry)

118 freez(entry->old_value_string);

119

120 freez(entry->rendered_info);

121 + freez(entry->chart_context);

122 }

123

124 static void fill_alarm_log_entry(struct alarm_log_entry *data, AlarmLogEntry *proto)

@@ -166,6 +167,8 @@ static void fill_alarm_log_entry(struct alarm_log_entry *data, AlarmLogEntry *pr

167 proto->set_updated(data->updated);

168

169 proto->set_rendered_info(data->rendered_info);

170 +

171 + proto->set_chart_context(data->chart_context);

172 }

173

174 char *generate_alarm_log_entry(size_t *len, struct alarm_log_entry *data)

aclk/schema-wrappers/alarm_stream.h

@@ -97,6 +97,8 @@ struct alarm_log_entry {

98 // rendered_info

99 char *rendered_info;

100 +

101 + char *chart_context;

102 };

103

104 struct send_alarm_snapshot {

database/rrd.h

@@ -694,6 +694,7 @@ struct alarm_entry {

694

695 char *chart;

696 uint32_t hash_chart;

697 + char *chart_context;

698

699 char *family;

700

database/rrdcalc.c

@@ -92,6 +92,7 @@ static void rrdsetcalc_link(RRDSET *st, RRDCALC *rc) {

92 now,

93 rc->name,

94 rc->rrdset->id,

95 + rc->rrdset->context,

96 rc->rrdset->family,

97 rc->classification,

98 rc->component,

@@ -164,6 +165,7 @@ inline void rrdsetcalc_unlink(RRDCALC *rc) {

165 now,

166 rc->name,

167 rc->rrdset->id,

168 + rc->rrdset->context,

169 rc->rrdset->family,

170 rc->classification,

171 rc->component,

database/sqlite/sqlite_aclk_alert.c

+5 -1

@@ -261,7 +261,7 @@ void aclk_push_alert_event(struct aclk_database_worker_config *wc, struct aclk_d

261 buffer_sprintf(sql, "select aa.sequence_id, hl.unique_id, hl.alarm_id, hl.config_hash_id, hl.updated_by_id, hl.when_key, \

262 hl.duration, hl.non_clear_duration, hl.flags, hl.exec_run_timestamp, hl.delay_up_to_timestamp, hl.name, \

263 hl.chart, hl.family, hl.exec, hl.recipient, hl.source, hl.units, hl.info, hl.exec_code, hl.new_status, \

264 - hl.old_status, hl.delay, hl.new_value, hl.old_value, hl.last_repeat \

264 + hl.old_status, hl.delay, hl.new_value, hl.old_value, hl.last_repeat, hl.chart_context \

265 from health_log_%s hl, aclk_alert_%s aa \

266 where hl.unique_id = aa.alert_unique_id and aa.date_submitted is null \

267 order by aa.sequence_id asc limit %d;", wc->uuid_str, wc->uuid_str, limit);

@@ -344,6 +344,10 @@ void aclk_push_alert_event(struct aclk_database_worker_config *wc, struct aclk_d

344 strdupz((char *)"") :

345 strdupz((char *)sqlite3_column_text(res, 18));

346

347 + alarm_log.chart_context = sqlite3_column_type(res, 26) == SQLITE_NULL ?

348 + strdupz((char *)"") :

349 + strdupz((char *)sqlite3_column_text(res, 26));

350 +

351 aclk_send_alarm_log_entry(&alarm_log);

352

353 if (first_sequence_id == 0)

database/sqlite/sqlite_db_migration.c

+32 -1

@@ -84,6 +84,37 @@ static int do_migration_v2_v3(sqlite3 *database, const char *name)

84 return 0;

85 }

87 +static int do_migration_v3_v4(sqlite3 *database, const char *name)

88 +{

89 + UNUSED(name);

90 + info("Running database migration %s", name);

91 +

92 + char sql[256];

93 +

94 + int rc;

95 + sqlite3_stmt *res = NULL;

96 + snprintfz(sql, 255, "SELECT name FROM sqlite_schema WHERE type ='table' AND name LIKE 'health_log_%%';");

97 + rc = sqlite3_prepare_v2(database, sql, -1, &res, 0);

98 + if (rc != SQLITE_OK) {

99 + error_report("Failed to prepare statement to alter health_log tables");

100 + return 1;

101 + }

102 +

103 + while (sqlite3_step(res) == SQLITE_ROW) {

104 + char *table = strdupz((char *) sqlite3_column_text(res, 0));

105 + if (!column_exists_in_table(table, "chart_context")) {

106 + snprintfz(sql, 255, "ALTER TABLE %s ADD chart_context text", table);

107 + sqlite3_exec(database, sql, 0, 0, NULL);

108 + }

109 + freez(table);

110 + }

111 +

112 + rc = sqlite3_finalize(res);

113 + if (unlikely(rc != SQLITE_OK))

114 + error_report("Failed to finalize statement when altering health_log tables, rc = %d", rc);

115 +

116 + return 0;

117 +}

118

119 static int do_migration_noop(sqlite3 *database, const char *name)

120 {

@@ -131,13 +162,13 @@ DATABASE_FUNC_MIGRATION_LIST migration_action[] = {

162 {.name = "v0 to v1", .func = do_migration_noop},

163 {.name = "v1 to v2", .func = do_migration_v1_v2},

164 {.name = "v2 to v3", .func = do_migration_v2_v3},

165 + {.name = "v3 to v4", .func = do_migration_v3_v4},

166 // the terminator of this array

167 {.name = NULL, .func = NULL}

168 };

169

170 DATABASE_FUNC_MIGRATION_LIST context_migration_action[] = {

171 {.name = "v0 to v1", .func = do_migration_noop},

140 -

172 // the terminator of this array

173 {.name = NULL, .func = NULL}

174 };

database/sqlite/sqlite_functions.c

+1 -1

@@ -3,7 +3,7 @@

3 #include "sqlite_functions.h"

4 #include "sqlite_db_migration.h"

6 -#define DB_METADATA_VERSION 3

6 +#define DB_METADATA_VERSION 4

8 const char *database_config[] = {

9 "CREATE TABLE IF NOT EXISTS host(host_id BLOB PRIMARY KEY, hostname TEXT NOT NULL, "

database/sqlite/sqlite_health.c

+16 -5

@@ -8,7 +8,7 @@

8 /* Health related SQL queries

9 Creates a health log table in sqlite, one per host guid

10 */

11 -#define SQL_CREATE_HEALTH_LOG_TABLE(guid) "CREATE TABLE IF NOT EXISTS health_log_%s(hostname text, unique_id int, alarm_id int, alarm_event_id int, config_hash_id blob, updated_by_id int, updates_id int, when_key int, duration int, non_clear_duration int, flags int, exec_run_timestamp int, delay_up_to_timestamp int, name text, chart text, family text, exec text, recipient text, source text, units text, info text, exec_code int, new_status real, old_status real, delay int, new_value double, old_value double, last_repeat int, class text, component text, type text);", guid

11 +#define SQL_CREATE_HEALTH_LOG_TABLE(guid) "CREATE TABLE IF NOT EXISTS health_log_%s(hostname text, unique_id int, alarm_id int, alarm_event_id int, config_hash_id blob, updated_by_id int, updates_id int, when_key int, duration int, non_clear_duration int, flags int, exec_run_timestamp int, delay_up_to_timestamp int, name text, chart text, family text, exec text, recipient text, source text, units text, info text, exec_code int, new_status real, old_status real, delay int, new_value double, old_value double, last_repeat int, class text, component text, type text, chart_context text);", guid

12 int sql_create_health_log_table(RRDHOST *host) {

13 int rc;

14 char *err_msg = NULL, command[MAX_HEALTH_SQL_SIZE + 1];

@@ -113,7 +113,7 @@ void sql_health_alarm_log_update(RRDHOST *host, ALARM_ENTRY *ae) {

113 "config_hash_id, updated_by_id, updates_id, when_key, duration, non_clear_duration, flags, " \

114 "exec_run_timestamp, delay_up_to_timestamp, name, chart, family, exec, recipient, source, " \

115 "units, info, exec_code, new_status, old_status, delay, new_value, old_value, last_repeat, " \

116 - "class, component, type) values (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?);", guid

116 + "class, component, type, chart_context) values (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?);", guid

117

118 void sql_health_alarm_log_insert(RRDHOST *host, ALARM_ENTRY *ae) {

119 sqlite3_stmt *res = NULL;

@@ -323,6 +323,12 @@ void sql_health_alarm_log_insert(RRDHOST *host, ALARM_ENTRY *ae) {

323 goto failed;

324 }

325

326 + rc = sqlite3_bind_text(res, 32, ae->chart_context, -1, SQLITE_STATIC);

327 + if (unlikely(rc != SQLITE_OK)) {

328 + error_report("Failed to bind chart_context parameter for SQL_INSERT_HEALTH_LOG");

329 + goto failed;

330 + }

331 +

332 rc = execute_insert(res);

333 if (unlikely(rc != SQLITE_DONE)) {

334 error_report("HEALTH [%s]: Failed to execute SQL_INSERT_HEALTH_LOG, rc = %d", host->hostname, rc);

@@ -434,9 +440,9 @@ void sql_health_alarm_log_count(RRDHOST *host) {

440 }

441

442 #define SQL_INJECT_REMOVED(guid, guid2) "insert into health_log_%s (hostname, unique_id, alarm_id, alarm_event_id, config_hash_id, updated_by_id, updates_id, when_key, duration, non_clear_duration, flags, exec_run_timestamp, " \

437 -"delay_up_to_timestamp, name, chart, family, exec, recipient, source, units, info, exec_code, new_status, old_status, delay, new_value, old_value, last_repeat, class, component, type) " \

443 +"delay_up_to_timestamp, name, chart, family, exec, recipient, source, units, info, exec_code, new_status, old_status, delay, new_value, old_value, last_repeat, class, component, type, chart_context) " \

444 "select hostname, ?1, ?2, ?3, config_hash_id, 0, ?4, unixepoch(), 0, 0, flags, exec_run_timestamp, " \

439 -"unixepoch(), name, chart, family, exec, recipient, source, units, info, exec_code, -2, new_status, delay, NULL, new_value, 0, class, component, type " \

445 +"unixepoch(), name, chart, family, exec, recipient, source, units, info, exec_code, -2, new_status, delay, NULL, new_value, 0, class, component, type, chart_context " \

446 "from health_log_%s where unique_id = ?5", guid, guid2

447 #define SQL_INJECT_REMOVED_UPDATE(guid) "update health_log_%s set flags = flags | ?1, updated_by_id = ?2 where unique_id = ?3; ", guid

448 void sql_inject_removed_status(char *uuid_str, uint32_t alarm_id, uint32_t alarm_event_id, uint32_t unique_id, uint32_t max_unique_id)

@@ -598,7 +604,7 @@ void sql_check_removed_alerts_state(char *uuid_str)

604 /* Health related SQL queries

605 Load from the health log table

606 */

601 -#define SQL_LOAD_HEALTH_LOG(guid,limit) "SELECT hostname, unique_id, alarm_id, alarm_event_id, config_hash_id, updated_by_id, updates_id, when_key, duration, non_clear_duration, flags, exec_run_timestamp, delay_up_to_timestamp, name, chart, family, exec, recipient, source, units, info, exec_code, new_status, old_status, delay, new_value, old_value, last_repeat, class, component, type FROM (SELECT hostname, unique_id, alarm_id, alarm_event_id, config_hash_id, updated_by_id, updates_id, when_key, duration, non_clear_duration, flags, exec_run_timestamp, delay_up_to_timestamp, name, chart, family, exec, recipient, source, units, info, exec_code, new_status, old_status, delay, new_value, old_value, last_repeat, class, component, type FROM health_log_%s order by unique_id desc limit %u) order by unique_id asc;", guid, limit

607 +#define SQL_LOAD_HEALTH_LOG(guid,limit) "SELECT hostname, unique_id, alarm_id, alarm_event_id, config_hash_id, updated_by_id, updates_id, when_key, duration, non_clear_duration, flags, exec_run_timestamp, delay_up_to_timestamp, name, chart, family, exec, recipient, source, units, info, exec_code, new_status, old_status, delay, new_value, old_value, last_repeat, class, component, type, chart_context FROM (SELECT hostname, unique_id, alarm_id, alarm_event_id, config_hash_id, updated_by_id, updates_id, when_key, duration, non_clear_duration, flags, exec_run_timestamp, delay_up_to_timestamp, name, chart, family, exec, recipient, source, units, info, exec_code, new_status, old_status, delay, new_value, old_value, last_repeat, class, component, type, chart_context FROM health_log_%s order by unique_id desc limit %u) order by unique_id asc;", guid, limit

608 void sql_health_alarm_log_load(RRDHOST *host) {

609 sqlite3_stmt *res = NULL;

610 int rc;

@@ -771,6 +777,11 @@ void sql_health_alarm_log_load(RRDHOST *host) {

777 else

778 ae->type = NULL;

779

780 + if (sqlite3_column_type(res, 31) != SQLITE_NULL)

781 + ae->chart_context = strdupz((char *) sqlite3_column_text(res, 31));

782 + else

783 + ae->chart_context = NULL;

784 +

785 char value_string[100 + 1];

786 freez(ae->old_value_string);

787 freez(ae->new_value_string);

health/health.c

+3 -3

@@ -853,7 +853,7 @@ void *health_main(void *ptr) {

853 worker_is_busy(WORKER_HEALTH_JOB_ALARM_LOG_ENTRY);

854 time_t now = now_realtime_sec();

855 ALARM_ENTRY *ae = health_create_alarm_entry(

856 - host, rc->id, rc->next_event_id++, rc->config_hash_id, now, rc->name, rc->rrdset->id,

856 + host, rc->id, rc->next_event_id++, rc->config_hash_id, now, rc->name, rc->rrdset->id, rc->rrdset->context,

857 rc->rrdset->family, rc->classification, rc->component, rc->type, rc->exec, rc->recipient, now - rc->last_status_change,

858 rc->value, NAN, rc->status, RRDCALC_STATUS_REMOVED, rc->source, rc->units, rc->info, 0, 0);

859 if (ae) {

@@ -1113,7 +1113,7 @@ void *health_main(void *ptr) {

1113

1114

1115 ALARM_ENTRY *ae = health_create_alarm_entry(

1116 - host, rc->id, rc->next_event_id++, rc->config_hash_id, now, rc->name, rc->rrdset->id,

1116 + host, rc->id, rc->next_event_id++, rc->config_hash_id, now, rc->name, rc->rrdset->id, rc->rrdset->context,

1117 rc->rrdset->family, rc->classification, rc->component, rc->type, rc->exec, rc->recipient, now - rc->last_status_change,

1118 rc->old_value, rc->value, rc->status, status, rc->source, rc->units, rc->info,

1119 rc->delay_last,

@@ -1165,7 +1165,7 @@ void *health_main(void *ptr) {

1165 rc->last_repeat = now;

1166 if (likely(rc->times_repeat < UINT32_MAX)) rc->times_repeat++;

1167 ALARM_ENTRY *ae = health_create_alarm_entry(

1168 - host, rc->id, rc->next_event_id++, rc->config_hash_id, now, rc->name, rc->rrdset->id,

1168 + host, rc->id, rc->next_event_id++, rc->config_hash_id, now, rc->name, rc->rrdset->id, rc->rrdset->context,

1169 rc->rrdset->family, rc->classification, rc->component, rc->type, rc->exec, rc->recipient, now - rc->last_status_change,

1170 rc->old_value, rc->value, rc->old_status, rc->status, rc->source, rc->units, rc->info,

1171 rc->delay_last,

health/health.h

+3 -2

@@ -56,6 +56,7 @@ extern ALARM_ENTRY* health_create_alarm_entry(

56 time_t when,

57 const char *name,

58 const char *chart,

59 + const char *chart_context,

60 const char *family,

61 const char *classification,

62 const char *component,

@@ -63,8 +64,8 @@ extern ALARM_ENTRY* health_create_alarm_entry(

64 const char *exec,

65 const char *recipient,

66 time_t duration,

66 - NETDATA_DOUBLE old_value,

67 - NETDATA_DOUBLE new_value,

67 + NETDATA_DOUBLE old_value,

68 + NETDATA_DOUBLE new_value,

69 RRDCALC_STATUS old_status,

70 RRDCALC_STATUS new_status,

71 const char *source,

health/health_json.c

@@ -29,6 +29,7 @@ void health_alarm_entry2json_nolock(BUFFER *wb, ALARM_ENTRY *ae, RRDHOST *host)

29 "\t\t\"config_hash_id\": \"%s\",\n"

30 "\t\t\"name\": \"%s\",\n"

31 "\t\t\"chart\": \"%s\",\n"

32 + "\t\t\"context\": \"%s\",\n"

33 "\t\t\"family\": \"%s\",\n"

34 "\t\t\"class\": \"%s\",\n"

35 "\t\t\"component\": \"%s\",\n"

@@ -65,6 +66,7 @@ void health_alarm_entry2json_nolock(BUFFER *wb, ALARM_ENTRY *ae, RRDHOST *host)

66 , config_hash_id

67 , ae->name

68 , ae->chart

69 + , ae->chart_context

70 , ae->family

71 , ae->classification?ae->classification:"Unknown"

72 , ae->component?ae->component:"Unknown"

health/health_log.c

+7 -2

@@ -450,6 +450,7 @@ inline ALARM_ENTRY* health_create_alarm_entry(

450 time_t when,

451 const char *name,

452 const char *chart,

453 + const char *chart_context,

454 const char *family,

455 const char *class,

456 const char *component,

@@ -457,8 +458,8 @@ inline ALARM_ENTRY* health_create_alarm_entry(

458 const char *exec,

459 const char *recipient,

460 time_t duration,

460 - NETDATA_DOUBLE old_value,

461 - NETDATA_DOUBLE new_value,

461 + NETDATA_DOUBLE old_value,

462 + NETDATA_DOUBLE new_value,

463 RRDCALC_STATUS old_status,

464 RRDCALC_STATUS new_status,

465 const char *source,

@@ -478,6 +479,9 @@ inline ALARM_ENTRY* health_create_alarm_entry(

479 ae->hash_chart = simple_hash(ae->chart);

480 }

481

482 + if(chart_context)

483 + ae->chart_context = strdupz(chart_context);

484 +

485 uuid_copy(ae->config_hash_id, *((uuid_t *) config_hash_id));

486

487 if(family)

@@ -583,6 +587,7 @@ inline void health_alarm_log(

587 inline void health_alarm_log_free_one_nochecks_nounlink(ALARM_ENTRY *ae) {

588 freez(ae->name);

589 freez(ae->chart);

590 + freez(ae->chart_context);

591 freez(ae->family);

592 freez(ae->classification);

593 freez(ae->component);