@cryptotaxi247 / netdata-1 / commits / 5607d21c0

Store and transmit chart_name to cloud in alert events (#15441)

Emmanuel Vasilakis committed Jul 20, 2023 at 23:23 UTC 5607d21c025bf79aa80fccb5ad07842a31066aef

13 files changed +158 -56

aclk/aclk-schemas

+1 -1

@@ -1 +1 @@

1 -Subproject commit 2aba113db56ac32deccc1e83784c4b3b6fcfe1cd

1 +Subproject commit 40703f5e0258b3f4a97d3767e0f4e6833801b7d1

aclk/schema-wrappers/alarm_stream.cc

+2 -4

@@ -87,6 +87,7 @@ void destroy_alarm_log_entry(struct alarm_log_entry *entry)

87 freez(entry->rendered_info);

88 freez(entry->chart_context);

89 freez(entry->transition_id);

90 + freez(entry->chart_name);

91 }

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

@@ -129,15 +130,12 @@ static void fill_alarm_log_entry(struct alarm_log_entry *data, AlarmLogEntry *pr

130

131 proto->set_value(data->value);

132 proto->set_old_value(data->old_value);

132 -

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

134 -

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

136 -

135 proto->set_chart_context(data->chart_context);

138 -

136 proto->set_event_id(data->event_id);

137 proto->set_transition_id(data->transition_id);

138 + proto->set_chart_name(data->chart_name);

139 }

140

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

aclk/schema-wrappers/alarm_stream.h

@@ -73,6 +73,7 @@ struct alarm_log_entry {

73 char *rendered_info;

75 char *chart_context;

76 + char *chart_name;

78 uint64_t event_id;

79 char *transition_id;

database/rrd.h

+3 -1

@@ -1045,6 +1045,7 @@ struct alarm_entry {

1045 STRING *name;

1046 STRING *chart;

1047 STRING *chart_context;

1048 + STRING *chart_name;

1049 STRING *family;

1050

1051 STRING *classification;

@@ -1086,7 +1087,8 @@ struct alarm_entry {

1087 };

1088

1089 #define ae_name(ae) string2str((ae)->name)

1089 -#define ae_chart_name(ae) string2str((ae)->chart)

1090 +#define ae_chart_id(ae) string2str((ae)->chart)

1091 +#define ae_chart_name(ae) string2str((ae)->chart_name)

1092 #define ae_chart_context(ae) string2str((ae)->chart_context)

1093 #define ae_family(ae) string2str((ae)->family)

1094 #define ae_classification(ae) string2str((ae)->classification)

database/rrdcalc.c

@@ -292,6 +292,7 @@ static void rrdcalc_link_to_rrdset(RRDSET *st, RRDCALC *rc) {

292 rc->name,

293 rc->rrdset->id,

294 rc->rrdset->context,

295 + rc->rrdset->name,

296 rc->rrdset->family,

297 rc->classification,

298 rc->component,

@@ -336,6 +337,7 @@ static void rrdcalc_unlink_from_rrdset(RRDCALC *rc, bool having_ll_wrlock) {

337 rc->name,

338 rc->rrdset->id,

339 rc->rrdset->context,

340 + rc->rrdset->name,

341 rc->rrdset->family,

342 rc->classification,

343 rc->component,

database/sqlite/sqlite_aclk_alert.c

+4 -5

@@ -268,7 +268,7 @@ void aclk_push_alert_event(struct aclk_sync_host_config *wc)

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

269 " hld.duration, hld.non_clear_duration, hld.flags, hld.exec_run_timestamp, hld.delay_up_to_timestamp, hl.name, " \

270 " hl.chart, hl.family, hl.exec, hl.recipient, ha.source, hl.units, hld.info, hld.exec_code, hld.new_status, " \

271 - " hld.old_status, hld.delay, hld.new_value, hld.old_value, hld.last_repeat, hl.chart_context, hld.transition_id, hld.alarm_event_id " \

271 + " hld.old_status, hld.delay, hld.new_value, hld.old_value, hld.last_repeat, hl.chart_context, hld.transition_id, hld.alarm_event_id, hl.chart_name " \

272 " from health_log hl, aclk_alert_%s aa, alert_hash ha, health_log_detail hld " \

273 " where hld.unique_id = aa.alert_unique_id and hl.config_hash_id = ha.hash_id and aa.date_submitted is null " \

274 " and hl.host_id = @host_id and hl.health_log_id = hld.health_log_id " \

@@ -371,14 +371,12 @@ void aclk_push_alert_event(struct aclk_sync_host_config *wc)

371

372 alarm_log.value = (NETDATA_DOUBLE) sqlite3_column_double(res, 23);

373 alarm_log.old_value = (NETDATA_DOUBLE) sqlite3_column_double(res, 24);

374 -

374 alarm_log.updated = (sqlite3_column_int64(res, 8) & HEALTH_ENTRY_FLAG_UPDATED) ? 1 : 0;

375 alarm_log.rendered_info = sqlite3_text_strdupz_empty(res, 18);

377 -

376 alarm_log.chart_context = sqlite3_text_strdupz_empty(res, 26);

377 alarm_log.transition_id = sqlite3_uuid_unparse_strdupz(res, 27);

380 -

378 alarm_log.event_id = (time_t) sqlite3_column_int64(res, 28);

379 + alarm_log.chart_name = sqlite3_text_strdupz_empty(res, 29);

380

381 aclk_send_alarm_log_entry(&alarm_log);

382

@@ -788,7 +786,7 @@ void health_alarm_entry2proto_nolock(struct alarm_log_entry *alarm_log, ALARM_EN

786 char transition_id[UUID_STR_LEN];

787 uuid_unparse_lower(ae->transition_id, transition_id);

788

791 - alarm_log->chart = strdupz(ae_chart_name(ae));

789 + alarm_log->chart = strdupz(ae_chart_id(ae));

790 alarm_log->name = strdupz(ae_name(ae));

791 alarm_log->family = strdupz(ae_family(ae));

792

@@ -827,6 +825,7 @@ void health_alarm_entry2proto_nolock(struct alarm_log_entry *alarm_log, ALARM_EN

825 alarm_log->updated = (ae->flags & HEALTH_ENTRY_FLAG_UPDATED) ? 1 : 0;

826 alarm_log->rendered_info = strdupz(ae_info(ae));

827 alarm_log->chart_context = strdupz(ae_chart_context(ae));

828 + alarm_log->chart_name = strdupz(ae_chart_name(ae));

829

830 alarm_log->transition_id = strdupz((char *)transition_id);

831 alarm_log->event_id = (uint64_t) ae->alarm_event_id;

database/sqlite/sqlite_db_migration.c

+16 -1

@@ -83,6 +83,11 @@ const char *database_migrate_v9_v10[] = {

83 NULL

84 };

86 +const char *database_migrate_v10_v11[] = {

87 + "ALTER TABLE health_log ADD chart_name TEXT;",

88 + NULL

89 +};

90 +

91 static int do_migration_v1_v2(sqlite3 *database, const char *name)

92 {

93 UNUSED(name);

@@ -293,7 +298,6 @@ static int do_migration_v8_v9(sqlite3 *database, const char *name)

298

299 static int do_migration_v9_v10(sqlite3 *database, const char *name)

300 {

296 - UNUSED(name);

301 netdata_log_info("Running \"%s\" database migration", name);

302

303 if (table_exists_in_database("alert_hash") && !column_exists_in_table("alert_hash", "chart_labels"))

@@ -301,6 +305,16 @@ static int do_migration_v9_v10(sqlite3 *database, const char *name)

305 return 0;

306 }

307

308 +static int do_migration_v10_v11(sqlite3 *database, const char *name)

309 +{

310 + netdata_log_info("Running \"%s\" database migration", name);

311 +

312 + if (table_exists_in_database("health_log") && !column_exists_in_table("health_log", "chart_name"))

313 + return init_database_batch(database, DB_CHECK_NONE, 0, &database_migrate_v10_v11[0]);

314 +

315 + return 0;

316 +}

317 +

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

319 {

320 UNUSED(database);

@@ -354,6 +368,7 @@ DATABASE_FUNC_MIGRATION_LIST migration_action[] = {

368 {.name = "v7 to v8", .func = do_migration_v7_v8},

369 {.name = "v8 to v9", .func = do_migration_v8_v9},

370 {.name = "v9 to v10", .func = do_migration_v9_v10},

371 + {.name = "v10 to v11", .func = do_migration_v10_v11},

372 // the terminator of this array

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

374 };

database/sqlite/sqlite_functions.c

+2 -2

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

3 #include "sqlite_functions.h"

4 #include "sqlite_db_migration.h"

6 -#define DB_METADATA_VERSION 10

6 +#define DB_METADATA_VERSION 11

8 const char *database_config[] = {

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

@@ -47,7 +47,7 @@ const char *database_config[] = {

48 "CREATE TABLE IF NOT EXISTS health_log (health_log_id INTEGER PRIMARY KEY, host_id blob, alarm_id int, "

49 "config_hash_id blob, name text, chart text, family text, recipient text, units text, exec text, "

50 - "chart_context text, last_transition_id blob, UNIQUE (host_id, alarm_id)) ;",

50 + "chart_context text, last_transition_id blob, chart_name text, UNIQUE (host_id, alarm_id)) ;",

52 "CREATE INDEX IF NOT EXISTS health_log_ind_1 ON health_log (host_id);",

database/sqlite/sqlite_health.c

+105 -29

@@ -83,8 +83,8 @@ failed:

83 Inserts an entry in the table

84 */

85 #define SQL_INSERT_HEALTH_LOG "INSERT INTO health_log (host_id, alarm_id, " \

86 - "config_hash_id, name, chart, family, exec, recipient, units, chart_context, last_transition_id) " \

87 - "VALUES (?,?,?,?,?,?,?,?,?,?,?) " \

86 + "config_hash_id, name, chart, family, exec, recipient, units, chart_context, last_transition_id, chart_name) " \

87 + "VALUES (?,?,?,?,?,?,?,?,?,?,?,?) " \

88 "ON CONFLICT (host_id, alarm_id) DO UPDATE SET last_transition_id = excluded.last_transition_id RETURNING health_log_id; "

90 #define SQL_INSERT_HEALTH_LOG_DETAIL "INSERT INTO health_log_detail (health_log_id, unique_id, alarm_id, alarm_event_id, " \

@@ -174,6 +174,12 @@ void sql_health_alarm_log_insert(RRDHOST *host, ALARM_ENTRY *ae) {

174 goto failed;

175 }

176

177 + rc = sqlite3_bind_string_or_null(res, ae->chart_name, 12);

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

179 + error_report("Failed to bind chart_name parameter for SQL_INSERT_HEALTH_LOG");

180 + goto failed;

181 + }

182 +

183 rc = sqlite3_step_monitored(res);

184 if (likely(rc == SQLITE_ROW))

185 health_log_id = (size_t) sqlite3_column_int64(res, 0);

@@ -766,7 +772,7 @@ void sql_check_removed_alerts_state(RRDHOST *host)

772 "hld.updates_id, hld.when_key, hld.duration, hld.non_clear_duration, hld.flags, hld.exec_run_timestamp, " \

773 "hld.delay_up_to_timestamp, hl.name, hl.chart, hl.family, hl.exec, hl.recipient, ah.source, hl.units, " \

774 "hld.info, hld.exec_code, hld.new_status, hld.old_status, hld.delay, hld.new_value, hld.old_value, " \

769 - "hld.last_repeat, ah.class, ah.component, ah.type, hl.chart_context, hld.transition_id, hld.global_id " \

775 + "hld.last_repeat, ah.class, ah.component, ah.type, hl.chart_context, hld.transition_id, hld.global_id, hl.chart_name " \

776 "FROM health_log hl, alert_hash ah, health_log_detail hld " \

777 "WHERE hl.config_hash_id = ah.hash_id and hl.host_id = @host_id and hl.last_transition_id = hld.transition_id;"

778 void sql_health_alarm_log_load(RRDHOST *host) {

@@ -945,6 +951,11 @@ void sql_health_alarm_log_load(RRDHOST *host) {

951 if (sqlite3_column_type(res, 32) != SQLITE_NULL)

952 ae->global_id = sqlite3_column_int64(res, 32);

953

954 + if (sqlite3_column_type(res, 33) != SQLITE_NULL)

955 + ae->chart_name = string_strdupz((char *) sqlite3_column_text(res, 33));

956 + else

957 + ae->chart_name = NULL;

958 +

959 char value_string[100 + 1];

960 string_freez(ae->old_value_string);

961 string_freez(ae->new_value_string);

@@ -1767,7 +1778,7 @@ fail:

1778 #define SQL_POPULATE_TEMP_ALERT_TRANSITION_TABLE "INSERT INTO v_%p (host_id) VALUES (@host_id)"

1779

1780 #define SQL_SEARCH_ALERT_TRANSITION_SELECT "SELECT " \

1770 - "h.host_id, h.alarm_id, h.config_hash_id, h.name, h.chart, h.family, h.recipient, h.units, h.exec, " \

1781 + "h.host_id, h.alarm_id, h.config_hash_id, h.name, h.chart, h.chart_name, h.family, h.recipient, h.units, h.exec, " \

1782 "h.chart_context, d.when_key, d.duration, d.non_clear_duration, d.flags, d.delay_up_to_timestamp, " \

1783 "d.info, d.exec_code, d.new_status, d.old_status, d.delay, d.new_value, d.old_value, d.last_repeat, " \

1784 "d.transition_id, d.global_id, ah.class, ah.type, ah.component, d.exec_run_timestamp"

@@ -1913,31 +1924,31 @@ run_query:;

1924 atd.config_hash_id = (uuid_t *)sqlite3_column_blob(res, 2);

1925 atd.alert_name = (const char *) sqlite3_column_text(res, 3);

1926 atd.chart = (const char *) sqlite3_column_text(res, 4);

1916 - atd.chart_name = (const char *) sqlite3_column_text(res, 4); // FIXME don't copy the id, find the name

1917 - atd.family = (const char *) sqlite3_column_text(res, 5);

1918 - atd.recipient = (const char *) sqlite3_column_text(res, 6);

1919 - atd.units = (const char *) sqlite3_column_text(res, 7);

1920 - atd.exec = (const char *) sqlite3_column_text(res, 8);

1921 - atd.chart_context = (const char *) sqlite3_column_text(res, 9);

1922 - atd.when_key = sqlite3_column_int64(res, 10);

1923 - atd.duration = sqlite3_column_int64(res, 11);

1924 - atd.non_clear_duration = sqlite3_column_int64(res, 12);

1925 - atd.flags = sqlite3_column_int64(res, 13);

1926 - atd.delay_up_to_timestamp = sqlite3_column_int64(res, 14);

1927 - atd.info = (const char *) sqlite3_column_text(res, 15);

1928 - atd.exec_code = sqlite3_column_int(res, 16);

1929 - atd.new_status = sqlite3_column_int(res, 17);

1930 - atd.old_status = sqlite3_column_int(res, 18);

1931 - atd.delay = (int) sqlite3_column_int(res, 19);

1932 - atd.new_value = (NETDATA_DOUBLE) sqlite3_column_double(res, 20);

1933 - atd.old_value = (NETDATA_DOUBLE) sqlite3_column_double(res, 21);

1934 - atd.last_repeat = sqlite3_column_int64(res, 22);

1935 - atd.transition_id = (uuid_t *) sqlite3_column_blob(res, 23);

1936 - atd.global_id = sqlite3_column_int64(res, 24);

1937 - atd.classification = (const char *) sqlite3_column_text(res, 25);

1938 - atd.type = (const char *) sqlite3_column_text(res, 26);

1939 - atd.component = (const char *) sqlite3_column_text(res, 27);

1940 - atd.exec_run_timestamp = sqlite3_column_int64(res, 28);

1927 + atd.chart_name = (const char *) sqlite3_column_text(res, 5);

1928 + atd.family = (const char *) sqlite3_column_text(res, 6);

1929 + atd.recipient = (const char *) sqlite3_column_text(res, 7);

1930 + atd.units = (const char *) sqlite3_column_text(res, 8);

1931 + atd.exec = (const char *) sqlite3_column_text(res, 9);

1932 + atd.chart_context = (const char *) sqlite3_column_text(res, 10);

1933 + atd.when_key = sqlite3_column_int64(res, 11);

1934 + atd.duration = sqlite3_column_int64(res, 12);

1935 + atd.non_clear_duration = sqlite3_column_int64(res, 13);

1936 + atd.flags = sqlite3_column_int64(res, 14);

1937 + atd.delay_up_to_timestamp = sqlite3_column_int64(res, 15);

1938 + atd.info = (const char *) sqlite3_column_text(res, 16);

1939 + atd.exec_code = sqlite3_column_int(res, 17);

1940 + atd.new_status = sqlite3_column_int(res, 18);

1941 + atd.old_status = sqlite3_column_int(res, 19);

1942 + atd.delay = (int) sqlite3_column_int(res, 20);

1943 + atd.new_value = (NETDATA_DOUBLE) sqlite3_column_double(res, 21);

1944 + atd.old_value = (NETDATA_DOUBLE) sqlite3_column_double(res, 22);

1945 + atd.last_repeat = sqlite3_column_int64(res, 23);

1946 + atd.transition_id = (uuid_t *) sqlite3_column_blob(res, 24);

1947 + atd.global_id = sqlite3_column_int64(res, 25);

1948 + atd.classification = (const char *) sqlite3_column_text(res, 26);

1949 + atd.type = (const char *) sqlite3_column_text(res, 27);

1950 + atd.component = (const char *) sqlite3_column_text(res, 28);

1951 + atd.exec_run_timestamp = sqlite3_column_int64(res, 29);

1952

1953 cb(&atd, data);

1954 }

@@ -2087,3 +2098,68 @@ fail_only_drop:

2098 return added;

2099 }

2100

2101 +#define SQL_FETCH_CHART_NAME "SELECT chart_name FROM health_log where host_id = @host_id LIMIT 1;"

2102 +bool is_chart_name_populated(uuid_t *host_uuid)

2103 +{

2104 + sqlite3_stmt *res = NULL;

2105 + int rc;

2106 +

2107 + bool status = true;

2108 +

2109 + rc = sqlite3_prepare_v2(db_meta, SQL_FETCH_CHART_NAME, -1, &res, 0);

2110 + if (unlikely(rc != SQLITE_OK)) {

2111 + error_report("Failed to prepare statement to check health_log chart_name");

2112 + return true;

2113 + }

2114 +

2115 + rc = sqlite3_bind_blob(res, 1, host_uuid, sizeof(*host_uuid), SQLITE_STATIC);

2116 + if (unlikely(rc != SQLITE_OK)) {

2117 + error_report("Failed to bind host_id for health_log chart_name check");

2118 + goto fail;

2119 + }

2120 +

2121 + rc = sqlite3_step_monitored(res);

2122 + if (likely(rc == SQLITE_ROW))

2123 + status = sqlite3_column_type(res, 0) != SQLITE_NULL;

2124 +fail:

2125 +

2126 + rc = sqlite3_finalize(res);

2127 + if (unlikely(rc != SQLITE_OK))

2128 + error_report("Failed to finalize the prepared statement for health_log chart_name check");

2129 +

2130 + return status;

2131 +}

2132 +

2133 +#define SQL_POPULATE_CHART_NAME " UPDATE health_log SET chart_name = upd.chart_name FROM " \

2134 + "(SELECT c.type || '.' || IFNULL(c.name, c.id) AS chart_name, hl.host_id, hl.health_log_id FROM " \

2135 + "chart c, health_log hl WHERE (c.type || '.' || c.id) = hl.chart AND c.host_id = hl.host_id " \

2136 + "AND hl.host_id = @host_id) AS upd WHERE health_log.host_id = upd.host_id " \

2137 + "AND health_log.health_log_id = upd.health_log_id"

2138 +

2139 +void chart_name_populate(uuid_t *host_uuid)

2140 +{

2141 + sqlite3_stmt *res = NULL;

2142 + int rc;

2143 +

2144 + rc = sqlite3_prepare_v2(db_meta, SQL_POPULATE_CHART_NAME, -1, &res, 0);

2145 + if (unlikely(rc != SQLITE_OK)) {

2146 + error_report("Failed to prepare statement to update health_log chart_name");

2147 + return;

2148 + }

2149 +

2150 + rc = sqlite3_bind_blob(res, 1, host_uuid, sizeof(*host_uuid), SQLITE_STATIC);

2151 + if (unlikely(rc != SQLITE_OK)) {

2152 + error_report("Failed to bind host_id for health_log chart_name update");

2153 + goto fail;

2154 + }

2155 +

2156 + rc = execute_insert(res);

2157 + if (unlikely(rc != SQLITE_DONE))

2158 + error_report("Failed to update chart name in health_log, rc = %d", rc);

2159 +

2160 +fail:

2161 +

2162 + rc = sqlite3_finalize(res);

2163 + if (unlikely(rc != SQLITE_OK))

2164 + error_report("Failed to finalize the prepared statement for health_log chart_name update");

2165 +}

database/sqlite/sqlite_health.h

@@ -37,4 +37,6 @@ int sql_get_alert_configuration(

37 bool debug __maybe_unused);

39 bool sql_find_alert_transition(const char *transition, void (*cb)(const char *machine_guid, const char *context, time_t alert_id, void *data), void *data);

40 +bool is_chart_name_populated(uuid_t *host_uuid);

41 +void chart_name_populate(uuid_t *host_uuid);

42 #endif //NETDATA_SQLITE_HEALTH_H

health/health.c

+17 -13

@@ -433,14 +433,14 @@ static inline void health_alarm_execute(RRDHOST *host, ALARM_ENTRY *ae) {

433

434 if(unlikely(ae->new_status < RRDCALC_STATUS_CLEAR)) {

435 // do not send notifications for internal statuses

436 - netdata_log_debug(D_HEALTH, "Health not sending notification for alarm '%s.%s' status %s (internal statuses)", ae_chart_name(ae), ae_name(ae), rrdcalc_status2string(ae->new_status));

436 + netdata_log_debug(D_HEALTH, "Health not sending notification for alarm '%s.%s' status %s (internal statuses)", ae_chart_id(ae), ae_name(ae), rrdcalc_status2string(ae->new_status));

437 goto done;

438 }

439

440 if(unlikely(ae->new_status <= RRDCALC_STATUS_CLEAR && (ae->flags & HEALTH_ENTRY_FLAG_NO_CLEAR_NOTIFICATION))) {

441 // do not send notifications for disabled statuses

442 - netdata_log_debug(D_HEALTH, "Health not sending notification for alarm '%s.%s' status %s (it has no-clear-notification enabled)", ae_chart_name(ae), ae_name(ae), rrdcalc_status2string(ae->new_status));

443 - netdata_log_health("[%s]: Health not sending notification for alarm '%s.%s' status %s (it has no-clear-notification enabled)", rrdhost_hostname(host), ae_chart_name(ae), ae_name(ae), rrdcalc_status2string(ae->new_status));

442 + netdata_log_debug(D_HEALTH, "Health not sending notification for alarm '%s.%s' status %s (it has no-clear-notification enabled)", ae_chart_id(ae), ae_name(ae), rrdcalc_status2string(ae->new_status));

443 + netdata_log_health("[%s]: Health not sending notification for alarm '%s.%s' status %s (it has no-clear-notification enabled)", rrdhost_hostname(host), ae_chart_id(ae), ae_name(ae), rrdcalc_status2string(ae->new_status));

444

445 // mark it as run, so that we will send the same alarm if it happens again

446 goto done;

@@ -457,9 +457,9 @@ static inline void health_alarm_execute(RRDHOST *host, ALARM_ENTRY *ae) {

457 // we have executed this alarm notification in the past

458 if(last_executed_status == ae->new_status && !(ae->flags & HEALTH_ENTRY_FLAG_IS_REPEATING)) {

459 // don't send the notification for the same status again

460 - netdata_log_debug(D_HEALTH, "Health not sending again notification for alarm '%s.%s' status %s", ae_chart_name(ae), ae_name(ae)

460 + netdata_log_debug(D_HEALTH, "Health not sending again notification for alarm '%s.%s' status %s", ae_chart_id(ae), ae_name(ae)

461 , rrdcalc_status2string(ae->new_status));

462 - netdata_log_health("[%s]: Health not sending again notification for alarm '%s.%s' status %s", rrdhost_hostname(host), ae_chart_name(ae), ae_name(ae)

462 + netdata_log_health("[%s]: Health not sending again notification for alarm '%s.%s' status %s", rrdhost_hostname(host), ae_chart_id(ae), ae_name(ae)

463 , rrdcalc_status2string(ae->new_status));

464 goto done;

465 }

@@ -470,7 +470,7 @@ static inline void health_alarm_execute(RRDHOST *host, ALARM_ENTRY *ae) {

470 if(unlikely(ae->new_status == RRDCALC_STATUS_CLEAR)) {

471 if((!(ae->flags & HEALTH_ENTRY_RUN_ONCE)) || (ae->flags & HEALTH_ENTRY_RUN_ONCE && ae->old_status < RRDCALC_STATUS_RAISED) ) {

472 netdata_log_debug(D_HEALTH, "Health not sending notification for first initialization of alarm '%s.%s' status %s"

473 - , ae_chart_name(ae), ae_name(ae), rrdcalc_status2string(ae->new_status));

473 + , ae_chart_id(ae), ae_name(ae), rrdcalc_status2string(ae->new_status));

474 goto done;

475 }

476 }

@@ -479,11 +479,11 @@ static inline void health_alarm_execute(RRDHOST *host, ALARM_ENTRY *ae) {

479

480 // Check if alarm notifications are silenced

481 if (ae->flags & HEALTH_ENTRY_FLAG_SILENCED) {

482 - netdata_log_health("[%s]: Health not sending notification for alarm '%s.%s' status %s (command API has disabled notifications)", rrdhost_hostname(host), ae_chart_name(ae), ae_name(ae), rrdcalc_status2string(ae->new_status));

482 + netdata_log_health("[%s]: Health not sending notification for alarm '%s.%s' status %s (command API has disabled notifications)", rrdhost_hostname(host), ae_chart_id(ae), ae_name(ae), rrdcalc_status2string(ae->new_status));

483 goto done;

484 }

485

486 - netdata_log_health("[%s]: Sending notification for alarm '%s.%s' status %s.", rrdhost_hostname(host), ae_chart_name(ae), ae_name(ae), rrdcalc_status2string(ae->new_status));

486 + netdata_log_health("[%s]: Sending notification for alarm '%s.%s' status %s.", rrdhost_hostname(host), ae_chart_id(ae), ae_name(ae), rrdcalc_status2string(ae->new_status));

487

488 const char *exec = (ae->exec) ? ae_exec(ae) : string2str(host->health.health_default_exec);

489 const char *recipient = (ae->recipient) ? ae_recipient(ae) : string2str(host->health.health_default_recipient);

@@ -562,7 +562,7 @@ static inline void health_alarm_execute(RRDHOST *host, ALARM_ENTRY *ae) {

562 ae->alarm_event_id,

563 (unsigned long)ae->when,

564 ae_name(ae),

565 - ae->chart?ae_chart_name(ae):"NOCHART",

565 + ae->chart?ae_chart_id(ae):"NOCHART",

566 ae->family?ae_family(ae):"NOFAMILY",

567 rrdcalc_status2string(ae->new_status),

568 rrdcalc_status2string(ae->old_status),

@@ -627,7 +627,7 @@ static inline void health_alarm_wait_for_execution(ALARM_ENTRY *ae) {

627

628 static inline void health_process_notifications(RRDHOST *host, ALARM_ENTRY *ae) {

629 netdata_log_debug(D_HEALTH, "Health alarm '%s.%s' = " NETDATA_DOUBLE_FORMAT_AUTO " - changed status from %s to %s",

630 - ae->chart?ae_chart_name(ae):"NOCHART", ae_name(ae),

630 + ae->chart?ae_chart_id(ae):"NOCHART", ae_name(ae),

631 ae->new_value,

632 rrdcalc_status2string(ae->old_status),

633 rrdcalc_status2string(ae->new_status)

@@ -844,8 +844,9 @@ static void initialize_health(RRDHOST *host)

844 host->health.health_default_exec = string_strdupz(config_get(CONFIG_SECTION_HEALTH, "script to execute on alarm", filename));

845 host->health.health_default_recipient = string_strdupz("root");

846

847 - // TODO: This needs to go to the metadata thread

848 - // Health should wait before accessing the table (needs to be created by the metadata thread)

847 + if (!is_chart_name_populated(&host->host_uuid))

848 + chart_name_populate(&host->host_uuid);

849 +

850 sql_health_alarm_log_load(host);

851

852 // ------------------------------------------------------------------------

@@ -1156,6 +1157,7 @@ void *health_main(void *ptr) {

1157 rc->name,

1158 rc->rrdset->id,

1159 rc->rrdset->context,

1160 + rc->rrdset->name,

1161 rc->rrdset->family,

1162 rc->classification,

1163 rc->component,

@@ -1422,6 +1424,7 @@ void *health_main(void *ptr) {

1424 rc->name,

1425 rc->rrdset->id,

1426 rc->rrdset->context,

1427 + rc->rrdset->name,

1428 rc->rrdset->family,

1429 rc->classification,

1430 rc->component,

@@ -1446,7 +1449,7 @@ void *health_main(void *ptr) {

1449

1450 health_alarm_log_add_entry(host, ae);

1451

1449 - netdata_log_health("[%s]: Alert event for [%s.%s], value [%s], status [%s].", rrdhost_hostname(host), ae_chart_name(ae), ae_name(ae), ae_new_value_string(ae), rrdcalc_status2string(ae->new_status));

1452 + netdata_log_health("[%s]: Alert event for [%s.%s], value [%s], status [%s].", rrdhost_hostname(host), ae_chart_id(ae), ae_name(ae), ae_new_value_string(ae), rrdcalc_status2string(ae->new_status));

1453

1454 rc->last_status_change_value = rc->value;

1455 rc->last_status_change = now;

@@ -1508,6 +1511,7 @@ void *health_main(void *ptr) {

1511 rc->name,

1512 rc->rrdset->id,

1513 rc->rrdset->context,

1514 + rc->rrdset->name,

1515 rc->rrdset->family,

1516 rc->classification,

1517 rc->component,

health/health.h

@@ -70,6 +70,7 @@ ALARM_ENTRY* health_create_alarm_entry(

70 STRING *name,

71 STRING *chart,

72 STRING *chart_context,

73 + STRING *chart_id,

74 STRING *family,

75 STRING *classification,

76 STRING *component,

health/health_log.c

@@ -20,6 +20,7 @@ inline ALARM_ENTRY* health_create_alarm_entry(

20 STRING *name,

21 STRING *chart,

22 STRING *chart_context,

23 + STRING *chart_name,

24 STRING *family,

25 STRING *class,

26 STRING *component,

@@ -43,6 +44,7 @@ inline ALARM_ENTRY* health_create_alarm_entry(

44 ae->name = string_dup(name);

45 ae->chart = string_dup(chart);

46 ae->chart_context = string_dup(chart_context);

47 + ae->chart_name = string_dup(chart_name);

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