diff options
Diffstat (limited to 'health/health_log.c')
| -rw-r--r-- | health/health_log.c | 316 |
1 files changed, 141 insertions, 175 deletions
diff --git a/health/health_log.c b/health/health_log.c index f0a05531d..8105e01ae 100644 --- a/health/health_log.c +++ b/health/health_log.c @@ -14,11 +14,11 @@ inline int health_alarm_log_open(RRDHOST *host) { if(host->health_log_fp) { if (setvbuf(host->health_log_fp, NULL, _IOLBF, 0) != 0) - error("HEALTH [%s]: cannot set line buffering on health log file '%s'.", host->hostname, host->health_log_filename); + error("HEALTH [%s]: cannot set line buffering on health log file '%s'.", rrdhost_hostname(host), host->health_log_filename); return 0; } - error("HEALTH [%s]: cannot open health log file '%s'. Health data will be lost in case of netdata or server crash.", host->hostname, host->health_log_filename); + error("HEALTH [%s]: cannot open health log file '%s'. Health data will be lost in case of netdata or server crash.", rrdhost_hostname(host), host->health_log_filename); return -1; } @@ -45,13 +45,13 @@ static inline void health_log_rotate(RRDHOST *host) { snprintfz(old_filename, FILENAME_MAX, "%s.old", host->health_log_filename); if(unlink(old_filename) == -1 && errno != ENOENT) - error("HEALTH [%s]: cannot remove old alarms log file '%s'", host->hostname, old_filename); + error("HEALTH [%s]: cannot remove old alarms log file '%s'", rrdhost_hostname(host), old_filename); if(link(host->health_log_filename, old_filename) == -1 && errno != ENOENT) - error("HEALTH [%s]: cannot move file '%s' to '%s'.", host->hostname, host->health_log_filename, old_filename); + error("HEALTH [%s]: cannot move file '%s' to '%s'.", rrdhost_hostname(host), host->health_log_filename, old_filename); if(unlink(host->health_log_filename) == -1 && errno != ENOENT) - error("HEALTH [%s]: cannot remove old alarms log file '%s'", host->hostname, host->health_log_filename); + error("HEALTH [%s]: cannot remove old alarms log file '%s'", rrdhost_hostname(host), host->health_log_filename); // open it with truncate host->health_log_fp = fopen(host->health_log_filename, "w"); @@ -59,7 +59,7 @@ static inline void health_log_rotate(RRDHOST *host) { if(host->health_log_fp) fclose(host->health_log_fp); else - error("HEALTH [%s]: cannot truncate health log '%s'", host->hostname, host->health_log_filename); + error("HEALTH [%s]: cannot truncate health log '%s'", rrdhost_hostname(host), host->health_log_filename); host->health_log_fp = NULL; @@ -75,12 +75,12 @@ inline void health_label_log_save(RRDHOST *host) { if(unlikely(host->health_log_fp)) { BUFFER *wb = buffer_create(1024); - rrdlabels_to_buffer(localhost->host_labels, wb, "", "=", "", "\t ", NULL, NULL, NULL, NULL); + rrdlabels_to_buffer(localhost->rrdlabels, wb, "", "=", "", "\t ", NULL, NULL, NULL, NULL); char *write = (char *) buffer_tostring(wb); if (unlikely(fprintf(host->health_log_fp, "L\t%s", write) < 0)) error("HEALTH [%s]: failed to save alarm log entry to '%s'. Health data may be lost in case of abnormal restart.", - host->hostname, host->health_log_filename); + rrdhost_hostname(host), host->health_log_filename); else host->health_log_entries_written++; @@ -103,7 +103,7 @@ inline void health_alarm_log_save(RRDHOST *host, ALARM_ENTRY *ae) { "\t%s\t%s\t%s" "\n" , (ae->flags & HEALTH_ENTRY_FLAG_SAVED)?'U':'A' - , host->hostname + , rrdhost_hostname(host) , ae->unique_id , ae->alarm_id @@ -118,14 +118,14 @@ inline void health_alarm_log_save(RRDHOST *host, ALARM_ENTRY *ae) { , (uint32_t)ae->exec_run_timestamp , (uint32_t)ae->delay_up_to_timestamp - , (ae->name)?ae->name:"" - , (ae->chart)?ae->chart:"" - , (ae->family)?ae->family:"" - , (ae->exec)?ae->exec:"" - , (ae->recipient)?ae->recipient:"" - , (ae->source)?ae->source:"" - , (ae->units)?ae->units:"" - , (ae->info)?ae->info:"" + , ae_name(ae) + , ae_chart_name(ae) + , ae_family(ae) + , ae_exec(ae) + , ae_recipient(ae) + , ae_source(ae) + , ae_units(ae) + , ae_info(ae) , ae->exec_code , ae->new_status @@ -135,11 +135,11 @@ inline void health_alarm_log_save(RRDHOST *host, ALARM_ENTRY *ae) { , ae->new_value , ae->old_value , (uint64_t)ae->last_repeat - , (ae->classification)?ae->classification:"Unknown" - , (ae->component)?ae->component:"Unknown" - , (ae->type)?ae->type:"Unknown" + , (ae->classification)?ae_classification(ae):"Unknown" + , (ae->component)?ae_component(ae):"Unknown" + , (ae->type)?ae_type(ae):"Unknown" ) < 0)) - error("HEALTH [%s]: failed to save alarm log entry to '%s'. Health data may be lost in case of abnormal restart.", host->hostname, host->health_log_filename); + error("HEALTH [%s]: failed to save alarm log entry to '%s'. Health data may be lost in case of abnormal restart.", rrdhost_hostname(host), host->health_log_filename); else { ae->flags |= HEALTH_ENTRY_FLAG_SAVED; host->health_log_entries_written++; @@ -156,18 +156,23 @@ inline void health_alarm_log_save(RRDHOST *host, ALARM_ENTRY *ae) { static uint32_t is_valid_alarm_id(RRDHOST *host, const char *chart, const char *name, uint32_t alarm_id) { - uint32_t hash_chart = simple_hash(chart); - uint32_t hash_name = simple_hash(name); + STRING *chart_string = string_strdupz(chart); + STRING *name_string = string_strdupz(name); + + uint32_t ret = 1; ALARM_ENTRY *ae; for(ae = host->health_log.alarms; ae ;ae = ae->next) { - if (unlikely( - ae->alarm_id == alarm_id && (!(ae->hash_name == hash_name && ae->hash_chart == hash_chart && - !strcmp(name, ae->name) && !strcmp(chart, ae->chart))))) { - return 0; + if (unlikely(ae->alarm_id == alarm_id && (!(chart_string == ae->chart && name_string == ae->name)))) { + ret = 0; + break; } } - return 1; + + string_freez(chart_string); + string_freez(name_string); + + return ret; } static inline ssize_t health_alarm_log_read(RRDHOST *host, FILE *fp, const char *filename) { @@ -177,6 +182,14 @@ static inline ssize_t health_alarm_log_read(RRDHOST *host, FILE *fp, const char size_t line = 0, len = 0; ssize_t loaded = 0, updated = 0, errored = 0, duplicate = 0; + DICTIONARY *all_rrdcalcs = dictionary_create( + DICT_OPTION_NAME_LINK_DONT_CLONE | DICT_OPTION_VALUE_LINK_DONT_CLONE | DICT_OPTION_DONT_OVERWRITE_VALUE); + RRDCALC *rc; + foreach_rrdcalc_in_rrdhost_read(host, rc) { + dictionary_set(all_rrdcalcs, rrdcalc_name(rc), rc, sizeof(*rc)); + } + foreach_rrdcalc_in_rrdhost_done(rc); + netdata_rwlock_rdlock(&host->health_log.alarm_log_rwlock); while((s = fgets_trim_len(buf, 65536, fp, &len))) { @@ -192,7 +205,7 @@ static inline ssize_t health_alarm_log_read(RRDHOST *host, FILE *fp, const char *s = '\0'; pointers[entries++] = ++s; if(entries >= max_entries) { - error("HEALTH [%s]: line %zu of file '%s' has more than %d entries. Ignoring excessive entries.", host->hostname, line, filename, max_entries); + error("HEALTH [%s]: line %zu of file '%s' has more than %d entries. Ignoring excessive entries.", rrdhost_hostname(host), line, filename, max_entries); break; } } @@ -206,7 +219,7 @@ static inline ssize_t health_alarm_log_read(RRDHOST *host, FILE *fp, const char ALARM_ENTRY *ae = NULL; if(entries < 27) { - error("HEALTH [%s]: line %zu of file '%s' should have at least 27 entries, but it has %d. Ignoring it.", host->hostname, line, filename, entries); + error("HEALTH [%s]: line %zu of file '%s' should have at least 27 entries, but it has %d. Ignoring it.", rrdhost_hostname(host), line, filename, entries); errored++; continue; } @@ -214,14 +227,14 @@ static inline ssize_t health_alarm_log_read(RRDHOST *host, FILE *fp, const char // check that we have valid ids uint32_t unique_id = (uint32_t)strtoul(pointers[2], NULL, 16); if(!unique_id) { - error("HEALTH [%s]: line %zu of file '%s' states alarm entry with invalid unique id %u (%s). Ignoring it.", host->hostname, line, filename, unique_id, pointers[2]); + error("HEALTH [%s]: line %zu of file '%s' states alarm entry with invalid unique id %u (%s). Ignoring it.", rrdhost_hostname(host), line, filename, unique_id, pointers[2]); errored++; continue; } uint32_t alarm_id = (uint32_t)strtoul(pointers[3], NULL, 16); if(!alarm_id) { - error("HEALTH [%s]: line %zu of file '%s' states alarm entry for invalid alarm id %u (%s). Ignoring it.", host->hostname, line, filename, alarm_id, pointers[3]); + error("HEALTH [%s]: line %zu of file '%s' states alarm entry for invalid alarm id %u (%s). Ignoring it.", rrdhost_hostname(host), line, filename, alarm_id, pointers[3]); errored++; continue; } @@ -232,18 +245,7 @@ static inline ssize_t health_alarm_log_read(RRDHOST *host, FILE *fp, const char char* alarm_name = pointers[13]; last_repeat = (time_t)strtoul(pointers[27], NULL, 16); - RRDCALC *rc = alarm_max_last_repeat(host, alarm_name,simple_hash(alarm_name)); - if (!rc) { - for(rc = host->alarms; rc ; rc = rc->next) { - RRDCALC *rdcmp = (RRDCALC *) avl_insert_lock(&(host)->alarms_idx_name, (avl_t *)rc); - if(rdcmp != rc) { - error("Cannot insert the alarm index ID using log %s", rc->name); - } - } - - rc = alarm_max_last_repeat(host, alarm_name,simple_hash(alarm_name)); - } - + rc = dictionary_get(all_rrdcalcs, alarm_name); if(unlikely(rc)) { if (rrdcalc_isrepeating(rc)) { rc->last_repeat = last_repeat; @@ -259,7 +261,7 @@ static inline ssize_t health_alarm_log_read(RRDHOST *host, FILE *fp, const char // make sure it is properly numbered if(unlikely(host->health_log.alarms && unique_id < host->health_log.alarms->unique_id)) { error( "HEALTH [%s]: line %zu of file '%s' has alarm log entry %u in wrong order. Ignoring it." - , host->hostname, line, filename, unique_id); + , rrdhost_hostname(host), line, filename, unique_id); errored++; continue; } @@ -272,7 +274,7 @@ static inline ssize_t health_alarm_log_read(RRDHOST *host, FILE *fp, const char if(unlikely(unique_id == ae->unique_id)) { if(unlikely(*pointers[0] == 'A')) { error("HEALTH [%s]: line %zu of file '%s' adds duplicate alarm log entry %u. Using the later." - , host->hostname, line, filename, unique_id); + , rrdhost_hostname(host), line, filename, unique_id); *pointers[0] = 'U'; duplicate++; } @@ -298,8 +300,13 @@ static inline ssize_t health_alarm_log_read(RRDHOST *host, FILE *fp, const char // error("HEALTH [%s]: line %zu of file '%s' provides an alarm for host '%s' but this is named '%s'.", host->hostname, line, filename, pointers[1], host->hostname); ae->unique_id = unique_id; - if (!is_valid_alarm_id(host, pointers[14], pointers[13], alarm_id)) - alarm_id = rrdcalc_get_unique_id(host, pointers[14], pointers[13], NULL); + if (!is_valid_alarm_id(host, pointers[14], pointers[13], alarm_id)) { + STRING *chart = string_strdupz(pointers[14]); + STRING *name = string_strdupz(pointers[13]); + alarm_id = rrdcalc_get_unique_id(host, chart, name, NULL); + string_freez(chart); + string_freez(name); + } ae->alarm_id = alarm_id; ae->alarm_event_id = (uint32_t)strtoul(pointers[4], NULL, 16); ae->updated_by_id = (uint32_t)strtoul(pointers[5], NULL, 16); @@ -315,36 +322,29 @@ static inline ssize_t health_alarm_log_read(RRDHOST *host, FILE *fp, const char ae->exec_run_timestamp = (uint32_t)strtoul(pointers[11], NULL, 16); ae->delay_up_to_timestamp = (uint32_t)strtoul(pointers[12], NULL, 16); - freez(ae->name); - ae->name = strdupz(pointers[13]); - ae->hash_name = simple_hash(ae->name); + string_freez(ae->name); + ae->name = string_strdupz(pointers[13]); - freez(ae->chart); - ae->chart = strdupz(pointers[14]); - ae->hash_chart = simple_hash(ae->chart); + string_freez(ae->chart); + ae->chart = string_strdupz(pointers[14]); - freez(ae->family); - ae->family = strdupz(pointers[15]); + string_freez(ae->family); + ae->family = string_strdupz(pointers[15]); - freez(ae->exec); - ae->exec = strdupz(pointers[16]); - if(!*ae->exec) { freez(ae->exec); ae->exec = NULL; } + string_freez(ae->exec); + ae->exec = string_strdupz(pointers[16]); - freez(ae->recipient); - ae->recipient = strdupz(pointers[17]); - if(!*ae->recipient) { freez(ae->recipient); ae->recipient = NULL; } + string_freez(ae->recipient); + ae->recipient = string_strdupz(pointers[17]); - freez(ae->source); - ae->source = strdupz(pointers[18]); - if(!*ae->source) { freez(ae->source); ae->source = NULL; } + string_freez(ae->source); + ae->source = string_strdupz(pointers[18]); - freez(ae->units); - ae->units = strdupz(pointers[19]); - if(!*ae->units) { freez(ae->units); ae->units = NULL; } + string_freez(ae->units); + ae->units = string_strdupz(pointers[19]); - freez(ae->info); - ae->info = strdupz(pointers[20]); - if(!*ae->info) { freez(ae->info); ae->info = NULL; } + string_freez(ae->info); + ae->info = string_strdupz(pointers[20]); ae->exec_code = str2i(pointers[21]); ae->new_status = str2i(pointers[22]); @@ -357,24 +357,21 @@ static inline ssize_t health_alarm_log_read(RRDHOST *host, FILE *fp, const char ae->last_repeat = last_repeat; if (likely(entries > 30)) { - freez(ae->classification); - ae->classification = strdupz(pointers[28]); - if(!*ae->classification) { freez(ae->classification); ae->classification = NULL; } + string_freez(ae->classification); + ae->classification = string_strdupz(pointers[28]); - freez(ae->component); - ae->component = strdupz(pointers[29]); - if(!*ae->component) { freez(ae->component); ae->component = NULL; } + string_freez(ae->component); + ae->component = string_strdupz(pointers[29]); - freez(ae->type); - ae->type = strdupz(pointers[30]); - if(!*ae->type) { freez(ae->type); ae->type = NULL; } + string_freez(ae->type); + ae->type = string_strdupz(pointers[30]); } char value_string[100 + 1]; - freez(ae->old_value_string); - freez(ae->new_value_string); - ae->old_value_string = strdupz(format_value_and_unit(value_string, 100, ae->old_value, ae->units, -1)); - ae->new_value_string = strdupz(format_value_and_unit(value_string, 100, ae->new_value, ae->units, -1)); + string_freez(ae->old_value_string); + string_freez(ae->new_value_string); + ae->old_value_string = string_strdupz(format_value_and_unit(value_string, 100, ae->old_value, ae_units(ae), -1)); + ae->new_value_string = string_strdupz(format_value_and_unit(value_string, 100, ae->new_value, ae_units(ae), -1)); // add it to host if not already there if(unlikely(*pointers[0] == 'A')) { @@ -395,13 +392,16 @@ static inline ssize_t health_alarm_log_read(RRDHOST *host, FILE *fp, const char host->health_max_alarm_id = ae->alarm_id; } else { - error("HEALTH [%s]: line %zu of file '%s' is invalid (unrecognized entry type '%s').", host->hostname, line, filename, pointers[0]); + error("HEALTH [%s]: line %zu of file '%s' is invalid (unrecognized entry type '%s').", rrdhost_hostname(host), line, filename, pointers[0]); errored++; } } netdata_rwlock_unlock(&host->health_log.alarm_log_rwlock); + dictionary_destroy(all_rrdcalcs); + all_rrdcalcs = NULL; + freez(buf); if(!host->health_max_unique_id) host->health_max_unique_id = (uint32_t)now_realtime_sec(); @@ -411,7 +411,7 @@ static inline ssize_t health_alarm_log_read(RRDHOST *host, FILE *fp, const char if (unlikely(!host->health_log.next_alarm_id || host->health_log.next_alarm_id <= host->health_max_alarm_id)) host->health_log.next_alarm_id = host->health_max_alarm_id + 1; - debug(D_HEALTH, "HEALTH [%s]: loaded file '%s' with %zd new alarm entries, updated %zd alarms, errors %zd entries, duplicate %zd", host->hostname, filename, loaded, updated, errored, duplicate); + debug(D_HEALTH, "HEALTH [%s]: loaded file '%s' with %zd new alarm entries, updated %zd alarms, errors %zd entries, duplicate %zd", rrdhost_hostname(host), filename, loaded, updated, errored, duplicate); return loaded; } @@ -422,7 +422,7 @@ inline void health_alarm_log_load(RRDHOST *host) { snprintfz(filename, FILENAME_MAX, "%s.old", host->health_log_filename); FILE *fp = fopen(filename, "r"); if(!fp) - error("HEALTH [%s]: cannot open health file: %s", host->hostname, filename); + error("HEALTH [%s]: cannot open health file: %s", rrdhost_hostname(host), filename); else { health_alarm_log_read(host, fp, filename); fclose(fp); @@ -431,7 +431,7 @@ inline void health_alarm_log_load(RRDHOST *host) { host->health_log_entries_written = 0; fp = fopen(host->health_log_filename, "r"); if(!fp) - error("HEALTH [%s]: cannot open health file: %s", host->hostname, host->health_log_filename); + error("HEALTH [%s]: cannot open health file: %s", rrdhost_hostname(host), host->health_log_filename); else { health_alarm_log_read(host, fp, host->health_log_filename); fclose(fp); @@ -443,63 +443,48 @@ inline void health_alarm_log_load(RRDHOST *host) { // health alarm log management inline ALARM_ENTRY* health_create_alarm_entry( - RRDHOST *host, - uint32_t alarm_id, - uint32_t alarm_event_id, - uuid_t config_hash_id, - time_t when, - const char *name, - const char *chart, - const char *chart_context, - const char *family, - const char *class, - const char *component, - const char *type, - const char *exec, - const char *recipient, - time_t duration, - NETDATA_DOUBLE old_value, - NETDATA_DOUBLE new_value, - RRDCALC_STATUS old_status, - RRDCALC_STATUS new_status, - const char *source, - const char *units, - const char *info, - int delay, - uint32_t flags + RRDHOST *host, + uint32_t alarm_id, + uint32_t alarm_event_id, + const uuid_t config_hash_id, + time_t when, + STRING *name, + STRING *chart, + STRING *chart_context, + STRING *family, + STRING *class, + STRING *component, + STRING *type, + STRING *exec, + STRING *recipient, + time_t duration, + NETDATA_DOUBLE old_value, + NETDATA_DOUBLE new_value, + RRDCALC_STATUS old_status, + RRDCALC_STATUS new_status, + STRING *source, + STRING *units, + STRING *info, + int delay, + uint32_t flags ) { debug(D_HEALTH, "Health adding alarm log entry with id: %u", host->health_log.next_log_id); ALARM_ENTRY *ae = callocz(1, sizeof(ALARM_ENTRY)); - ae->name = strdupz(name); - ae->hash_name = simple_hash(ae->name); - - if(chart) { - ae->chart = strdupz(chart); - ae->hash_chart = simple_hash(ae->chart); - } - - if(chart_context) - ae->chart_context = strdupz(chart_context); + ae->name = string_dup(name); + ae->chart = string_dup(chart); + ae->chart_context = string_dup(chart_context); uuid_copy(ae->config_hash_id, *((uuid_t *) config_hash_id)); - if(family) - ae->family = strdupz(family); - - if (class) - ae->classification = strdupz(class); - - if (component) - ae->component = strdupz(component); - - if (type) - ae->type = strdupz(type); - - if(exec) ae->exec = strdupz(exec); - if(recipient) ae->recipient = strdupz(recipient); - if(source) ae->source = strdupz(source); - if(units) ae->units = strdupz(units); + ae->family = string_dup(family); + ae->classification = string_dup(class); + ae->component = string_dup(component); + ae->type = string_dup(type); + ae->exec = string_dup(exec); + ae->recipient = string_dup(recipient); + ae->source = string_dup(source); + ae->units = string_dup(units); ae->unique_id = host->health_log.next_log_id++; ae->alarm_id = alarm_id; @@ -509,27 +494,10 @@ inline ALARM_ENTRY* health_create_alarm_entry( ae->new_value = new_value; char value_string[100 + 1]; - ae->old_value_string = strdupz(format_value_and_unit(value_string, 100, ae->old_value, ae->units, -1)); - ae->new_value_string = strdupz(format_value_and_unit(value_string, 100, ae->new_value, ae->units, -1)); - - char *replaced_info = NULL; - if (likely(info)) { - char *m; - replaced_info = strdupz(info); - size_t pos = 0; - while ((m = strstr(replaced_info + pos, "$family"))) { - char *buf = NULL; - pos = m - replaced_info; - buf = find_and_replace(replaced_info, "$family", (ae->family) ? ae->family : "", m); - freez(replaced_info); - replaced_info = strdupz(buf); - freez(buf); - } - } - - if(replaced_info) ae->info = strdupz(replaced_info); - freez(replaced_info); + ae->old_value_string = string_strdupz(format_value_and_unit(value_string, 100, ae->old_value, ae_units(ae), -1)); + ae->new_value_string = string_strdupz(format_value_and_unit(value_string, 100, ae->new_value, ae_units(ae), -1)); + ae->info = string_dup(info); ae->old_status = old_status; ae->new_status = new_status; ae->duration = duration; @@ -545,7 +513,7 @@ inline ALARM_ENTRY* health_create_alarm_entry( return ae; } -inline void health_alarm_log( +inline void health_alarm_log_add_entry( RRDHOST *host, ALARM_ENTRY *ae ) { @@ -585,26 +553,24 @@ inline void health_alarm_log( } inline void health_alarm_log_free_one_nochecks_nounlink(ALARM_ENTRY *ae) { - freez(ae->name); - freez(ae->chart); - freez(ae->chart_context); - freez(ae->family); - freez(ae->classification); - freez(ae->component); - freez(ae->type); - freez(ae->exec); - freez(ae->recipient); - freez(ae->source); - freez(ae->units); - freez(ae->info); - freez(ae->old_value_string); - freez(ae->new_value_string); + string_freez(ae->name); + string_freez(ae->chart); + string_freez(ae->chart_context); + string_freez(ae->family); + string_freez(ae->classification); + string_freez(ae->component); + string_freez(ae->type); + string_freez(ae->exec); + string_freez(ae->recipient); + string_freez(ae->source); + string_freez(ae->units); + string_freez(ae->info); + string_freez(ae->old_value_string); + string_freez(ae->new_value_string); freez(ae); } inline void health_alarm_log_free(RRDHOST *host) { - rrdhost_check_wrlock(host); - netdata_rwlock_wrlock(&host->health_log.alarm_log_rwlock); ALARM_ENTRY *ae; |