Add alarm status counter api call (#6554)

##### Summary
This is implementation of a prerequisite for the requested feature #6536 (Generate an overall status badge/chart for the health of category)

##### Component Name
web/api/
health/

##### Details
Provide a new, `alarm_count` API call that returns the total number of alarms for given contexts and alarm states. Default is the total number of raised alarms, for all contexts.

6 files changed, 171 insertions, 0 deletions

diff --git a/health/health.h b/health/health.h
index 0bf3b8ffa8..6c000bf458 100644
--- a/health/health.h
+++ b/health/health.h
@@ -58,6 +58,7 @@ extern void *health_main(void *ptr);
 extern void health_reload(void);
 
 extern int health_variable_lookup(const char *variable, uint32_t hash, RRDCALC *rc, calculated_number *result);
+extern void health_aggregate_alarms(RRDHOST *host, BUFFER *wb, BUFFER* context, RRDCALC_STATUS status);
 extern void health_alarms2json(RRDHOST *host, BUFFER *wb, int all);
 extern void health_alarm_log2json(RRDHOST *host, BUFFER *wb, uint32_t after);
 
diff --git a/health/health_json.c b/health/health_json.c
index e923b05c6c..f6ff1b1a74 100644
--- a/health/health_json.c
+++ b/health/health_json.c
@@ -231,6 +231,43 @@ static inline void health_rrdcalc2json_nolock(RRDHOST *host, BUFFER *wb, RRDCALC
 //
 //}
 
+void health_aggregate_alarms(RRDHOST *host, BUFFER *wb, BUFFER* contexts, RRDCALC_STATUS status) {
+    RRDCALC *rc;
+    int numberOfAlarms = 0;
+    char *tok = NULL;
+    char *p = NULL;
+
+    rrdhost_rdlock(host);
+
+    if (contexts) {
+        p = (char*)buffer_tostring(contexts);
+        while(p && *p && (tok = mystrsep(&p, ", |"))) {
+            if(!*tok) continue;
+
+            for(rc = host->alarms; rc ; rc = rc->next) {
+                if(unlikely(!rc->rrdset || !rc->rrdset->last_collected_time.tv_sec))
+                    continue;
+                if(unlikely(rc->rrdset && rc->rrdset->hash_context == simple_hash(tok)
+                            && !strcmp(rc->rrdset->context, tok)
+                            && ((status==RRDCALC_STATUS_RAISED)?(rc->status >= RRDCALC_STATUS_WARNING):rc->status == status)))
+                    numberOfAlarms++;
+            }
+        }
+    }
+    else {
+        for(rc = host->alarms; rc ; rc = rc->next) {
+            if(unlikely(!rc->rrdset || !rc->rrdset->last_collected_time.tv_sec))
+                continue;
+
+            if(unlikely((status==RRDCALC_STATUS_RAISED)?(rc->status >= RRDCALC_STATUS_WARNING):rc->status == status))
+                numberOfAlarms++;
+        }
+    }
+
+    buffer_sprintf(wb, "%d", numberOfAlarms);
+    rrdhost_unlock(host);
+}
+
 void health_alarms2json(RRDHOST *host, BUFFER *wb, int all) {
     int i;
 
diff --git a/web/api/netdata-swagger.json b/web/api/netdata-swagger.json
index 44e862d174..1a0ec86385 100644
--- a/web/api/netdata-swagger.json
+++ b/web/api/netdata-swagger.json
@@ -657,6 +657,61 @@
         }
       }
     },
+    "/alarm_count": {
+      "get": {
+        "summary": "Get an overall status of the chart",
+        "description": "Checks multiple charts with the same context and counts number of alarms with given status.",
+        "parameters": [
+          {
+            "in": "query",
+            "name": "context",
+            "description": "Specify context which should be checked",
+            "required": false,
+            "allowEmptyValue": true,
+            "type": "array",
+            "items": {
+              "type": "string",
+              "collectionFormat": "pipes"
+            },
+            "default": [
+              "system.cpu"
+            ]
+          },
+          {
+            "in": "query",
+            "name": "status",
+            "description": "Specify alarm status to count",
+            "required": false,
+            "allowEmptyValue": true,
+            "type": "string",
+            "enum": [
+              "REMOVED",
+              "UNDEFINED",
+              "UNINITIALIZED",
+              "CLEAR",
+              "RAISED",
+              "WARNING",
+              "CRITICAL"
+            ],
+            "default": "RAISED"
+          }
+        ],
+        "responses": {
+          "200": {
+            "description": "An object containing a count of alarms with given status for given contexts",
+            "schema": {
+              "type": "array",
+              "items": {
+                "type": "number"
+              }
+            }
+          },
+          "500": {
+            "description": "Internal server error. This usually means the server is out of memory."
+          }
+        }
+      }
+    },
     "/manage/health": {
       "get": {
         "summary": "Accesses the health management API to control health checks and notifications at runtime.",
diff --git a/web/api/netdata-swagger.yaml b/web/api/netdata-swagger.yaml
index 0052e646c2..50e6614006 100644
--- a/web/api/netdata-swagger.yaml
+++ b/web/api/netdata-swagger.yaml
@@ -430,6 +430,38 @@ paths:
             type: array
             items:
               $ref: '#/definitions/alarm_log_entry'
+  /alarm_count:
+    get:
+      summary: 'Get an overall status of the chart'
+      description: "Checks multiple charts with the same context and counts number of alarms with given status."
+      parameters:
+        - in: query
+          name: context
+          description: "Specify context which should be checked"
+          required: false
+          allowEmptyValue: true
+          type: array
+          items:
+            type: string
+            collectionFormat: pipes
+          default: ['system.cpu']
+        - in: query
+          name: status
+          description: "Specify alarm status to count"
+          required: false
+          allowEmptyValue: true
+          type: string
+          enum: ['REMOVED', 'UNDEFINED', 'UNINITIALIZED', 'CLEAR', 'RAISED', 'WARNING', 'CRITICAL']
+          default: 'RAISED'
+      responses:
+        '200':
+          description: 'An object containing a count of alarms with given status for given contexts'
+          schema:
+            type: array
+            items: 
+              type: number
+        '500':
+          description: 'Internal server error. This usually means the server is out of memory.'
   /manage/health:
     get:
       summary: 'Accesses the health management API to control health checks and notifications at runtime.'
diff --git a/web/api/web_api_v1.c b/web/api/web_api_v1.c
index 2d12049632..f34f020090 100644
--- a/web/api/web_api_v1.c
+++ b/web/api/web_api_v1.c
@@ -213,6 +213,50 @@ inline int web_client_api_request_v1_alarms(RRDHOST *host, struct web_client *w,
     return 200;
 }
 
+inline int web_client_api_request_v1_alarm_count(RRDHOST *host, struct web_client *w, char *url) {
+    RRDCALC_STATUS status = RRDCALC_STATUS_RAISED;
+    BUFFER *contexts = NULL;
+
+    buffer_flush(w->response.data);
+    buffer_sprintf(w->response.data, "[");
+
+    while(url) {
+        char *value = mystrsep(&url, "&");
+        if(!value || !*value) continue;
+
+        char *name = mystrsep(&value, "=");
+        if(!name || !*name) continue;
+        if(!value || !*value) continue;
+
+        debug(D_WEB_CLIENT, "%llu: API v1 alarm_count query param '%s' with value '%s'", w->id, name, value);
+
+        char* p = value;
+        if(!strcmp(name, "status")) {
+            while ((*p = toupper(*p))) p++;
+            if (!strcmp("CRITICAL", value)) status = RRDCALC_STATUS_CRITICAL;
+            else if (!strcmp("WARNING", value)) status = RRDCALC_STATUS_WARNING;
+            else if (!strcmp("UNINITIALIZED", value)) status = RRDCALC_STATUS_UNINITIALIZED;
+            else if (!strcmp("UNDEFINED", value)) status = RRDCALC_STATUS_UNDEFINED;
+            else if (!strcmp("REMOVED", value)) status = RRDCALC_STATUS_REMOVED;
+            else if (!strcmp("CLEAR", value)) status = RRDCALC_STATUS_CLEAR;
+        }
+        else if(!strcmp(name, "context") || !strcmp(name, "ctx")) {
+            if(!contexts) contexts = buffer_create(255);
+            buffer_strcat(contexts, "|");
+            buffer_strcat(contexts, value);
+        }
+    }
+
+    health_aggregate_alarms(host, w->response.data, contexts, status);
+
+    buffer_sprintf(w->response.data, "]\n");
+    w->response.data->contenttype = CT_APPLICATION_JSON;
+    buffer_no_cacheable(w->response.data);
+
+    buffer_free(contexts);
+    return 200;
+}
+
 inline int web_client_api_request_v1_alarm_log(RRDHOST *host, struct web_client *w, char *url) {
     uint32_t after = 0;
 
@@ -780,6 +824,7 @@ static struct api_command {
         { "alarms",          0, WEB_CLIENT_ACL_DASHBOARD, web_client_api_request_v1_alarms          },
         { "alarm_log",       0, WEB_CLIENT_ACL_DASHBOARD, web_client_api_request_v1_alarm_log       },
         { "alarm_variables", 0, WEB_CLIENT_ACL_DASHBOARD, web_client_api_request_v1_alarm_variables },
+        { "alarm_count",     0, WEB_CLIENT_ACL_DASHBOARD, web_client_api_request_v1_alarm_count     },
         { "allmetrics",      0, WEB_CLIENT_ACL_DASHBOARD, web_client_api_request_v1_allmetrics      },
         { "manage/health",   0, WEB_CLIENT_ACL_MGMT,      web_client_api_request_v1_mgmt_health     },
         // terminator
diff --git a/web/api/web_api_v1.h b/web/api/web_api_v1.h
index 70b7817804..3edb47e3c7 100644
--- a/web/api/web_api_v1.h
+++ b/web/api/web_api_v1.h
@@ -16,6 +16,7 @@ extern int web_client_api_request_v1_alarms(RRDHOST *host, struct web_client *w,
 extern int web_client_api_request_v1_alarm_log(RRDHOST *host, struct web_client *w, char *url);
 extern int web_client_api_request_single_chart(RRDHOST *host, struct web_client *w, char *url, void callback(RRDSET *st, BUFFER *buf));
 extern int web_client_api_request_v1_alarm_variables(RRDHOST *host, struct web_client *w, char *url);
+extern int web_client_api_request_v1_alarm_count(RRDHOST *host, struct web_client *w, char *url);
 extern int web_client_api_request_v1_charts(RRDHOST *host, struct web_client *w, char *url);
 extern int web_client_api_request_v1_chart(RRDHOST *host, struct web_client *w, char *url);
 extern int web_client_api_request_v1_data(RRDHOST *host, struct web_client *w, char *url);