Implement fine-grained error replies to cloud queries (#12460)

4 files changed, 38 insertions, 9 deletions

diff --git a/aclk/aclk_query.c b/aclk/aclk_query.c
index b30e6be14b..ae56593103 100644
--- a/aclk/aclk_query.c
+++ b/aclk/aclk_query.c
@@ -115,20 +115,20 @@ static int http_api_v2(struct aclk_query_thread *query_thr, aclk_query_t query)
         char *node_uuid = query->data.http_api_v2.query + strlen(NODE_ID_QUERY);
         char nodeid[UUID_STR_LEN];
         if (strlen(node_uuid) < (UUID_STR_LEN - 1)) {
-            error("URL requests node_id but there is not enough chars following. Returning 404 to Cloud.");
+            error_report(CLOUD_EMSG_MALFORMED_NODE_ID);
             retval = 1;
             w->response.code = 404;
-            aclk_http_msg_v2_err(query_thr->client, query->callback_topic, query->msg_id, w->response.code, NULL, 0);
+            aclk_http_msg_v2_err(query_thr->client, query->callback_topic, query->msg_id, w->response.code, CLOUD_EC_MALFORMED_NODE_ID, CLOUD_EMSG_MALFORMED_NODE_ID, NULL, 0);
             goto cleanup;
         }
         strncpyz(nodeid, node_uuid, UUID_STR_LEN - 1);
 
         query_host = node_id_2_rrdhost(nodeid);
         if (!query_host) {
-            error("Host with node_id \"%s\" not found! Returning 404 to Cloud!", node_uuid);
+            error_report("Host with node_id \"%s\" not found! Returning 404 to Cloud!", nodeid);
             retval = 1;
             w->response.code = 404;
-            aclk_http_msg_v2_err(query_thr->client, query->callback_topic, query->msg_id, w->response.code, NULL, 0);
+            aclk_http_msg_v2_err(query_thr->client, query->callback_topic, query->msg_id, w->response.code, CLOUD_EC_NODE_NOT_FOUND, CLOUD_EMSG_NODE_NOT_FOUND, NULL, 0);
             goto cleanup;
         }
     }
@@ -187,7 +187,7 @@ static int http_api_v2(struct aclk_query_thread *query_thr, aclk_query_t query)
                     error("Unknown error during zlib compression.");
                 retval = 1;
                 w->response.code = 500;
-                aclk_http_msg_v2_err(query_thr->client, query->callback_topic, query->msg_id, w->response.code, NULL, 0);
+                aclk_http_msg_v2_err(query_thr->client, query->callback_topic, query->msg_id, w->response.code, CLOUD_EC_ZLIB_ERROR, CLOUD_EMSG_ZLIB_ERROR, NULL, 0);
                 goto cleanup;
             }
             int bytes_to_cpy = NETDATA_WEB_RESPONSE_ZLIB_CHUNK_SIZE - w->response.zstream.avail_out;
diff --git a/aclk/aclk_tx_msgs.c b/aclk/aclk_tx_msgs.c
index 9ade2e12ef..185f5d7968 100644
--- a/aclk/aclk_tx_msgs.c
+++ b/aclk/aclk_tx_msgs.c
@@ -325,12 +325,19 @@ void aclk_send_alarm_metadata(mqtt_wss_client client, int metadata_submitted)
     buffer_free(local_buffer);
 }
 
-void aclk_http_msg_v2_err(mqtt_wss_client client, const char *topic, const char *msg_id, int http_code, const char *payload, size_t payload_len)
+void aclk_http_msg_v2_err(mqtt_wss_client client, const char *topic, const char *msg_id, int http_code, int ec, const char* emsg, const char *payload, size_t payload_len)
 {
     json_object *tmp, *msg;
     msg = create_hdr("http", msg_id, 0, 0, 2);
     tmp = json_object_new_int(http_code);
     json_object_object_add(msg, "http-code", tmp);
+
+    tmp = json_object_new_int(ec);
+    json_object_object_add(msg, "error-code", tmp);
+
+    tmp = json_object_new_string(emsg);
+    json_object_object_add(msg, "error-description", tmp);
+
     if (aclk_send_message_with_bin_payload(client, msg, topic, payload, payload_len)) {
         error("Failed to send cancelation message for http reply");
     }
@@ -355,8 +362,17 @@ void aclk_http_msg_v2(mqtt_wss_client client, const char *topic, const char *msg
     int rc = aclk_send_message_with_bin_payload(client, msg, topic, payload, payload_len);
     json_object_put(msg);
 
-    if (rc)
-        aclk_http_msg_v2_err(client, topic, msg_id, rc, payload, payload_len);
+    switch (rc) {
+    case 403:
+        aclk_http_msg_v2_err(client, topic, msg_id, rc, CLOUD_EC_REQ_REPLY_TOO_BIG, CLOUD_EMSG_REQ_REPLY_TOO_BIG, payload, payload_len);
+        break;
+    case 500:
+        aclk_http_msg_v2_err(client, topic, msg_id, rc, CLOUD_EC_FAIL_TOPIC, CLOUD_EMSG_FAIL_TOPIC, payload, payload_len);
+        break;
+    case 503:
+        aclk_http_msg_v2_err(client, topic, msg_id, rc, CLOUD_EC_SND_TIMEOUT, CLOUD_EMSG_SND_TIMEOUT, payload, payload_len);
+        break;
+    }
 }
 
 void aclk_chart_msg(mqtt_wss_client client, RRDHOST *host, const char *chart)
diff --git a/aclk/aclk_tx_msgs.h b/aclk/aclk_tx_msgs.h
index d8f7978ddd..402f13fb65 100644
--- a/aclk/aclk_tx_msgs.h
+++ b/aclk/aclk_tx_msgs.h
@@ -14,7 +14,7 @@ uint16_t aclk_send_bin_message_subtopic_pid(mqtt_wss_client client, char *msg, s
 void aclk_send_info_metadata(mqtt_wss_client client, int metadata_submitted, RRDHOST *host);
 void aclk_send_alarm_metadata(mqtt_wss_client client, int metadata_submitted);
 
-void aclk_http_msg_v2_err(mqtt_wss_client client, const char *topic, const char *msg_id, int http_code, const char *payload, size_t payload_len);
+void aclk_http_msg_v2_err(mqtt_wss_client client, const char *topic, const char *msg_id, int http_code, int ec, const char* emsg, const char *payload, size_t payload_len);
 void aclk_http_msg_v2(mqtt_wss_client client, const char *topic, const char *msg_id, usec_t t_exec, usec_t created, int http_code, const char *payload, size_t payload_len);
 
 void aclk_chart_msg(mqtt_wss_client client, RRDHOST *host, const char *chart);
diff --git a/aclk/aclk_util.h b/aclk/aclk_util.h
index 1f0ecdd114..7a72020763 100644
--- a/aclk/aclk_util.h
+++ b/aclk/aclk_util.h
@@ -5,6 +5,19 @@
 #include "libnetdata/libnetdata.h"
 #include "mqtt_wss_client.h"
 
+#define CLOUD_EC_MALFORMED_NODE_ID   1
+#define CLOUD_EMSG_MALFORMED_NODE_ID "URL requests node_id but there is not enough chars following (for it to be valid uuid)."
+#define CLOUD_EC_NODE_NOT_FOUND      2
+#define CLOUD_EMSG_NODE_NOT_FOUND    "Node with requested node_id not found"
+#define CLOUD_EC_ZLIB_ERROR          3
+#define CLOUD_EMSG_ZLIB_ERROR        "Error during zlib compression"
+#define CLOUD_EC_REQ_REPLY_TOO_BIG   4
+#define CLOUD_EMSG_REQ_REPLY_TOO_BIG "Request reply produces message bigger than allowed maximum"
+#define CLOUD_EC_FAIL_TOPIC          5
+#define CLOUD_EMSG_FAIL_TOPIC        "Internal Topic Error"
+#define CLOUD_EC_SND_TIMEOUT         6
+#define CLOUD_EMSG_SND_TIMEOUT       "Timeout sending binpacked message"
+
 // Helper stuff which should not have any further inside ACLK dependency
 // and are supposed not to be needed outside of ACLK
 extern int aclk_use_new_cloud_arch;