🎈 perf: Refine counting structure in health check node.

 Refine counting structure in health check node.

2 files changed, 81 insertions, 48 deletions

diff --git a/service/src/monit.c b/service/src/monit.c
index 16bad63..d758c64 100644
--- a/service/src/monit.c
+++ b/service/src/monit.c
@@ -257,7 +257,7 @@ static cJSON * monit_root(struct sc_main * sc)
     cJSON_AddItemToObject(j_root, "shmdev_rx", shmdev_rx_node_monit_loop(sc));
     cJSON_AddItemToObject(j_root, "eth-egress", eth_egress_node_monit_loop(sc));
     cJSON_AddItemToObject(j_root, "health-check-ask", health_check_ask_node_monit_loop(sc));
-    cJSON_AddItemToObject(j_root, "health-check-dealanswer", health_check_deal_answer_node_monit_loop(sc));
+    cJSON_AddItemToObject(j_root, "health-check-deal-answer", health_check_deal_answer_node_monit_loop(sc));
     cJSON_AddItemToObject(j_root, "health-session-state", health_check_session_monit_loop(sc));
     cJSON_AddItemToObject(j_root, "forwarder-table", forwarder_table_monit_loop(sc));
     cJSON_AddItemToObject(j_root, "ef-peer-table", ef_peer_monit_loop(sc));
diff --git a/service/src/node_health_check.c b/service/src/node_health_check.c
index 20fee8b..5ad2b34 100644
--- a/service/src/node_health_check.c
+++ b/service/src/node_health_check.c
@@ -1,4 +1,5 @@
 #include "sc_trace.h"
+#include <bits/stdint-uintn.h>
 #include <rte_graph.h>
 #include <rte_graph_worker.h>
 #include <rte_hash_crc.h>
@@ -45,12 +46,26 @@ struct health_check_ask_stats
     volatile uint64_t buf_alloc_err;
 } __rte_cache_aligned;
 
+/* Health check deal answer exception reason */
+enum health_chk_deal_ans_exc_reason
+{
+    HEALTH_CHK_DEAL_ANS_EXC_RSN_SESS_MISS = 0,
+    HEALTH_CHK_DEAL_ANS_EXC_RSN_TIMEOUT,
+    HEALTH_CHK_DEAL_ANS_EXC_RSN_MAX
+};
+
+/* Health check deal answer exception reason string */
+static const char * health_chk_deal_ans_exc_reason_str[HEALTH_CHK_DEAL_ANS_EXC_RSN_MAX] = {
+    "session lookup miss",
+    "timeout",
+};
+
 /* Health check deal answer stats */
 struct health_check_deal_answer_stats
 {
-    volatile uint64_t deal_pkts;
-    volatile uint64_t session_lookup_miss;
-    volatile uint64_t timeout;
+    volatile uint64_t total_pkts;
+    volatile uint64_t pkts_per_batch;
+    volatile uint64_t excpt_reason[HEALTH_CHK_DEAL_ANS_EXC_RSN_MAX];
 } __rte_cache_aligned;
 
 /* Private health check hdr */
@@ -503,29 +518,50 @@ cJSON * health_check_ask_node_monit_loop(struct sc_main * sc)
 /************************************ Health check deal answer statistics *************************************/
 cJSON * health_check_deal_answer_node_monit_loop(struct sc_main * sc)
 {
-    unsigned int nr_graph_total = sc->nr_io_thread;
     cJSON * json_root = cJSON_CreateObject();
+    unsigned int nr_graphs = sc->nr_io_thread;
 
-    for (uint32_t graph_id = 0; graph_id < nr_graph_total; graph_id++)
+    uint64_t total_pkts[nr_graphs];
+    uint64_t pkts_per_batch[nr_graphs];
+    uint64_t excpt_reason[nr_graphs][HEALTH_CHK_DEAL_ANS_EXC_RSN_MAX];
+
+    for (uint32_t graph_id = 0; graph_id < nr_graphs; graph_id++)
     {
         struct health_check_deal_answer_stats * stats = &deal_answer_stats[graph_id];
 
-        if (stats->deal_pkts == 0)
+        if (stats->total_pkts == 0)
+        {
+            total_pkts[graph_id] = 0;
+            pkts_per_batch[graph_id] = 0;
+
+            memset(excpt_reason[graph_id], 0, sizeof(excpt_reason[graph_id]));
             continue;
+        }
 
-        /* Fill public stats */
-        cJSON * graph_obj = cJSON_CreateObject();
-        cJSON_AddNumberToObject(graph_obj, "deal_pkts", stats->deal_pkts);
-        cJSON_AddNumberToObject(graph_obj, "session_lookup_miss", stats->session_lookup_miss);
-        cJSON_AddNumberToObject(graph_obj, "timeout", stats->timeout);
+        total_pkts[graph_id] = stats->total_pkts;
+        pkts_per_batch[graph_id] = stats->pkts_per_batch;
 
-        /* Fill graph id */
-        char graph_index[MR_STRING_MAX] = {};
-        snprintf(graph_index, sizeof(graph_index), "graph-%u", graph_id);
-        cJSON_AddItemToObject(json_root, graph_index, graph_obj);
+        for (int i = 0; i < HEALTH_CHK_DEAL_ANS_EXC_RSN_MAX; i++)
+        {
+            excpt_reason[graph_id][i] = stats->excpt_reason[i];
+        }
+    }
+
+    cJSON * json_total_pkts = cJSON_CreateIntArray((const int *)total_pkts, nr_graphs);
+    cJSON_AddItemToObject(json_root, "health_chk_deal_ans, total pkts", json_total_pkts);
+
+    cJSON * json_pkts_per_batch = cJSON_CreateIntArray((const int *)pkts_per_batch, nr_graphs);
+    cJSON_AddItemToObject(json_root, "health_chk_deal_ans, pkts per batch", json_pkts_per_batch);
+
+    for (int i = 0; i < HEALTH_CHK_DEAL_ANS_EXC_RSN_MAX; i++)
+    {
+        char str_title[MR_STRING_MAX];
+        snprintf(str_title, sizeof(str_title), "health_chk_deal_ans, %s", health_chk_deal_ans_exc_reason_str[i]);
+
+        cJSON * json_excpt_reason = cJSON_CreateIntArray((const int *)excpt_reason[i], nr_graphs);
+        cJSON_AddItemToObject(json_root, str_title, json_excpt_reason);
     }
 
-    cJSON_AddNumberToObject(json_root, "total_graph_num", nr_graph_total);
     return json_root;
 }
 
@@ -714,43 +750,32 @@ RTE_NODE_REGISTER(health_check_ask_node_base);
 /************************************* Health check deal answer node **************************************/
 
 /* Generate and store the trace information */
-static __rte_always_inline void gen_store_trace_info_answer_node(
-    struct rte_node * node, struct rte_mbuf * mbuf, uint16_t next_node_index,
-    struct health_check_deal_answer_stats * stats, struct health_check_deal_answer_stats * stats_for_trace,
-    struct health_check_session * session)
+static __rte_always_inline void gen_store_trace_info_answer_node(struct rte_node * node, struct rte_mbuf * mbuf,
+                                                                 uint16_t next_node_index,
+                                                                 struct health_check_deal_answer_stats * stats,
+                                                                 struct health_check_session * session,
+                                                                 enum health_chk_deal_ans_exc_reason excpt_reason)
 {
-    struct dp_trace_record_meta meta = {.appsym = MR_TRACE_APPSYM, .module = node->name};
-
     /* Populate the next node infomation */
     char str_record[MR_STRING_MAX];
     int len = snprintf(str_record, sizeof(str_record), "next node:%s", node->nodes[next_node_index]->name);
 
-    /* Populate the lookup result */
-    if (stats_for_trace->session_lookup_miss != stats->session_lookup_miss)
+    if (excpt_reason != HEALTH_CHK_DEAL_ANS_EXC_RSN_MAX)
     {
-        stats_for_trace->session_lookup_miss = stats->session_lookup_miss;
-        len += snprintf(str_record + len, sizeof(str_record) - len, ", missed session lookup");
-    }
-
-    /* Populate the timeout information */
-    if (stats_for_trace->timeout != stats->timeout)
-    {
-        stats_for_trace->timeout = stats->timeout;
-        len += snprintf(str_record + len, sizeof(str_record) - len, ", timeout");
+        len += snprintf(str_record + len, sizeof(str_record) - len, ", %s",
+                        health_chk_deal_ans_exc_reason_str[excpt_reason]);
     }
 
     /* Populate the health check session infomation */
     if (likely(session != NULL))
     {
-        len += snprintf(str_record + len, sizeof(str_record) - len, ", session name:%s", session->name);
-        len += snprintf(str_record + len, sizeof(str_record) - len, ", listening dev:%u,%s", session->port_id,
-                        session->device);
-
-        len += snprintf(str_record + len, sizeof(str_record) - len, ", remote_state:%ld",
-                        rte_atomic64_read(&session->remote_state));
+        len += snprintf(str_record + len, sizeof(str_record) - len,
+                        ", session name:%s, listening dev:%u,%s, remote_state:%ld", session->name, session->port_id,
+                        session->device, rte_atomic64_read(&session->remote_state));
     }
 
     /* Emit the trace record */
+    struct dp_trace_record_meta meta = {.appsym = MR_TRACE_APPSYM, .module = node->name};
     dp_trace_record_emit_str(sc_main_get()->trace, mbuf, rte_lcore_id(), &meta, str_record);
 }
 
@@ -764,7 +789,7 @@ static __rte_always_inline uint16_t health_check_deal_answer_node_process(struct
     struct rte_mbuf ** pkts = (struct rte_mbuf **)objs;
     void ** batch_pkts = objs;
     struct health_check_deal_answer_stats stats = {};
-    struct health_check_deal_answer_stats stats_for_trace = {};
+    enum health_chk_deal_ans_exc_reason excpt_reason = HEALTH_CHK_DEAL_ANS_EXC_RSN_MAX;
 
     /* Single packet processing */
     while (n_left_from > 0)
@@ -785,19 +810,23 @@ static __rte_always_inline uint16_t health_check_deal_answer_node_process(struct
             /* Deal answer */
             int ret = session->deal_answer_func(session, eth_hdr);
             if (unlikely(ret != RT_SUCCESS))
-                stats.timeout++;
+            {
+                excpt_reason = HEALTH_CHK_DEAL_ANS_EXC_RSN_TIMEOUT;
+                stats.excpt_reason[excpt_reason]++;
+            }
         }
         else
         {
             /* Update no match session pkts */
-            stats.session_lookup_miss++;
+            excpt_reason = HEALTH_CHK_DEAL_ANS_EXC_RSN_SESS_MISS;
+            stats.excpt_reason[excpt_reason]++;
         }
 
         /* Check if tracing is enabled for the current Mbuf */
         if (unlikely(dp_trace_record_can_emit(mbuf)))
         {
-            gen_store_trace_info_answer_node(node, mbuf, HEALTH_CHECK_DEAL_ANSWER_NEXT_PKT_DROP, &stats,
-                                             &stats_for_trace, session);
+            gen_store_trace_info_answer_node(node, mbuf, HEALTH_CHECK_DEAL_ANSWER_NEXT_PKT_DROP, &stats, session,
+                                             excpt_reason);
         }
     }
 
@@ -806,9 +835,13 @@ static __rte_always_inline uint16_t health_check_deal_answer_node_process(struct
 
     /* Update graph stats */
     struct health_check_deal_answer_stats * graph_stats = &deal_answer_stats[graph->id];
-    graph_stats->deal_pkts += cnt;
-    graph_stats->session_lookup_miss += stats.session_lookup_miss;
-    graph_stats->timeout += stats.timeout;
+    graph_stats->total_pkts += cnt;
+    graph_stats->pkts_per_batch = cnt;
+
+    for (int i = 0; i < HEALTH_CHK_DEAL_ANS_EXC_RSN_MAX; i++)
+    {
+        graph_stats->excpt_reason[i] += stats.excpt_reason[i];
+    }
 
     return cnt;
 }