gala-gopher/repair-stackprobe-caused-cpu-rush.patch

From 8051b4711f4c93e6f6858847555ef0c4c4dd1db5 Mon Sep 17 00:00:00 2001
From: wo_cow <niuqianqian@huawei.com>
Date: Fri, 6 Jan 2023 17:18:07 +0800
Subject: [PATCH] repair stackprobe caused cpu rush

---
 .../ebpf.probe/src/stackprobe/flame_graph.c   | 32 +++++++++++--------
 .../extends/ebpf.probe/src/stackprobe/stack.h |  4 +--
 .../ebpf.probe/src/stackprobe/stackprobe.c    | 16 +++++-----
 .../ebpf.probe/src/stackprobe/stackprobe.h    |  2 +-
 4 files changed, 29 insertions(+), 25 deletions(-)

diff --git a/src/probes/extends/ebpf.probe/src/stackprobe/flame_graph.c b/src/probes/extends/ebpf.probe/src/stackprobe/flame_graph.c
index 126b98d..40c6dcb 100644
--- a/src/probes/extends/ebpf.probe/src/stackprobe/flame_graph.c
+++ b/src/probes/extends/ebpf.probe/src/stackprobe/flame_graph.c
@@ -38,6 +38,8 @@
 #include "bpf.h"
 #include "flame_graph.h"
 
+#define POST_MAX_LEN 131072
+
 struct post_info_s {
     int post_flag;
     int sk;
@@ -193,7 +195,6 @@ static void __reopen_flame_graph_file(struct stack_svg_mng_s *svg_mng)
 #define HISTO_TMP_LEN   (2 * STACK_SYMBS_LEN)
 static char __histo_tmp_str[HISTO_TMP_LEN];
 
-#define POST_MAX_LEN 2048
 static int __do_wr_stack_histo(struct stack_svg_mng_s *svg_mng,
                                struct stack_trace_histo_s *stack_trace_histo, int first, struct post_info_s *post_info)
 {
@@ -246,7 +247,7 @@ static int __build_url(char *url, struct post_server_s *post_server, int en_type
     time_t now, before;
     (void)time(&now);
     if (post_server->last_post_ts == 0) {
-        before = now - 60; // 60s
+        before = now - 30; // 60s
     } else {
         before = post_server->last_post_ts + 1;
     }
@@ -267,12 +268,13 @@ static void __curl_post(struct post_server_s *post_server, struct post_info_s *p
     CURLcode res;
     CURL *curl = post_info->curl;
     if (curl == NULL) {
-        return;
+        goto end2;
     }
+
     long post_len = (long)strlen(post_info->buf_start);
     if (post_len == 0) {
         DEBUG("[FLAMEGRAPH]: buf is null. No need to curl post post to %s\n", appname[en_type]);
-        return;
+        goto end1;
     }
 
     char url[LINE_BUF_LEN] = {0};
@@ -306,26 +308,28 @@ static void __curl_post(struct post_server_s *post_server, struct post_info_s *p
     res = curl_easy_perform(curl);
     /* Check for errors */
     if(res != CURLE_OK) {
-        ERROR("[FLAMEGRAPH]: curl post failed: %s\n", curl_easy_strerror(res));
+        ERROR("[FLAMEGRAPH]: curl post to %s failed: %s\n", url, curl_easy_strerror(res));
     } else {
-        INFO("[FLAMEGRAPH]: curl post post to %s success\n", appname[en_type]);
+        INFO("[FLAMEGRAPH]: curl post post to %s success\n", url, post_info->remain_size);
     }
 
-    /* always cleanup */
-    curl_easy_cleanup(curl);
     if (chunk.memory) {
         free(chunk.memory);
-        chunk.memory = NULL;
     }
-    free(post_info->buf_start);
-    post_info->buf_start = NULL;
-
+end1:
+    /* always cleanup */
+    curl_easy_cleanup(curl);
+end2:
+    if (post_info->buf_start != NULL) {
+        free(post_info->buf_start);
+        post_info->buf_start = NULL;
+    }
     return;
 }
 
 static void __init_curl_handle(struct post_server_s *post_server, struct post_info_s *post_info)
 {
-    if (post_server == NULL || post_server->post_flag == 0) {
+    if (post_server == NULL || post_server->post_enable == 0) {
         return;
     }
 
@@ -419,7 +423,7 @@ int set_post_server(struct post_server_s *post_server, const char *server_str)
     }
 
     curl_global_init(CURL_GLOBAL_ALL);
-    post_server->post_flag = 1;
+    post_server->post_enable = 1;
     post_server->timeout = 3;
     (void)strcpy(post_server->host, server_str);
 
diff --git a/src/probes/extends/ebpf.probe/src/stackprobe/stack.h b/src/probes/extends/ebpf.probe/src/stackprobe/stack.h
index 5a932b1..976ff81 100644
--- a/src/probes/extends/ebpf.probe/src/stackprobe/stack.h
+++ b/src/probes/extends/ebpf.probe/src/stackprobe/stack.h
@@ -19,10 +19,10 @@
 
 #include "common.h"
 
-#define AGGRE_PERIOD    (1 * 60 * 1000) // 1min
+#define AGGRE_PERIOD    (1 * 30 * 1000) // 30s
 #define SAMPLE_PERIOD   (10)            // 10ms
 #define TMOUT_PERIOD    (AGGRE_PERIOD / 1000) // Second as unit
-#define PROC_CACHE_MAX_COUNT    10    // Cache 10 proc symbols
+#define PROC_CACHE_MAX_COUNT    100    // Cache 100 proc symbols
 #define DIV_ROUND_UP(NUM, DEN) ((NUM + DEN - 1) / DEN)
 
 #define PERCPU_SAMPLE_COUNT     (2 * DIV_ROUND_UP(AGGRE_PERIOD, SAMPLE_PERIOD))
diff --git a/src/probes/extends/ebpf.probe/src/stackprobe/stackprobe.c b/src/probes/extends/ebpf.probe/src/stackprobe/stackprobe.c
index a4733e6..9fa5079 100644
--- a/src/probes/extends/ebpf.probe/src/stackprobe/stackprobe.c
+++ b/src/probes/extends/ebpf.probe/src/stackprobe/stackprobe.c
@@ -753,7 +753,7 @@ static void destroy_stack_trace(struct stack_trace_s **ptr_st)
         return;
     }
 
-    if (st->post_server.post_flag) {
+    if (st->post_server.post_enable) {
         clean_post_server();
     }
 
@@ -849,7 +849,7 @@ static struct stack_trace_s *create_stack_trace(StackprobeConfig *conf)
 
     if (set_post_server(&st->post_server, conf->generalConfig->pyroscopeServer) != 0) {
         INFO("[STACKPROBE]: Do not post to Pyroscope Server.\n");
-        st->post_server.post_flag = 0;
+        st->post_server.post_enable = 0;
     } else {
         INFO("[STACKPROBE]: Will post to Pyroscope Server: %s.\n", conf->generalConfig->pyroscopeServer);
     }
@@ -1085,9 +1085,9 @@ static int add_pids()
         // find_bpf_link and add_bpf_link will set bpf_link status
         if (!find_bpf_link(pid)) {
             if (add_bpf_link(pid) != 0) {
-                fprintf(stderr, "add pid %u failed\n", pid);
+                ERROR("[STACKPROBE]: add pid %u failed\n", pid);
             } else {
-                printf("add of pid %u success\n", pid);
+                INFO("[STACKPROBE]: add of pid %u success\n", pid);
             }
         }
     }
@@ -1103,7 +1103,7 @@ static void clear_invalid_pids()
     }
     H_ITER(bpf_link_head, pid_bpf_links, tmp) {
         if (pid_bpf_links->v.pid_state == PID_NOEXIST) {
-            printf("clear bpf link of pid %u\n", pid_bpf_links->pid);
+            INFO("[STACKPROBE]: clear bpf link of pid %u\n", pid_bpf_links->pid);
             H_DEL(bpf_link_head, pid_bpf_links);
             (void)free(pid_bpf_links);
         }
@@ -1344,11 +1344,11 @@ static void init_wr_flame_pthreads(struct svg_stack_trace_s *svg_st, const char
 
     err = pthread_create(&wr_flame_thd, NULL, __running, (void *)svg_st);
     if (err != 0) {
-        fprintf(stderr, "Failed to create %s wr_flame_pthread.\n", flame_name);
+        ERROR("[STACKPROBE]: Failed to create %s wr_flame_pthread.\n", flame_name);
         g_stop = 1;
         return;
     }
-    printf("%s wr_flame_pthread successfully started!\n", flame_name);
+    INFO("[STACKPROBE]: %s wr_flame_pthread successfully started!\n", flame_name);
 
     return;
 }
@@ -1410,7 +1410,7 @@ int main(int argc, char **argv)
     StackprobeConfig *conf;
 
     if (signal(SIGINT, sig_int) == SIG_ERR) {
-        fprintf(stderr, "can't set signal handler: %d\n", errno);
+        ERROR("[STACKPROBE]: can't set signal handler: %d\n", errno);
         return errno;
     }
 
diff --git a/src/probes/extends/ebpf.probe/src/stackprobe/stackprobe.h b/src/probes/extends/ebpf.probe/src/stackprobe/stackprobe.h
index 657b6e7..97e5ea5 100644
--- a/src/probes/extends/ebpf.probe/src/stackprobe/stackprobe.h
+++ b/src/probes/extends/ebpf.probe/src/stackprobe/stackprobe.h
@@ -103,7 +103,7 @@ struct svg_stack_trace_s {
 };
 
 struct post_server_s {
-    char post_flag;
+    char post_enable;
     long timeout; // sec
     char host[PATH_LEN];
     time_t last_post_ts;
-- 
2.33.0