From dea58a559f3dbad3dbce3b681639ee89c20b1cee Mon Sep 17 00:00:00 2001
From: zhuofeng <zhuofeng2@huawei.com>
Date: Fri, 20 Sep 2024 14:35:39 +0800
Subject: [PATCH] fix some about collect module and avg block io

---
 config/tasks/avg_block_io.mod                  |  4 ++--
 src/python/sentryCollector/collect_io.py       | 18 +++++++++++-------
 src/python/sentryCollector/collect_plugin.py   | 17 ++++++++---------
 src/python/sentryCollector/collect_server.py   |  6 +++---
 src/python/sentryCollector/collectd.py         |  2 --
 .../sentryPlugins/avg_block_io/avg_block_io.py | 13 ++++++++++---
 6 files changed, 34 insertions(+), 26 deletions(-)

diff --git a/config/tasks/avg_block_io.mod b/config/tasks/avg_block_io.mod
index 814c483..b9b6f34 100644
--- a/config/tasks/avg_block_io.mod
+++ b/config/tasks/avg_block_io.mod
@@ -1,5 +1,5 @@
 [common]
 enabled=yes
 task_start=/usr/bin/python3 /usr/bin/avg_block_io
-task_stop=pkill avg_block_io
-type=oneshot
\ No newline at end of file
+task_stop=pkill -f /usr/bin/avg_block_io
+type=oneshot
diff --git a/src/python/sentryCollector/collect_io.py b/src/python/sentryCollector/collect_io.py
index b826dc4..104b734 100644
--- a/src/python/sentryCollector/collect_io.py
+++ b/src/python/sentryCollector/collect_io.py
@@ -175,8 +175,7 @@ class CollectIo():
 
         threading.Timer(self.period_time, self.task_loop).start()
 
-    def main_loop(self):
-        logging.info("collect io thread start")
+    def is_kernel_avaliable(self):
         base_path = '/sys/kernel/debug/block'
         for disk_name in os.listdir(base_path):
             if not self.loop_all and disk_name not in self.disk_list:
@@ -198,8 +197,13 @@ class CollectIo():
                     self.window_value[disk_name] = {}
                     IO_GLOBAL_DATA[disk_name] = {}
 
-        if len(self.disk_map_stage) == 0:
-            logging.warning("no disks meet the requirements. the thread exits")
+        return len(IO_GLOBAL_DATA) != 0
+
+    def main_loop(self):
+        logging.info("collect io thread start")
+        
+        if not self.is_kernel_avaliable() or len(self.disk_map_stage) == 0:
+            logging.warning("no disks meet the requirements. collect io thread exits")
             return
 
         for disk_name, stage_list in self.disk_map_stage.items():
@@ -213,7 +217,7 @@ class CollectIo():
             start_time = time.time()
 
             if self.stop_event.is_set():
-                logging.info("collect io thread exit")
+                logging.debug("collect io thread exit")
                 return
 
             for disk_name, stage_list in self.disk_map_stage.items():
@@ -227,7 +231,7 @@ class CollectIo():
                 continue
             while sleep_time > 1:
                 if self.stop_event.is_set():
-                    logging.info("collect io thread exit")
+                    logging.debug("collect io thread exit")
                     return
                 time.sleep(1)
                 sleep_time -= 1
@@ -235,5 +239,5 @@ class CollectIo():
 
     # set stop event, notify thread exit
     def stop_thread(self):
-        logging.info("collect io thread is preparing to exit")
+        logging.debug("collect io thread is preparing to exit")
         self.stop_event.set() 
diff --git a/src/python/sentryCollector/collect_plugin.py b/src/python/sentryCollector/collect_plugin.py
index 49ce0a8..9132473 100644
--- a/src/python/sentryCollector/collect_plugin.py
+++ b/src/python/sentryCollector/collect_plugin.py
@@ -142,22 +142,21 @@ def validate_parameters(param, len_limit, char_limit):
         ret =  ResultMessage.RESULT_INVALID_LENGTH
         return [False, ret]
 
-    if len(param) > len_limit:
-        print(f"{param} length more than {len_limit}")
-        ret =  ResultMessage.RESULT_EXCEED_LIMIT
-        return [False, ret]
-
     pattern = r'^[a-zA-Z0-9_-]+$'
     for info in param:
-        if len(info) > char_limit:
-            print(f"{info} length more than {char_limit}")
-            ret =  ResultMessage.RESULT_EXCEED_LIMIT
-            return [False, ret]
         if not re.match(pattern, info):
             print(f"{info} is invalid char")
             ret =  ResultMessage.RESULT_INVALID_CHAR
             return [False, ret]
 
+    # length of len_limit is exceeded, keep len_limit
+    if len(param) > len_limit:
+        print(f"{param} length more than {len_limit}, keep the first {len_limit}")
+        param[:] = param[0:len_limit]
+
+    # only keep elements under the char_limit length
+    param[:] = [elem for elem in param if len(elem) <= char_limit]
+
     return [True, ret]
 
 def is_iocollect_valid(period, disk_list=None, stage=None):
diff --git a/src/python/sentryCollector/collect_server.py b/src/python/sentryCollector/collect_server.py
index fa49781..bab4e56 100644
--- a/src/python/sentryCollector/collect_server.py
+++ b/src/python/sentryCollector/collect_server.py
@@ -256,7 +256,7 @@ class CollectServer():
 
     def server_loop(self):
         """main loop"""
-        logging.info("collect server thread start")
+        logging.info("collect listen thread start")
         server_fd = self.server_fd_create()
         if not server_fd:
             return
@@ -267,7 +267,7 @@ class CollectServer():
         logging.debug("start server_loop loop")
         while True:
             if self.stop_event.is_set():
-                logging.info("collect server thread exit")
+                logging.debug("collect listen thread exit")
                 server_fd = None
                 return
             try:
@@ -281,5 +281,5 @@ class CollectServer():
                 pass
 
     def stop_thread(self):
-        logging.info("collect server thread is preparing to exit")
+        logging.debug("collect listen thread is preparing to exit")
         self.stop_event.set() 
diff --git a/src/python/sentryCollector/collectd.py b/src/python/sentryCollector/collectd.py
index b77c642..3a836df 100644
--- a/src/python/sentryCollector/collectd.py
+++ b/src/python/sentryCollector/collectd.py
@@ -49,7 +49,6 @@ def sig_handler(signum, _f):
         Thread_List[i][0].stop_thread()
 
     remove_sock_file()
-    sys.exit(0)
 
 def main():
     """main
@@ -64,7 +63,6 @@ def main():
     try:
         signal.signal(signal.SIGINT, sig_handler)
         signal.signal(signal.SIGTERM, sig_handler)
-        signal.signal(signal.SIGHUP, sig_handler)
 
         logging.info("finish main parse_args")
 
diff --git a/src/python/sentryPlugins/avg_block_io/avg_block_io.py b/src/python/sentryPlugins/avg_block_io/avg_block_io.py
index ff2071d..73f0b22 100644
--- a/src/python/sentryPlugins/avg_block_io/avg_block_io.py
+++ b/src/python/sentryPlugins/avg_block_io/avg_block_io.py
@@ -21,7 +21,7 @@ CONFIG_FILE = "/etc/sysSentry/plugins/avg_block_io.ini"
 
 def log_invalid_keys(not_in_list, keys_name, config_list, default_list):
     """print invalid log"""
-    if config_list and default_list:
+    if config_list and not_in_list:
         logging.warning("{} in common.{} are not valid, set {}={}".format(not_in_list, keys_name, keys_name, default_list))
     elif config_list == ["default"]:
         logging.warning("Default {} use {}".format(keys_name, default_list))
@@ -144,9 +144,11 @@ def init_io_win(io_dic, config, common_param):
 
                 if avg_lim_value and avg_time_value and tot_lim_value:
                     io_data[disk_name][stage_name][rw]["latency"] = IoWindow(window_size=io_dic["win_size"], window_threshold=io_dic["win_threshold"], abnormal_multiple=avg_time_value, abnormal_multiple_lim=avg_lim_value, abnormal_time=tot_lim_value)
+                    logging.debug("Successfully create {}-{}-{} latency window".format(disk_name, stage_name, rw))
 
                 if iodump_lim_value is not None:
                     io_data[disk_name][stage_name][rw]["iodump"] = IoDumpWindow(window_size=io_dic["win_size"], window_threshold=io_dic["win_threshold"], abnormal_time=iodump_lim_value)
+                    logging.debug("Successfully create {}-{}-{} iodump window".format(disk_name, stage_name, rw))
     return io_data, io_avg_value
 
 
@@ -159,10 +161,10 @@ def get_valid_disk_stage_list(io_dic, config_disk, config_stage):
     for disk_stage_list in json_data.values():
         all_stage_set.update(disk_stage_list)
 
-    disk_list = [key for key in config_disk if key in all_disk_set]
+    disk_list = [key for key in all_disk_set if key in config_disk]
     not_in_disk_list = [key for key in config_disk if key not in all_disk_set]
 
-    stage_list = [key for key in config_stage if key in all_stage_set]
+    stage_list = [key for key in all_stage_set if key in config_stage]
     not_in_stage_list = [key for key in config_stage if key not in all_stage_set]
 
     if not config_disk:
@@ -171,6 +173,9 @@ def get_valid_disk_stage_list(io_dic, config_disk, config_stage):
     if not config_stage:
         stage_list = [key for key in all_stage_set]
 
+    disk_list = disk_list[:10] if len(disk_list) > 10 else disk_list
+    stage_list = stage_list[:15] if len(stage_list) > 15 else stage_list
+
     if config_disk and not disk_list:
         logging.warning("Cannot get valid disk by disk={}, set to default".format(config_disk))
         disk_list, stage_list = get_valid_disk_stage_list(io_dic, [], config_stage)
@@ -228,6 +233,8 @@ def main():
     signal.signal(signal.SIGINT, sig_handler)
     signal.signal(signal.SIGTERM, sig_handler)
 
+    logging.basicConfig(level=logging.INFO)
+
     # 初始化配置读取
     config = configparser.ConfigParser(comment_prefixes=('#', ';'))
     try:
-- 
2.33.0