Merge pull request #882 from netenglabs/poller-status-normalize

Poller status normalize

Author: ddutt

suzieq/config/schema/sqPoller.avsc

@@ -21,7 +21,7 @@
                 }
             },
             "display": 4,
-            "description": "[min, max, avg] of time taken to get data, computed over the greater between the last 5 mins and the polling period"
+            "description": "[min, max, avg] of time(ms) taken to get data, computed over the greater between the last 5 mins and the polling period"
         },
         {
             "name": "totalTime",
@@ -33,7 +33,7 @@
                 }
             },
             "display": 5,
-            "description": "[min, max, avg] of time taken to get & process data,computed over the greater between the last 5 mins and the polling period"
+            "description": "[min, max, avg] of time(ms) taken to get & process data,computed over the greater between the last 5 mins and the polling period"
         },
         {
             "name": "svcQsize",
@@ -45,7 +45,7 @@
                 }
             },
             "display": 6,
-            "description": "[min, max, avg] of service queue size, computed over the greater between the last 5 mins and the polling period"
+            "description": "[min, max, avg] of service queue length, computed over the greater between the last 5 mins and the polling period"
         },
         {
             "name": "wrQsize",
@@ -57,7 +57,7 @@
                 }
             },
             "display": 7,
-            "description": "[min, max, avg] of write queue size, computed over the greater between the last 5 mins and the polling period"
+            "description": "[min, max, avg] of write queue length, computed over the greater between the last 5 mins and the polling period"
         },
         {
             "name": "nodeQsize",
@@ -69,7 +69,7 @@
                 }
             },
             "display": 8,
-            "description": "[min, max, avg] of node queue size, computed over the greater between the last 5 mins and the polling period"
+            "description": "[min, max, avg] of node queue length, computed over the greater between the last 5 mins and the polling period"
         },
         {
             "name": "rxBytes",
@@ -97,8 +97,14 @@
         {
             "name": "status",
             "type": "long",
+            "description": "Service polling status code. 0 or 200 means successfully polled, -1 means time timeout, other values mean failed to poll"
+        },
+        {
+            "name": "statusStr",
+            "type": "string",
             "display": 3,
-            "description": "Service polling status. 0 or 200 means successfully polled, -1 means time timeout, other values mean failed to poll"
+            "description": "Service polling status string",
+	    "depends": "status"
         },
         {
             "name": "nodesPolledCnt",

suzieq/engines/pandas/sqPoller.py

@@ -1,3 +1,4 @@
+from suzieq.shared.poller_error_codes import SqPollerStatusCode
 from suzieq.engines.pandas.engineobj import SqPandasEngine
 
 
@@ -16,8 +17,12 @@ def get(self, **kwargs):
         poll_period_exceeded = kwargs.pop('pollExcdPeriodCount', '')
         columns = kwargs.pop('columns', [])
         addnl_fields = kwargs.pop('addnl_fields', [])
+        user_query = kwargs.pop('query_str', '')
 
         fields = self.schema.get_display_fields(columns)
+        user_query_cols = self._get_user_query_cols(user_query)
+        addnl_fields += [x for x in user_query_cols if x not in addnl_fields]
+
         if status and 'status' not in fields:
             addnl_fields.append('status')
 
@@ -50,9 +55,13 @@ def get(self, **kwargs):
             else:
                 add_filter += f'{add_prefix}pollExcdPeriodCount == 0'
 
+        if 'statusStr' in fields:
+            df['statusStr'] = df.status.apply(self._get_status_str)
+
         if not df.empty and add_filter:
             df = df.query(add_filter).reset_index(drop=True)
 
+        df = self._handle_user_query_str(df, user_query)
         return df.reset_index(drop=True)[fields]
 
     def summarize(self, **kwargs):
@@ -72,3 +81,12 @@ def summarize(self, **kwargs):
         ]
 
         return super().summarize(**kwargs)
+
+    @staticmethod
+    def _get_status_str(code: int) -> str:
+        '''Function to catch exception'''
+        try:
+            # pylint: disable=no-value-for-parameter
+            return SqPollerStatusCode(code).errstr
+        except ValueError:
+            return f'{code}'

suzieq/shared/poller_error_codes.py

@@ -0,0 +1,63 @@
+# List of normalized error codes used in the poller
+
+from enum import IntEnum
+
+# Ensure you never copy backwards compat values
+
+
+class SqPollerStatusCode(IntEnum):
+    '''SuzieQ Poller Status Codes and their strings'''
+
+    OK = 0, 'OK'
+    HTTP_OK = 200, 'OK'
+    PERMISSION_DENIED = 1001, 'Permission Denied'
+    UNAUTHORIZED = 1002, 'Unauthorized'
+    LOGIN_FAIL = 1003, 'Login Failed'
+    PRIV_ESC_FAIL = 1004, 'Privilege Escalation Failed'
+    NON_FINAL_FAILURES = 1005, 'Dummy to mark boundary, never set'
+    CMD_NOT_FOUND = 2006, 'Command Not Found'
+    EMPTY_OUTPUT = 2007, 'Output Empty'
+    INCOMPLETE_CMD = 2008, 'Command Incomplete'
+    TIMEOUT = 2009, 'Command Timeout'
+    DISABLED_FEATURE = 2010, 'Feature Disabled'
+    CONNECT_FAIL = 2011, 'Connect Failed'
+    PARSING_ERROR = 2012, 'Parsing Failed'
+    UNSUPPORTED_SVC = 2013, 'Service Not Supported'
+    UNSUPPORTED_CMD = 2014, 'Command Not Supported'
+    NO_RT_TO_HOST = 2015, 'No Route to Host'
+    VAR_NOT_SET = 2016, 'Variable Not Set'
+    WAITING_FOR_VAR = 2017, 'Variable Waiting'
+    ERR_OUTPUT = 2018, 'Command Failed'
+    GEN_CLS_FAIL = 2019, 'Linux Cmd Failed'
+    UNREACH_NET = 2020, 'Network Unreachable'
+    NO_SVC_DEF = 2021, 'Service Not Defined'
+    UNKNOWN_FAILURE = 2022, 'Unknown Failure'
+    CONN_REFUSED = 2023, 'Connection Refused'
+    UNEXPECTED_DISCONNECT = 2024, 'Unexpected Disconnect'
+
+    # Everything below this is for backward compat
+    LINUX_OLD_CMD_NOT_FOUND = 1, 'Command Not Found'
+    NXOS_CMD_NOT_FOUND = 16, 'Command Not Found'
+    NXOS_YACNF = 20, 'Command Not Found'
+    LINUX_CMD_NOT_FOUND = 127, 'Command Not Found'
+    HTTP_EMPTY_OUTPUT = 204, 'Empty Output'
+    OLD_EMPTY_OUTPUT = 244, 'Empty Output'
+    HTTP_INCOMPLETE_CMD = 400, 'Incomplete Command'
+    HTTP_LOGIN_FAIL = 401, 'Login Failed'
+    HTTP_BAD_CMD = 404, 'Command Not Found'
+    HTTP_YAIC = 405, 'Incomplete Command'
+    HTTP_TIMEOUT = 408, 'Timeout'
+    OLD_FEATURE_DISABLED = 418, 'Feature Disabled'
+    OLD_PARSING_ERROR = 502, 'Parsing Error'
+    OLD_PRIV_ESC_FAIL = -1, 'Privilege Escalation Failed'
+
+    def __new__(cls, code: int, errstr: str):
+        obj = int.__new__(cls, code)
+        obj._value_ = code
+        obj.errstr = errstr
+
+        return obj
+
+    def is_error_code_final(self):
+        '''is the node in stopped state'''
+        return (self.value < self.NON_FINAL_FAILURES.value)