Feature 10296 Pip Standardization

[osm/LCM.git] / osm_lcm / prometheus.py

diff --git a/osm_lcm/prometheus.py b/osm_lcm/prometheus.py
index c1f49b0..5afa496 100644 (file)
--- a/osm_lcm/prometheus.py
+++ b/osm_lcm/prometheus.py
@@ -24,6 +24,7 @@ import yaml
 import os
 from osm_lcm.lcm_utils import LcmException
 from osm_common.dbbase import DbException
+from osm_lcm.data_utils.database.database import Database
 from jinja2 import Template, TemplateError, TemplateNotFound, TemplateSyntaxError
 
 __author__ = "Alfonso Tierno <alfonso.tiernosepulveda@telefonica.com>"
@@ -53,9 +54,9 @@ class Prometheus:
 
     PROMETHEUS_LOCKED_TIME = 120
 
-    def __init__(self, config, worker_id, db, loop, logger=None):
+    def __init__(self, config, worker_id, loop, logger=None):
         self.worker_id = worker_id
-        self.db = db
+        self.db = Database().instance.db
         self.loop = loop
         self.logger = logger or logging.getLogger("lcm.prometheus")
         self.server = config["uri"]
@@ -101,10 +102,10 @@ class Prometheus:
         :param remove_jobs: list with jobs to remove [job_id_1, job_id_2]
         :return: result. If false prometheus denies this configuration. Exception on error
         """
-        for retry in range(4):
+        for retry in range(20):
             result = True
             if retry:  # first time do not wait
-                await asyncio.sleep(self.PROMETHEUS_LOCKED_TIME / 2, loop=self.loop)
+                await asyncio.sleep(4 + retry, loop=self.loop)
 
             # lock database
             now = time()
@@ -132,9 +133,10 @@ class Prometheus:
                     for job_id in remove_jobs:
                         prometheus_data["scrape_configs"].pop(job_id, None)
                     pull_dict = {"scrape_configs." + job_id: None for job_id in remove_jobs}
-                self.logger.debug(". ".join(log_text_list))
+                self.logger.debug("Updating. " + ". ".join(log_text_list))
 
             if not await self.send_data(prometheus_data):
+                self.logger.error("Cannot update add_jobs: {}. remove_jobs: {}".format(add_jobs, remove_jobs))
                 push_dict = pull_dict = None
                 result = False
 
@@ -155,7 +157,7 @@ class Prometheus:
         del new_config["_id"]
         del new_config["_admin"]
         new_scrape_configs = []
-        
+
         # generate a list with the values of scrape_configs
         for scrape_config in new_config["scrape_configs"].values():
             scrape_config = scrape_config.copy()
@@ -190,7 +192,7 @@ class Prometheus:
                         restore_backup = False
             return True
         except Exception as e:
-            self.logger.error("Error updating prometheus configuration url={}: {}".format(self.server, e))
+            self.logger.error("Error updating configuration url={}: {}".format(self.server, e))
             return False
         finally:
             if restore_backup:
@@ -206,9 +208,14 @@ class Prometheus:
             current_config_yaml = yaml.safe_load(current_config['data']['yaml'])
             current_jobs = [j["job_name"] for j in current_config_yaml["scrape_configs"]]
             expected_jobs = [j["job_name"] for j in expected_config["scrape_configs"]]
-            return current_jobs == expected_jobs
+            if current_jobs == expected_jobs:
+                return True
+            else:
+                self.logger.error("Not all jobs have been loaded. Target jobs: {} Loaded jobs: {}".format(
+                    expected_jobs, current_jobs))
+                return False
         except Exception as e:
-            self.logger.error("Invalid obtained prometheus status. Error: '{}'. Obtained data: '{}'".format(
+            self.logger.error("Invalid obtained status from server. Error: '{}'. Obtained data: '{}'".format(
                 e, current_config))
             # if format is not understood, cannot be compared, assume it is ok
             return True