Updates branch sol006 with master
[osm/LCM.git] / osm_lcm / prometheus.py
diff --git a/osm_lcm/prometheus.py b/osm_lcm/prometheus.py
new file mode 100644 (file)
index 0000000..2be94af
--- /dev/null
@@ -0,0 +1,220 @@
+# -*- coding: utf-8 -*-
+
+##
+# Copyright 2020 Telefonica S.A.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"); you may
+# not use this file except in compliance with the License. You may obtain
+# a copy of the License at
+#
+#         http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+# License for the specific language governing permissions and limitations
+# under the License.
+##
+
+import asyncio
+from time import time
+import logging
+import aiohttp
+import yaml
+import os
+from osm_lcm.lcm_utils import LcmException
+from osm_common.dbbase import DbException
+from jinja2 import Template, TemplateError, TemplateNotFound, TemplateSyntaxError
+
+__author__ = "Alfonso Tierno <alfonso.tiernosepulveda@telefonica.com>"
+
+initial_prometheus_data = {
+    "_id": "prometheus",
+    "_admin": {
+        "locked_at": 0,
+        "locked_by": None,
+        "modified": 1593445184,  # 2020-06-29
+        "created": 1593445184,
+        "version": "1.0"    # to allow future version updates
+    },
+    'scrape_configs': {   # Dictionary at database. Converted to list before sending to prometheus
+        'mon_exporter': {'static_configs': [{'targets': ['mon:8000']}], 'job_name': 'mon_exporter'},
+    },
+    'global': {'evaluation_interval': '15s', 'scrape_interval': '15s'},
+    'rule_files': None,
+    'alerting': {'alertmanagers': [{'static_configs': [{'targets': None}]}]}
+}
+
+
+class Prometheus:
+    """
+    Implements a class to update Prometheus
+    """
+
+    PROMETHEUS_LOCKED_TIME = 120
+
+    def __init__(self, config, worker_id, db, loop, logger=None):
+        self.worker_id = worker_id
+        self.db = db
+        self.loop = loop
+        self.logger = logger or logging.getLogger("lcm.prometheus")
+        self.server = config["uri"]
+        self.path = config["path"]
+        if not self.path.endswith("/"):
+            self.path += "/"
+        self.cfg_file = self.path + "prometheus.yml"
+        self.cfg_file_backup = self.path + "prometheus.yml-backup"
+
+    @staticmethod
+    def parse_job(job_data: str, variables: dict) -> dict:
+        try:
+            template = Template(job_data)
+            job_parsed = template.render(variables or {})
+            return yaml.safe_load(job_parsed)
+        except (TemplateError, TemplateNotFound, TemplateSyntaxError) as e:
+            # TODO yaml exceptions
+            raise LcmException("Error parsing Jinja2 to prometheus job. job_data={}, variables={}. Error={}".format(
+                job_data, variables, e))
+
+    async def start(self):
+        for retry in range(4):
+            try:
+                # self.logger("Starting prometheus ")
+                # read from database
+                prometheus_data = self.db.get_one("admin", {"_id": "prometheus"}, fail_on_empty=False)
+                if not prometheus_data:
+                    self.logger.info("Init db.admin.prometheus content")
+                    self.db.create("admin", initial_prometheus_data)
+                # send database config file to prometheus. Ignore loading errors, as prometheus may be starting
+                # but at least an initial configuration file is set
+                await self.update()
+                return
+            except DbException as e:
+                if retry == 3:
+                    raise LcmException("Max retries trying to init prometheus configuration: {}".format(e))
+                await asyncio.sleep(5, loop=self.loop)
+
+    async def update(self, add_jobs: dict = None, remove_jobs: list = None) -> bool:
+        """
+
+        :param add_jobs: dictionary with {job_id_1: job_content, job_id_2: job_content}
+        :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(20):
+            result = True
+            if retry:  # first time do not wait
+                await asyncio.sleep(4 + retry, loop=self.loop)
+
+            # lock database
+            now = time()
+            if not self.db.set_one(
+                    "admin",
+                    q_filter={"_id": "prometheus", "_admin.locked_at.lt": now - self.PROMETHEUS_LOCKED_TIME},
+                    update_dict={"_admin.locked_at": now, "_admin.locked_by": self.worker_id},
+                    fail_on_empty=False):
+                continue
+            # read database
+            prometheus_data = self.db.get_one("admin", {"_id": "prometheus"})
+            update_dict = {"_admin.locked_at": 0,
+                           "_admin.locked_by": None}
+
+            # Make changes from prometheus_incremental
+            push_dict = pull_dict = None
+            if add_jobs or remove_jobs:
+                log_text_list = []
+                if add_jobs:
+                    log_text_list.append("adding jobs: {}".format(list(add_jobs.keys())))
+                    prometheus_data["scrape_configs"].update(add_jobs)
+                    push_dict = {"scrape_configs." + job_id: job_data for job_id, job_data in add_jobs.items()}
+                elif remove_jobs:
+                    log_text_list.append("removing jobs: {}".format(list(remove_jobs)))
+                    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("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
+
+            # unblock database
+            if push_dict:
+                update_dict.update(push_dict)
+            if push_dict or pull_dict:
+                update_dict["_admin.modified_at"] = now
+            if not self.db.set_one(
+                    "admin", {"_id": "prometheus", "_admin.locked_at": now, "_admin.locked_by": self.worker_id},
+                    update_dict=update_dict, unset=pull_dict, fail_on_empty=False):
+                continue
+            return result
+        raise LcmException("Cannot update prometheus database. Reached max retries")
+
+    async def send_data(self, new_config):
+        restore_backup = False
+        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()
+            # remove nsr_id metadata from scrape_configs
+            scrape_config.pop("nsr_id", None)
+            new_scrape_configs.append(scrape_config)
+        new_config["scrape_configs"] = new_scrape_configs
+
+        try:
+            if os.path.exists(self.cfg_file):
+                os.rename(self.cfg_file, self.cfg_file_backup)
+                restore_backup = True
+            with open(self.cfg_file, "w+") as f:
+                yaml.safe_dump(new_config, f, indent=4, default_flow_style=False)
+            # self.logger.debug("new configuration: {}".format(yaml.safe_dump(new_config, indent=4,
+            #                                                                 default_flow_style=False)))
+            async with aiohttp.ClientSession() as session:
+                async with session.post(self.server + "-/reload") as resp:
+                    if resp.status > 204:
+                        raise LcmException(await resp.text())
+                await asyncio.sleep(5, loop=self.loop)
+                # If prometheus does not admit this configuration, remains with the old one
+                # Then, to check if the configuration has been accepted, get the configuration from prometheus
+                # and compares with the inserted one
+                async with session.get(self.server + "api/v1/status/config") as resp:
+                    if resp.status > 204:
+                        raise LcmException(await resp.text())
+                    current_config = await resp.json()
+                    if not self._check_configuration_equal(current_config, new_config):
+                        return False
+                    else:
+                        restore_backup = False
+            return True
+        except Exception as e:
+            self.logger.error("Error updating configuration url={}: {}".format(self.server, e))
+            return False
+        finally:
+            if restore_backup:
+                try:
+                    os.rename(self.cfg_file_backup, self.cfg_file)
+                except Exception as e:
+                    self.logger.critical("Exception while rolling back: {}".format(e))
+
+    def _check_configuration_equal(self, current_config, expected_config):
+        try:
+            # self.logger.debug("Comparing current_config='{}' with expected_config='{}'".format(current_config,
+            #                                                                                    expected_config))
+            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"]]
+            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 status from server. Error: '{}'. Obtained data: '{}'".format(
+                e, current_config))
+            # if format is not understood, cannot be compared, assume it is ok
+            return True