[timeseries] Add index lifecycle policy to elasticsearch

openwisp_monitoring/db/backends/elasticsearch/client.py

@@ -17,6 +17,7 @@
 from .. import TIMESERIES_DB
 from .index import MetricIndex, Point, find_metric
 from .queries import default_chart_query, math_map, operator_lookup
+from .retention_policies import _make_policy, default_rp_policy
 
 logger = logging.getLogger(__name__)
 
@@ -70,7 +71,10 @@ def __init__(self, db_name='metric'):
     def create_database(self):
         """ creates connection to elasticsearch """
         connections.create_connection(hosts=[TIMESERIES_DB['HOST']])
-        self.get_db
+        db = self.get_db
+        # Skip if support for Index Lifecycle Management is disabled or no privileges
+        self.ilm_enabled = db.ilm.start()['acknowledged']
+        self.create_or_alter_retention_policy(name='default')
 
     def drop_database(self):
         """ deletes all indices """
@@ -81,27 +85,48 @@ def drop_database(self):
     @cached_property
     def get_db(self):
         """ Returns an ``Elasticsearch Client`` instance """
-        # TODO: AUTHENTICATION remains see `SecurityClient`
         return Elasticsearch(
             [f"{TIMESERIES_DB['HOST']}:{TIMESERIES_DB['PORT']}"],
             http_auth=(TIMESERIES_DB['USER'], TIMESERIES_DB['PASSWORD']),
             retry_on_timeout=True,
         )
 
-    def create_or_alter_retention_policy(self, name, duration):
-        """ creates or alters existing retention policy if necessary """
-        # TODO
-        pass
+    def create_or_alter_retention_policy(self, name, duration=None):
+        """
+        creates or alters existing retention policy if necessary
+
+        Note: default retention policy can't be altered with this function
+        """
+        if not self.ilm_enabled:
+            return
+        ilm = self.get_db.ilm
+        if not duration:
+            ilm.put_lifecycle(policy=name, body=default_rp_policy)
+            return
+        days = f'{int(duration.split("h")[0]) // 24}d'
+        duration_changed = False
+        try:
+            policy = ilm.get_lifecycle('default')
+            exists = True
+            current_duration = policy['default']['policy']['phases']['hot']['actions'][
+                'rollover'
+            ]['max_age']
+            duration_changed = current_duration != days
+        except NotFoundError:
+            exists = False
+        if not exists or duration_changed:
+            policy = _make_policy(days)
+            ilm.put_lifecycle(policy=name, body=policy)
 
     def query(self, query, precision=None):
         index = query.pop('key')
         return Search(index=index).from_dict(query).execute().to_dict()
 
     def write(self, name, values, **kwargs):
-        # TODO: Add support for retention policy
+        rp = kwargs.get('retention_policy')
         tags = kwargs.get('tags')
         timestamp = kwargs.get('timestamp')
-        metric_id = find_metric(name, tags, add=True)
+        metric_id = find_metric(self.get_db, name, tags, rp, add=True)
         metric_index = MetricIndex().get(metric_id, index=name)
         point = Point(fields=values, time=timestamp or datetime.now())
         metric_index.points.append(point)
@@ -111,13 +136,10 @@ def read(self, key, fields, tags, limit=1, order='-time', **kwargs):
         extra_fields = kwargs.get('extra_fields')
         time_format = kwargs.get('time_format')
         # since = kwargs.get('since')
-        metric_id = find_metric(key, tags)
+        metric_id = find_metric(self.get_db, key, tags)
         if not metric_id:
-            return list()
-        try:
-            metric_index = MetricIndex().get(metric_id, index=key)
-        except NotFoundError:
             return []
+        metric_index = self.get_db.get(index=key, id=metric_id)
         if order == 'time':
             points = list(metric_index.points[0:limit])
         elif order == '-time':
@@ -127,33 +149,27 @@ def read(self, key, fields, tags, limit=1, order='-time', **kwargs):
                 f'Invalid order "{order}" passed.\nYou may pass "time" / "-time" to get '
                 'result sorted in ascending /descending order respectively.'
             )
-        if not points:
-            return list()
         # distinguish between traffic and clients
         for point in list(points):
             if fields not in point.fields.to_dict():
                 points.remove(point)
         if extra_fields and extra_fields != '*':
             assert isinstance(extra_fields, list)
-            _points = []
-            for point in points:
-                point = point.to_dict()
-                _point = {
+            for count, point in enumerate(points):
+                fields_dict = point.to_dict()['fields']
+                point = {
                     'time': self._format_time(point['time'], time_format),
-                    fields: point['fields'][fields],
+                    fields: fields_dict[fields],
                 }
                 for extra_field in extra_fields:
-                    if point['fields'].get(extra_field) is not None:
-                        _point.update({extra_field: point['fields'][extra_field]})
-                _points.append(_point)
-            points = _points
+                    if fields_dict.get(extra_field) is not None:
+                        point.update({extra_field: fields_dict[extra_field]})
+                points[count] = point
         elif extra_fields == '*':
-            points = [
-                deep_merge_dicts(
-                    p.fields.to_dict(), {'time': self._format_time(p.time, time_format)}
+            for count, point in enumerate(points):
+                points[count] = deep_merge_dicts(
+                    point.fields.to_dict(), {'time': self._format_time(point.time, time_format)}
                 )
-                for p in points
-            ]
         else:
             points = [
                 deep_merge_dicts(
@@ -210,12 +226,14 @@ def _fill_points(self, query, points):
 
     def delete_metric_data(self, key=None, tags=None):
         """
-        deletes a specific metric based on the key and tags
-        provided, you may also choose to delete all metrics
+        deletes a specific metric based on given key and tags;
+        deletes all metrics if neither provided
         """
         if key and tags:
-            metric_id = find_metric(key, tags)
+            metric_id = find_metric(self.get_db, key, tags)
             self.get_db.delete(index=key, id=metric_id)
+        elif key:
+            self.get_db.indices.delete(index=key, ignore=[400, 404])
         else:
             self.get_db.indices.delete(index='*', ignore=[400, 404])
 
@@ -317,7 +335,7 @@ def default_chart_query(self, tags):
         return q
 
 
-# Old data - delete by query (inefficient) / retention policy - Index lifecycle management
+# TODO:
 # Fix Average - currently it's computing average over all fields!
 # Time Interval - fix range
 # Device query

openwisp_monitoring/db/backends/elasticsearch/index.py

@@ -1,12 +1,14 @@
 import uuid
 
 from django.conf import settings
-from elasticsearch import Elasticsearch
 from elasticsearch.exceptions import NotFoundError
 from elasticsearch.helpers import bulk
 from elasticsearch_dsl import Date, Document, InnerDoc, Nested, Q, Search
 from swapper import load_model
 
+# TODO: Remove this. Current Structure
+# Index Templates <-- Indexes <-- Documents <-- Points
+
 
 class Point(InnerDoc):
     time = Date(required=True, default_timezone=settings.TIME_ZONE)
@@ -15,20 +17,19 @@ class Point(InnerDoc):
 
 class MetricIndex(Document):
     tags = Nested(dynamic=True, required=True, multi=True)
-    # returns an empty list if not present
     points = Nested(Point)
 
     class Index:
-        # name gets replaced with metric's key
         name = 'metric'
         settings = {
-            "number_of_shards": 1,
-            "number_of_replicas": 0,
+            'number_of_shards': 1,
+            'number_of_replicas': 0,
+            'lifecycle.name': 'default',
+            'lifecycle.rollover_alias': 'metric',
         }
 
 
-def find_metric(index, tags, add=False):
-    client = Elasticsearch()
+def find_metric(client, index, tags, retention_policy=None, add=False):
     search = Search(using=client, index=index)
     if tags:
         tags_dict = dict()
@@ -40,28 +41,56 @@ def find_metric(index, tags, add=False):
     try:
         return list(search.query(q).execute())[0].meta['id']
     except (NotFoundError, AttributeError, IndexError):
-        return add_index(index, tags)['_id'] if add else None
+        return (
+            add_doc(client, index, tags, retention_policy=retention_policy)['_id']
+            if add
+            else None
+        )
 
 
-def add_index(key, tags, id=None):
+def add_doc(client, key, tags, _id=None, retention_policy=None):
     """
     Add index to elasticsearch using ``keys``, ``tags`` and ``id`` provided.
     If no ``id`` is provided a random ``uuid`` would be used.
     """
-    obj = MetricIndex(meta={'id': id or uuid.uuid1()}, tags=tags)
-    obj.meta.index = key
-    obj.save()
+    _id = str(_id or uuid.uuid1())
+    # Check if index exists
+    if client.indices.exists(index=key):
+        client.create(index=key, id=_id, body={'tags': tags})
+        return {'_id': _id}
+    # Create a new index if it doesn't exist
+    name = f'{key}-000001'
+    obj = MetricIndex()
+    obj._index = obj._index.clone(name)
+    # Create a new index template if it doesn't exist
+    if not client.indices.exists_template(name=key):
+        obj._index.settings(**{'lifecycle.rollover_alias': key})
+        if retention_policy:
+            obj._index.settings(**{'lifecycle.name': retention_policy})
+        # index pattern is added for Index Lifecycle Management
+        obj._index.as_template(key, f'{key}-*').save(using=client)
+    obj.init(using=client, index=name)
+    obj.meta = {'id': _id, 'index': name}
+    obj.tags = tags
+    obj.save(using=client, index=name)
+    if retention_policy:
+        client.indices.put_settings(
+            body={'lifecycle.name': retention_policy}, index=key
+        )
+    client.indices.put_settings(body={'lifecycle.rollover_alias': key}, index=name)
+    client.indices.put_alias(index=name, name=key, body={'is_write_index': True})
     return obj.to_dict(include_meta=True)
 
 
 def bulk_indexing():
     """ Index all existing metrics """
+    from ... import timeseries_db
+
     Metric = load_model('monitoring', 'Metric')
-    MetricIndex.init()
-    es = Elasticsearch()
     bulk(
-        client=es,
+        client=timeseries_db.get_db,
         actions=(
-            add_index(m.key, m.tags, m.id) for m in Metric.objects.all().iterator()
+            add_doc(timeseries_db.get_db, m.key, m.tags, m.id)
+            for m in Metric.objects.all().distinct('key').iterator()
         ),
     )

openwisp_monitoring/db/backends/elasticsearch/retention_policies.py

@@ -0,0 +1,41 @@
+# By default age is calculated from the date the index is created but if the
+# index has been rolled over than the rollover date is used to calculate the age
+
+default_rp_policy = {
+    'policy': {
+        'phases': {
+            'hot': {
+                'actions': {
+                    'rollover': {'max_age': '30d', 'max_size': '90G'},
+                    'set_priority': {'priority': 100},
+                }
+            },
+            'warm': {
+                'min_age': '30d',
+                'actions': {
+                    'forcemerge': {'max_num_segments': 1},
+                    'allocate': {'number_of_replicas': 0},
+                    'set_priority': {'priority': 50},
+                },
+            },
+            'cold': {'min_age': '150d', 'actions': {'freeze': {}}},
+            'delete': {'min_age': '335d', 'actions': {'delete': {}}},
+        }
+    }
+}
+
+
+def _make_policy(max_age):
+    return {
+        'policy': {
+            'phases': {
+                'hot': {
+                    'actions': {
+                        'rollover': {'max_age': max_age},
+                        'set_priority': {'priority': 100},
+                    }
+                },
+                'delete': {'actions': {'delete': {}}},
+            }
+        }
+    }

openwisp_monitoring/monitoring/tests/__init__.py

@@ -16,6 +16,7 @@
 Metric = load_model('monitoring', 'Metric')
 AlertSettings = load_model('monitoring', 'AlertSettings')
 
+# TODO: Queries relating to top_fields aren't yet
 # this custom chart configuration is used for automated testing purposes
 charts = {
     'histogram': {
@@ -125,7 +126,8 @@
                 "SELECT {fields|MEAN} FROM {key} "
                 "WHERE time >= '{time}' AND content_type = "
                 "'{content_type}' AND object_id = '{object_id}'"
-            )
+            ),
+            'elasticsearch': _make_query(),
         },
     },
 }