LP2053031: Adding tuning params (#213)

* Adding tuning params
* snapshot_count now accepts a string that can be an integer or auto

---------

Co-authored-by: Homayoon Alimohammadi <[email protected]>

config.yaml

@@ -8,13 +8,33 @@ options:
     default: 2380
     description: Port to run the ETCD Management service
   channel:
+    type: string
+    default: auto
+    description: "The snap channel from which to install etcd (e.g. '3.3/stable'), or 'auto'\nto accept the charm default. Choosing 'auto' will install the latest \nsupported version of etcd at deploy time, but will not automatically upgrade\nto a newer version thereafter.\n"
+  heartbeat_interval:
+    type: int
+    default: 100
+    description: |
+      The frequency with which the leader will notify followers that it is
+      still the leader. By default, etcd uses a 100ms heartbeat interval.
+  election_timeout:
+    type: int
+    default: 1000
+    description: |
+      How long a follower node will go without hearing a heartbeat before
+      attempting to become leader itself.
+      By default, etcd uses a 1000ms election timeout.
+  snapshot_count:
     type: string
     default: auto
     description: |
-      The snap channel from which to install etcd (e.g. '3.3/stable'), or 'auto'
-      to accept the charm default. Choosing 'auto' will install the latest 
-      supported version of etcd at deploy time, but will not automatically upgrade
-      to a newer version thereafter.
+      Creating snapshots with the V2 backend can be expensive,
+      so snapshots are only created after a given number of changes to etcd.
+      The value could be an integer number or auto.
+      if auto, the channel version is checked.
+      Any channel grather than 3.2, use 100000 as value otherwhise 10000 is set
+      If etcd’s memory usage and disk usage are too high,
+      try lowering the snapshot threshold.
   snapd_refresh:
     default: "max"
     type: string

lib/etcd_databag.py

@@ -4,8 +4,12 @@
 from charmhelpers.core.hookenv import leader_get, leader_set
 from charmhelpers.core import unitdata
 from charms.reactive import is_state
-from etcd_lib import get_ingress_address
-from etcd_lib import get_bind_address, build_uri
+from etcd_lib import (
+    get_ingress_address,
+    get_bind_address,
+    build_uri,
+    get_snapshot_count,
+)
 
 import string
 import random
@@ -24,6 +28,9 @@ class EtcdDatabag:
      'cluster_address': '127.0.0.1',
      'db_address': '127.0.0.1',
      'unit_name': 'etcd0',
+     'heartbeat-interval': '100',
+     'election-timeout': '1000',
+     'snapshot-count': '100000',
      'port': '2380',
      'management_port': '2379',
      'ca_certificate': '/etc/ssl/etcd/ca.crt',
@@ -46,6 +53,11 @@ def __init__(self):
             self.listen_client_urls.insert(0, build_uri("http", "127.0.0.1", 4001))
         self.advertise_urls = [build_uri("https", get_ingress_address("db"), self.port)]
         self.management_port = config("management_port")
+        self.heartbeat_interval = config("heartbeat_interval")
+        self.election_timeout = config("election_timeout")
+        self.snapshot_count = get_snapshot_count(
+            config("snapshot_count"), config("channel")
+        )
         # Live polled properties
         self.cluster_address = get_ingress_address("cluster")
         self.unit_name = os.getenv("JUJU_UNIT_NAME").replace("/", "")

lib/etcd_lib.py

@@ -1,4 +1,5 @@
 from ipaddress import ip_address
+from packaging.version import Version
 
 from charmhelpers.contrib.templating.jinja import render
 from charmhelpers.core.hookenv import (
@@ -47,6 +48,29 @@ def get_ingress_address(endpoint_name):
     return sorted(all_addrs, key=lambda i: ip_address(i).version)[0]
 
 
+def get_snapshot_count(snapshot_count: str, channel: str) -> int:
+    """Returns the snapshot count value
+
+    * check if the value is auto,
+        iff channel >=3.2 it will set 100'000 otherwhise it will set 10'000
+    * any other integer value will be set as it is
+
+    @param snapshot_count the value to set, could be a number or auto
+    @param channel the channel used by the charm
+    """
+    SNAPSHOT_COUNT_PRIOR_32 = 10000
+    SNAPSHOT_COUNT_BEYOND_32 = 100000
+    CHANNEL = channel.split("/")[0]
+    if snapshot_count == "auto":
+        if channel == "auto" or Version(CHANNEL) >= Version("3.2"):
+            return SNAPSHOT_COUNT_BEYOND_32
+        return SNAPSHOT_COUNT_PRIOR_32
+    try:
+        return int(snapshot_count)
+    except ValueError:
+        raise TypeError(f"{snapshot_count} value is not an integer number")
+
+
 def get_bind_address(endpoint_name):
     """Returns the first bind-address found in network info
     belonging to the named endpoint, if available.

reactive/etcd.py

@@ -209,7 +209,13 @@ def post_series_upgrade():
 
 @when("snap.installed.etcd")
 @when("leadership.is_leader")
-@when_any("config.changed.port", "config.changed.management_port")
+@when_any(
+    "config.changed.port",
+    "config.changed.management_port",
+    "config.changed.snapshot_count",
+    "config.changed.election_timeout",
+    "config.changed.heartbeat_interval",
+)
 @when_not("etcd.installed")
 @when_not("upgrade.series.in-progress")
 def leader_config_changed():
@@ -247,7 +253,13 @@ def leader_config_changed():
 
 @when("snap.installed.etcd")
 @when_not("leadership.is_leader")
-@when_any("config.changed.port", "config.changed.management_port")
+@when_any(
+    "config.changed.port",
+    "config.changed.management_port",
+    "config.changed.snapshot_count",
+    "config.changed.election_timeout",
+    "config.changed.heartbeat_interval",
+)
 @when_not("etcd.installed")
 def follower_config_changed():
     """Follower units need to render the configuration file, close and open

templates/etcd3.conf

@@ -15,13 +15,13 @@ data-dir: {{ etcd_data_dir }}
 wal-dir: {{ etcd_data_dir }}
 {% endif %}
 # Number of committed transactions to trigger a snapshot to disk.
-snapshot-count: 10000
+snapshot-count: {{ snapshot_count }}
 
 # Time (in milliseconds) of a heartbeat interval.
-heartbeat-interval: 100
+heartbeat-interval: {{ heartbeat_interval }}
 
 # Time (in milliseconds) for an election to timeout.
-election-timeout: 1000
+election-timeout:  {{ election_timeout }}
 
 # Raise alarms when backend size exceeds the given quota. 0 means use the
 # default quota.

tests/unit/lib/test_etcd_databag.py

@@ -43,6 +43,8 @@ def test_render_etcd2(
     config.set("management_port", 1234)
     config.set("port", 5678)
     config.set("bind_with_insecure_http", True)
+    config.set("channel", "3.2/stable")
+    config.set("snapshot_count", "auto")
     bag = etcd_databag.EtcdDatabag()
     template_env = Environment(loader=FileSystemLoader("templates"))
     config = template_env.get_template("etcd2.conf").render(bag.__dict__)
@@ -64,6 +66,8 @@ def test_render_etcd3(
     config.set("management_port", 1234)
     config.set("port", 5678)
     config.set("bind_with_insecure_http", True)
+    config.set("channel", "3.2/stable")
+    config.set("snapshot_count", "auto")
     bag = etcd_databag.EtcdDatabag()
     template_env = Environment(loader=FileSystemLoader("templates"))
     config = template_env.get_template("etcd3.conf").render(bag.__dict__)

tests/unit/lib/test_etcd_lib.py

@@ -5,7 +5,12 @@
 import charmhelpers.core.hookenv as hookenv
 import pytest
 
-from etcd_lib import build_uri, get_bind_address, render_grafana_dashboard
+from etcd_lib import (
+    build_uri,
+    get_bind_address,
+    render_grafana_dashboard,
+    get_snapshot_count,
+)
 
 
 def test_render_grafana_dashboard():
@@ -78,6 +83,24 @@ def test_get_bind_address_picks_v4_first(unit_private_ip):
     unit_private_ip.assert_not_called()
 
 
+def test_get_snapshot_count_auto_3_2():
+    channel = "3.2"
+    snapshot_count = "auto"
+    assert get_snapshot_count(snapshot_count, channel) == 100000
+
+
+def test_get_snapshot_count_auto_3_1():
+    channel = "3.1"
+    snapshot_count = "auto"
+    assert get_snapshot_count(snapshot_count, channel) == 10000
+
+
+def test_get_snapshot_count():
+    channel = "3.1"
+    snapshot_count = "100"
+    assert get_snapshot_count(snapshot_count, channel) == int(snapshot_count)
+
+
 def test_get_bind_address_picks_v6(unit_private_ip):
     ipv6 = "2002::1234:abcd:ffff:c0a8:101"
     bind_data = {

tests/unit/test_etcdctl.py

@@ -35,6 +35,8 @@ def config():
     reactive.etcd.config.reset_mock()
     reactive.etcd.config.side_effect = kv.get
     reactive.etcd.config.set = kv.set
+    reactive.etcd.config.set("channel", "3.2/stable")
+    reactive.etcd.config.set("snapshot_count", "auto")
     return reactive.etcd.config