this old ceilometer

Gnocchi is a great solution for large scale time series storage and in combination with Ceilometer, it provides a solid solution for tracking resources across your OpenStack environment that can be extended to solve your use cases such as billing or monitoring.

The above statement is correct for Gnocchi4. It’s even more accurate when speaking about Gnocchi4.1 which provides even better performance, more flexible query syntax, and less resource usage. That said, the prior versions of Gnocchi, like all things, are not as great. Gnocchi3.x is arguably only a good solution unless you acquire an in-depth knowledge of the code and Gnocchi2.x is arguably unusable for any medium or large cloud deployment.

So the obvious solution here is to just upgrade to Gnocchi4 and your monitoring solutions are resolved!… except depending on the packages you’re using, the available version of Gnocchi might be something unusable such as Gnocchi2.x.

As I currently have a Mitaka deployment running, let’s hack that to support Gnocchi4.1

What is known

As I know Ocata works with Gnocchi4.1.1 and I’m lazy, I’m just going to use Ocata code.

Installing Gnocchi

To install, I’m running through the Ocata install guide with the only difference being, rather than installing the Ocata packages, I’m installing Gnocchi and the client via pip:

pip install gnocchi[mysql,redis,keystone]
pip install “gnocchiclient<4.0.0”  # my fix doesn't support >=4.0

for reference, below is my gnocchi.conf

[api]
auth_mode = keystone

[indexer]
url = mysql+pymysql://gnocchi:password@controller/gnocchi

[keystone_authtoken]
# i copied this section from my ocata ceilometer.conf that packstack
# configured
auth_uri=http://controller:5000/v2.0
identity_uri=http://controller:35357
admin_user=gnocchi
admin_tenant_name=services
admin_password=password

[metricd]
workers = 18

[storage]
redis_url = redis://controller:6379
driver = redis
coordination_url = redis://controller:6379
aggregation_workers_number = 8
metric_processing_delay = 60
metric_reporting_delay = 10

[incoming]
driver = redis
redis_url = redis://controller:6379

with all that set, I initialise the Gnocchi service using:

gnocchi-upgrade

Creating Ceilometer resources

Now that I have Gnocchi4.1 installed, I’ve decided to backport Ceilometer’s Gnocchi integration from Ocata.

I first began by copying the gnocchi_client code from Ceilometer as it is not in Mitaka. The big difference between Ocata and Mitaka is that in Mitaka, Ceilometer uses a global configuration object and also does not use keystoneauth1. To get around this, I edited the get_gnocchiclient method:

def get_gnocchiclient(conf, endpoint_override=None):
    requests_session = requests.session()
    for scheme in list(requests_session.adapters.keys()):
        requests_session.mount(scheme, ka_session.TCPKeepAliveAdapter(
            pool_block=True))

session = keystone_client.get_session(requests_session=requests_session)
    return client.Client('1', session,
                         interface=conf.service_credentials.interface,
                         region_name=conf.service_credentials.region_name)

Once this is created, we can make Ceilometer create required resource types in Gnocchi. This can be skipped if you already have an existing Gnocchi deployment and just upgraded it to Gnocchi4.1.1 as Gnocchi previously created these resources-types

from oslo_config import cfg

from ceilometer import gnocchi_client
from ceilometer import service

service.prepare_service()
gnocchi_client.upgrade_resource_types(cfg.CONF)

The above script should trigger POST requests against the gnocchi-api, creating OpenStack specific resource-types.

Using Ocata Gnocchi dispatcher

First, I began by creating gnocchi_resources.yaml by copying the Ocata file verbatim.

Next, I overwrote the existing Gnocchi dispatcher with the Ocata version. Doing so provides support for capturing event data to track a resource’s lifespan. Quite a few tweaks were required in this module ranging from changing the dispatcher to handle a global configuration object, to fixing some import statements to handle different dependencies. The end result can be found in my fork.

With that, all the changes to required to leverage Gnocchi4.1 in Mitaka is complete. What remains is to edit ceilometer.conf and add:

[DEFAULT]
meter_dispatchers = gnocchi
event_dispatchers = gnocchi
# meter|event_dispatchers = database can still remain if you wish to
# continue publishing to ceilometer storage as well as Gnocchi.

[cache]
# this is used to minimise requests made to Gnocchi
backend_argument = redis_expiration_time:600
backend_argument = db:0
backend_argument = distrubuted_lock:True
backend_argument = url:redis://controller:6379
backend = dogpile.cache.redis

Once this is set, restarting openstack-ceilometer-collector service should start publishing to Gnocchi4, giving you a scalable metric storage solution in Mitaka.