openstack-ansible-ops/elk_metrics_6x

Install ELK with beats to gather metrics

tags

openstack, ansible

About this repository

This set of playbooks will deploy an elastic stack cluster (Elasticsearch, Logstash, Kibana) with beats to gather metrics from hosts and store them into the elastic stack.

These playbooks require Ansible 2.5+.

Highlevel overview of the Elastic-Stack infrastructure these playbooks will build and operate against.

OpenStack-Ansible Integration

These playbooks can be used as standalone inventory or as an integrated part of an OpenStack-Ansible deployment. For a simple example of standalone inventory, see [test-inventory.yml](tests/inventory/test-inventory.yml).

Optional | Load balancer configuration

Configure the Elasticsearch endpoints:

While the Elastic stack cluster does not need a load balancer to scale, it is useful when accessing the Elasticsearch cluster using external tooling. Tools like OSProfiler, Grafana, etc will all benefit from being able to interact with Elasticsearch using the load balancer. This provides better fault tolerance especially when compared to connecting to a single node. The following section can be added to the haproxy_extra_services list to create an Elasticsearch backend. The ingress port used to connect to Elasticsearch is 9201. The backend port is 9200. If this backend is setup make sure you set the internal_lb_vip_address on the CLI or within a known variable file which will be sourced at runtime. If using HAProxy, edit the /etc/openstack_deploy/user_variables.yml file and add the following lines.

haproxy_extra_services:
  - service:
      haproxy_service_name: elastic-logstash
      haproxy_ssl: False
      haproxy_backend_nodes: "{{ groups['Kibana'] | default([]) }}"  # Kibana nodes are also Elasticsearch coordination nodes
      haproxy_port: 9201  # This is set using the "elastic_hap_port" variable
      haproxy_check_port: 9200  # This is set using the "elastic_port" variable
      haproxy_backend_port: 9200  # This is set using the "elastic_port" variable
      haproxy_balance_type: tcp

Configure the Kibana endpoints:

It is recommended to use a load balancer with Kibana. Like Elasticsearch, a load balancer is not required however without one users will need to directly connect to a single Kibana node to access the dashboard. If a load balancer is present it can provide a highly available address for users to access a pool of Kibana nodes which will provide a much better user experience. If using HAProxy, edit the /etc/openstack_deploy/user_variables.yml file and add the following lines.

haproxy_extra_services:
  - service:
      haproxy_service_name: Kibana
      haproxy_ssl: False
      haproxy_backend_nodes: "{{ groups['Kibana'] | default([]) }}"
      haproxy_port: 81  # This is set using the "Kibana_nginx_port" variable
      haproxy_balance_type: tcp

Configure the APM endpoints:

It is recommented to use a load balancer for submitting Application Performance Monitoring data to the APM server. A load balancer will provide a highly available address which APM clients can use to connect to a pool of APM nodes. If using HAProxy, edit the /etc/openstack_deploy/user_variables.yml and add the following lines

haproxy_extra_services:
  - service:
      haproxy_service_name: apm-server
      haproxy_ssl: False
      haproxy_backend_nodes: "{{ groups['apm-server'] | default([]) }}"
      haproxy_port: 8200 # this is set using the "apm_port" variable
      haproxy_balance_type: tcp

Optional | add OSProfiler to an OpenStack-Ansible deployment

To initialize the OSProfiler module within openstack the following overrides can be applied to the to a user variables file. The hmac key needs to be defined consistently throughout the environment.

Full example to initialize the OSProfiler modules throughout an OpenStack-Ansible deployment.

profiler_overrides: &os_profiler
  profiler:
    enabled: true
    trace_sqlalchemy: true
    hmac_keys: "UNIQUE_HMACKEY"  # This needs to be set consistently throughout the deployment
    connection_string: "Elasticsearch://{{ internal_lb_vip_address }}:9201"
    es_doc_type: "notification"
    es_scroll_time: "2m"
    es_scroll_size: "10000"
    filter_error_trace: "false"

aodh_aodh_conf_overrides: *os_profiler
barbican_config_overrides: *os_profiler
ceilometer_ceilometer_conf_overrides: *os_profiler
cinder_cinder_conf_overrides: *os_profiler
designate_designate_conf_overrides: *os_profiler
glance_glance_api_conf_overrides: *os_profiler
gnocchi_conf_overrides: *os_profiler
heat_heat_conf_overrides: *os_profiler
horizon_config_overrides: *os_profiler
ironic_ironic_conf_overrides: *os_profiler
keystone_keystone_conf_overrides: *os_profiler
magnum_config_overrides: *os_profiler
neutron_neutron_conf_overrides: *os_profiler
nova_nova_conf_overrides: *os_profiler
octavia_octavia_conf_overrides: *os_profiler
rally_config_overrides: *os_profiler
sahara_conf_overrides: *os_profiler
swift_swift_conf_overrides: *os_profiler
tacker_tacker_conf_overrides: *os_profiler
trove_config_overrides: *os_profiler

If a deployer wishes to use multiple keys they can do so by with comma separated list.

profiler_overrides: &os_profiler
  profiler:
    hmac_keys: "key1,key2"

To add the OSProfiler section to an exist set of overrides, the yaml section can be added or dynamcally appended to a given hash using yaml tags.

profiler_overrides: &os_profiler
  profiler:
    enabled: true
    hmac_keys: "UNIQUE_HMACKEY"  # This needs to be set consistently throughout the deployment
    connection_string: "Elasticsearch://{{ internal_lb_vip_address }}:9201"
    es_doc_type: "notification"
    es_scroll_time: "2m"
    es_scroll_size: "10000"
    filter_error_trace: "false"

# Example to merge the os_profiler tag to into an existing override hash
nova_nova_conf_overrides:
  section1_override:
    key: "value"
  <<: *os_profiler

While the osprofiler and Elasticsearch libraries should be installed within all virtual environments by default, it's possible they're missing within a given deployment. To install these dependencies throughout the cluster without having to invoke a repo-build run the following adhoc Ansible command can by used.

The version of the Elasticsearch python library should match major version of of Elasticsearch being deployed within the environment.

ansible -m shell -a 'find /openstack/venvs/* -maxdepth 0 -type d -exec {}/bin/pip install osprofiler "elasticsearch>=6.0.0,<7.0.0" --isolated \;' all

Once the overrides are in-place the openstack-ansible playbooks will need to be rerun. To simply inject these options into the system a deployer will be able to use the *-config tags that are apart of all os_* roles. The following example will run the config tag on ALL openstack playbooks.

openstack-ansible setup-openstack.yml --tags "$(cat setup-openstack.yml | grep -wo 'os-.*' | awk -F'-' '{print $2 "-config"}' | tr '\n' ',')"

Once the OSProfiler module has been initialized tasks can be profiled on demand by using the --profile or --os-profile switch in the various openstack clients along with one of the given hmac keys defined.

Legacy profile example command.

glance --profile key1 image-list

Modern profile example command, requires python-openstackclient >= 3.4.1 and the osprofiler library.

openstack --os-profile key2 image list

If the client library is not installed in the same path as the python-openstackclient client, run the following command to install the required library.

pip install osprofiler

Optional | run the haproxy-install playbook

cd /opt/openstack-ansible/playbooks/
openstack-ansible haproxy-install.yml --tags=haproxy-service-config

Setup | system configuration

Clone the elk-osa repo

cd /opt
git clone https://github.com/openstack/openstack-ansible-ops

Copy the env.d file into place

cd /opt/openstack-ansible-ops/elk_metrics_6x
cp env.d/elk.yml /etc/openstack_deploy/env.d/

Copy the conf.d file into place

cp conf.d/elk.yml /etc/openstack_deploy/conf.d/

In elk.yml, list your logging hosts under elastic-logstash_hosts to create the Elasticsearch cluster in multiple containers and one logging host under kibana_hosts to create the Kibana container

vi /etc/openstack_deploy/conf.d/elk.yml

Create the containers

cd /opt/openstack-ansible/playbooks
openstack-ansible lxc-containers-create.yml --limit elk_all

Deploying | Installing with embedded Ansible

If this is being executed on a system that already has Ansible installed but is incompatible with these playbooks the script bootstrap-embedded-ansible.sh can be sourced to grab an embedded version of Ansible prior to executing the playbooks.

source bootstrap-embedded-ansible.sh

Deploying | Manually resolving the dependencies

This playbook has external role dependencies. If Ansible is not installed with the bootstrap-ansible.sh script these dependencies can be resolved with the ansible-galaxy command and the ansible-role-requirements.yml file.

• Example galaxy execution

ansible-galaxy install -r ansible-role-requirements.yml

Once the dependencies are set make sure to set the action plugin path to the location of the config_template action directory. This can be done using the environment variable ANSIBLE_ACTION_PLUGINS or through the use of an ansible.cfg file.

Deploying | The environment

Install master/data Elasticsearch nodes on the elastic-logstash containers, deploy logstash, deploy Kibana, and then deploy all of the service beats.

cd /opt/openstack-ansible-ops/elk_metrics_6x
ansible-playbook site.yml $USER_VARS

• The openstack-ansible command can be used if the version of ansible on the system is greater than 2.5. This will automatically pick up the necessary group_vars for hosts in an OSA deployment.
• If required add -e@/opt/openstack-ansible/inventory/group_vars/all/all.yml to import sufficient OSA group variables to define the OpenStack release. Journalbeat will then deploy onto all hosts/containers for releases prior to Rocky, and hosts only for Rocky onwards. If the variable openstack_release is undefined the default behaviour is to deploy Journalbeat to hosts only.
• Alternatively if using the embedded ansible, create a symlink to include all of the OSA group_vars. These are not available by default with the embedded ansible and can be symlinked into the ops repo.

ln -s /opt/openstack-ansible/inventory/group_vars /opt/openstack-ansible-ops/elk_metrics_6x/group_vars

The individual playbooks found within this repository can be independently run at anytime.

Architecture | Data flow

This diagram outlines the data flow from within an Elastic-Stack deployment.

Optional | Enable uwsgi stats

Config overrides can be used to make uwsgi stats available on unix domain sockets. Any /tmp/*uwsgi-stats.sock will be picked up by Metricsbeat.

keystone_uwsgi_ini_overrides:
  uwsgi:
    stats: "/tmp/keystone-uwsgi-stats.sock"

cinder_api_uwsgi_ini_overrides:
  uwsgi:
    stats: "/tmp/cinder-api-uwsgi-stats.sock"

glance_api_uwsgi_ini_overrides:
  uwsgi:
    stats: "/tmp/glance-api-uwsgi-stats.sock"

heat_api_uwsgi_ini_overrides:
  uwsgi:
    stats: "/tmp/heat-api-uwsgi-stats.sock"

heat_api_cfn_init_overrides:
  uwsgi:
    stats: "/tmp/heat-api-cfn-uwsgi-stats.sock"

nova_api_metadata_uwsgi_ini_overrides:
  uwsgi:
    stats: "/tmp/nova-api-metadata-uwsgi-stats.sock"

nova_api_os_compute_uwsgi_ini_overrides:
  uwsgi:
    stats: "/tmp/nova-api-os-compute-uwsgi-stats.sock"

nova_placement_uwsgi_ini_overrides:
  uwsgi:
    stats: "/tmp/nova-placement-uwsgi-stats.sock"

octavia_api_uwsgi_ini_overrides:
  uwsgi:
    stats: "/tmp/octavia-api-uwsgi-stats.sock"

sahara_api_uwsgi_ini_overrides:
  uwsgi:
    stats: "/tmp/sahara-api-uwsgi-stats.sock"

ironic_api_uwsgi_ini_overrides:
  uwsgi:
    stats: "/tmp/ironic-api-uwsgi-stats.sock"

magnum_api_uwsgi_ini_overrides:
  uwsgi:
    stats: "/tmp/magnum-api-uwsgi-stats.sock"

Rerun all of the openstack-ansible playbooks to enable these stats. Use the ${service_name}-config tags on all of the os_* roles. It's possible to auto-generate the tags list with the following command.

openstack-ansible setup-openstack.yml --tags "$(cat setup-openstack.yml | grep -wo 'os-.*' | awk -F'-' '{print $2 "-config"}' | tr '\n' ',')"

Optional | add Kafka Output format

To send data from Logstash to Kafka create the logstash_kafka_options variable. This variable will be used as a generator and create a Kafka output configuration file using the key/value pairs as options.

logstash_kafka_options:
  codec: json
  topic_id: "elk_kafka"
  ssl_key_password: "{{ logstash_kafka_ssl_key_password }}"
  ssl_keystore_password: "{{ logstash_kafka_ssl_keystore_password }}"
  ssl_keystore_location: "/var/lib/logstash/{{ logstash_kafka_ssl_keystore_location | basename }}"
  ssl_truststore_location: "/var/lib/logstash/{{ logstash_kafka_ssl_truststore_location | basename }}"
  ssl_truststore_password: "{{ logstash_kafka_ssl_truststore_password }}"
  bootstrap_servers:
    - server1.local:9092
    - server2.local:9092
    - server3.local:9092
  client_id: "elk_metrics_6x"
  compression_type: "gzip"
  security_protocol: "SSL"
  id: "UniqueOutputID"

For a complete list of all options available within the Logstash Kafka output plugin please review the following documentation.

Optional config:

The following variables are optional and correspond to the example logstash_kafka_options variable.

logstash_kafka_ssl_key_password: "secrete"
logstash_kafka_ssl_keystore_password: "secrete"
logstash_kafka_ssl_truststore_password: "secrete"

# SSL certificates in Java KeyStore format
logstash_kafka_ssl_keystore_location: "/root/kafka/keystore.jks"
logstash_kafka_ssl_truststore_location: "/root/kafka/truststore.jks"

When using the kafka output plugin the options, logstash_kafka_ssl_keystore_location and logstash_kafka_ssl_truststore_location will automatically copy a local SSL key to the logstash nodes. These options are string value and assume the deployment nodes have local access to the files.

Optional | add Grafana visualizations

See the grafana directory for more information on how to deploy grafana. Once When deploying grafana, source the variable file from ELK in order to automatically connect grafana to the Elasticsearch datastore and import dashboards. Including the variable file is as simple as adding -e @../elk_metrics_6x/vars/variables.yml to the grafana playbook run.

Included dashboards.

• https://grafana.com/dashboards/5569
• https://grafana.com/dashboards/5566

Example command using the embedded Ansible from within the grafana directory.

ansible-playbook ${USER_VARS} installGrafana.yml \
                              -e @../elk_metrics_6x/vars/variables.yml \
                              -e 'galera_root_user="root"' \
                              -e 'galera_address={{ internal_lb_vip_address }}'

Optional | add kibana custom dashboard

If you want to use a custom dashboard directly on your kibana, you can run the playbook bellow. The dashboard uses filebeat to collect the logs of your deployment.

ansible-playbook setupKibanaDashboard.yml $USER_VARS

Overview of kibana custom dashboard

Upgrading the cluster

To upgrade the packages throughout the elastic search cluster set the package state variable, elk_package_state, to latest.

cd /opt/openstack-ansible-ops/elk_metrics_6x
ansible-playbook site.yml $USER_VARS -e 'elk_package_state="latest"'

Trouble shooting

If everything goes bad, you can clean up with the following command

openstack-ansible /opt/openstack-ansible-ops/elk_metrics_6x/site.yml -e 'elk_package_state="absent"' --tags package_install
openstack-ansible /opt/openstack-ansible/playbooks/lxc-containers-destroy.yml --limit elk_all

Local testing

To test these playbooks within a local environment you will need a single server with at leasts 8GiB of RAM and 40GiB of storage on root. Running an m1.medium (openstack) flavor size is generally enough to get an environment online.

To run the local functional tests execute the run-tests.sh script out of the tests directory. This will create a 4 node elasaticsearch cluster, 1 kibana node with an elasticsearch coordination process, and 1 APM node. The beats will be deployed to the environment as if this was a production installation.

tests/run-tests.sh

After the test build is completed the cluster will test it's layout and ensure processes are functioning normally. Logs for the cluster can be found at /tmp/elk-metrics-6x-logs.

To rerun the playbooks after a test build, source the tests/manual-test.rc file and follow the onscreen instructions.

To clean-up a test environment and start from a bare server slate the run-cleanup.sh script can be used. This script is distructive and will purge all elk_metrics_6x related services within the local test environment.

tests/run-cleanup.sh