awx/docs/logging_integration.md

Integration with Third-Party Log Aggregators

This feature builds in the capability to send detailed logs to several kinds of 3rd party external log aggregation services. Services connected to this data feed should be useful in order to gain insights into Tower usage or technical trends. The data is intended to be sent in JSON format via three ways: over a HTTP connection, a direct TCP connection or a direct UDP connection. It uses minimal service-specific tweaks engineered in a custom handler or via an imported library.

Loggers

This features introduces several new loggers which are intended to deliver a large amount of information in a predictable structured format, following the same structure as one would expect if obtaining the data from the API. These data loggers are the following.

• awx.analytics.job_status
  • Summaries of status changes for jobs, project updates, inventory updates, and others
• awx.analytics.job_events
  • Data returned from the Ansible callback module
• awx.analytics.activity_stream
  • Record of changes to the objects within the Ansible Tower app
• awx.analytics.system_tracking
  • Data gathered by Ansible scan modules ran by scan job templates

These loggers only use log-level of INFO.

Additionally, the standard Tower logs are be deliverable through this same mechanism. It should be obvious to the user how to enable to disable each of these 5 sources of data without manipulating a complex dictionary in their local settings file, as well as adjust the log-level consumed from the standard Tower logs.

Supported Services

Committed to support:

• Splunk
• Elastic Stack / ELK Stack / Elastic Cloud

Have tested:

• Sumologic
• Loggly

Considered, but have not tested:

• Datadog
• Red Hat Common Logging via logstash connector

Elastic Search Instructions

In the development environment, the server can be started up with the log aggregation services attached via the Makefile targets. This starts up the 3 associated services of Logstash, Elastic Search, and Kibana as their own separate containers individually.

In addition to running these services, it establishes connections to the tower_tools containers as needed. This is derived from the docker-elk project. (https://github.com/deviantony/docker-elk)

# Start a single server with links
make docker-compose-elk
# Start the HA cluster with links
make docker-compose-cluster-elk

For more instructions on getting started with the environment this stands up, also refer to instructions in /tools/elastic/README.md.

If you were to start from scratch, standing up your own version the elastic stack, then the only change you should need is to add the following lines to the logstash logstash.conf file.

filter {
	json {
		source => "message"
	}
}

Debugging and Pitfalls

Backward-incompatible changes were introduced with Elastic 5.0.0, and customers may need different configurations depending on what versions they are using.

Log Message Schema

Common schema for all loggers:

Field	Information
cluster_host_id	(string) unique identifier of the host within the Tower cluster
level	(choice of DEBUG, INFO, WARNING, ERROR, etc.) Standard python log level, roughly reflecting the significance of the event All of the data loggers as a part of this feature use INFO level, but the other Tower logs will use different levels as appropriate
logger_name	(string) Name of the logger we use in the settings, for example, "awx.analytics.activity_stream"
@timestamp	(datetime) Time of log
path	(string) File path in code where the log was generated

Activity Stream Schema

Field	Information
(common)	this uses all the fields common to all loggers listed above
actor	(string) username of the user who took the action documented in the log
changes	(string) unique identifier of the host within the Tower cluster
operation	(choice of several options) the basic category of the changed logged in the activity stream, for instance, "associate".
object1	(string) Information about the primary object being operated on, consistent with what we show in the activity stream
object2	(string) if applicable, the second object involved in the action

Job Event Schema

This logger echoes the data being saved into job events, except when they would otherwise conflict with expected standard fields from the logger, in which case the fields are named differently. Notably, the field host on the job_event model is given as event_host. There is also a sub-dictionary field event_data within the payload, which will contain different fields depending on the specifics of the Ansible event.

This logger also includes the common fields.

Scan / Fact / System Tracking Data Schema

These contain a detailed dictionary-type field either services, packages, or files.

Field	Information
(common)	this uses all the fields common to all loggers listed above
services	(dict, optional) For services scans, this field is included and has keys based on the name of the service NOTE: Periods are disallowed by elastic search in names, and are replaced with "_" by our log formatter
packages	(dict, optional) Included for log messages from package scans
files	(dict, optional) Included for log messages from file scans
host	(str) name of host scan applies to
inventory_id	(int) inventory id host is inside of

Job Status Changes

This is a intended to be a lower-volume source of information about changes in job states compared to job events, and also intended to capture changes to types of unified jobs other than job template based jobs.

In addition to common fields, these logs include fields present on the job model.

Tower Logs

In addition to the common fields, this will contain a msg field with the log message. Errors contain a separate traceback field. These logs can be enabled or disabled in CTiT by adding or removing it to the setting LOG_AGGREGATOR_LOGGERS.

Configuring Inside of Tower

Parameters needed in order to configure the connection to the log aggregation service will include most of the following for all supported services:

• Host
• Port
• The type of service, allowing service-specific customizations
• Optional username for the connection, used by certain services
• Some kind of token or password
• A flag to indicate how system tracking records will be sent
• Selecting which loggers to send
• Enabling sending logs
• Connection type, which can be HTTPS, TCP and UDP.
• Timeout value if connection type is based on TCP protocol (HTTPS and TCP).

Some settings for the log handler will not be exposed to the user via this mechanism. For example, threading (enabled).

Parameters for the items listed above should be configurable through the Configure-Tower-in-Tower interface.

One note on configuring Host and Port: When entering URL it is customary to include port number, like https://localhost:4399/foo/bar. So for the convenience of users, when connection type is HTTPS, we allow entering hostname as a URL with port number and thus ignore Port field. In other words, Port field is optional in this case. On the other hand, TCP and UDP connections are determined by <hostname, port number> tuple rather than URL. So in the case of TCP/UDP connection, Port field is supposed to be provided and Host field is supposed to contain hostname only. If instead a URL is entered in Host field, its hostname portion will be extracted as the actual hostname.

Acceptance Criteria Notes

Connection: Testers need to replicate the documented steps for setting up and connecting with a destination log aggregation service, if that is an officially supported service. That will involve 1) configuring the settings, as documented, 2) taking some action in Tower that causes a log message from each type of data logger to be sent and 3) verifying that the content is present in the log aggregation service.

Schema: After the connection steps are completed, a tester will need to create an index. We need to confirm that no errors are thrown in this process. It also needs to be confirmed that the schema is consistent with the documentation. In the case of Splunk, we need basic confirmation that the data is compatible with the existing app schema.

Tower logs: Formatting of Traceback message is a known issue in several open-source log handlers, so we should confirm that server errors result in the log aggregator receiving a well-formatted multi-line string with the traceback message.

Log messages should be sent outside of the request-response cycle. For example, loggly examples use requests_futures.sessions.FuturesSession, which does some threading work to fire the message without interfering with other operations. A timeout on the part of the log aggregation service should not cause Tower operations to hang.