1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
|
# 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)
```bash
# 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.
|
