Telemetrics-client

This package provides the front end component of a complete telemetrics solution for Linux-based operating systems. Specifically, the front end component includes:

telemetrics probes that collect specific types of data from the operating system. For more info on probes go here.
a library, libtelemetry, that telemetrics probes use to create telemetrics records and send them to the daemon for further processing.
telemprobd, a daemon that handles communication with probes and forwards records to a second daemon.
telempostd, this daemon sends telemetry records to a telemetrics server (not included in this source tree), or spools the records on disk in case it's unable to successfully deliver them.

A telemetrics server implementation that works with this component is available from clearlinux/telemetrics-backend.

Build dependencies

libcheck
libcurl
elfutils, which provides libelf and libdwfl libraries..
(optional) libsystemd, for syslog-style logging to the systemd journal, and socket/path activation of telemprobd and telempostd by systemd.

Build and installation

$ ./autogen.sh
$ ./configure
$ make

Set up

There is a config file installed at

/usr/share/defaults/telemetrics/telemetrics.conf

To make modifications, copy that file to

/etc/telemetrics/telemetrics.conf

and modify the /etc version.

Descriptions of config options are listed below in the Usage section.

Starting the client

To use the telemetrics client a one time explicit opt-in is required (this is also true when the contents of the directory /etc/telemetrics/ are removed). To opt-in to telemetrics-client use the command:

 telemctl opt-in

Note this is a change from previous versions, before 2.3.0 installation of telemetrics client was enough to enable the client and if needed the client could be disabled with telemctl opt-out. This command in previous versions created /etc/telemetrics/opt-out file (after telemetrics-client version 2.3.0 this file can be safely removed).

If the client was compiled with systemd support the respective activation units should be already in place (after a make install invocation). In this case the client wil start automatically when data is made available to it. i.e. when executing an /usr/bin/hprobe command. Otherwise use the following command:

 telemctl start

Note: the above invocation technically readies the service for both socket and path activation, so you may not see an "active" status. To check the status of telemetrics-client use:

 telemctl is-active
   telemprobd : active
   telempostd : active

Starting individual service units telempostd.service or telemeprobd.service is discouraged.

Configure the client to autostart at boot

As longs as the first time opt-in was performed, the following methods are valid:

Method 1 (recommended):

Enable the socket-activated service and path unit:

systemctl enable telemprobd.socket telempostd.path

Method 2:

Enable services, which automatically enables the socket and path units as well:

systemctl enable telemprobd.service
systemctl enable telempostd.service

Usage

Once the client is running, the telemetrics probes will be ready to use.

Available probes:

hprobe: A test program that utilizes libtelemetry to ensure that telemetrics-client works. It sends a "hello world" record to the server.
crash probe: A handler for core files that sends the corresponding backtraces to the server.

The client uses the following configuration options from the configuration file:

server: This specifies the web server that the telempostd sends the telemetry records to
socket_path: This specifies the path of the unix domain socket that the telemprobd listens on for connections from the probes
spool_dir: This config option is related to spooling. If the daemon is not able to send the telemetry records to the backend server due to reasons such as the network availability, then it stores the records in a spool directory. This option specifies that path of the spool directory. This directory should be owned by the same user that the daemon is running as.

    mkdir -p /var/spool/telemetry
    chown -R telemetry:telemetry /var/spool/telemetry
    systemctl restart telemprobd.service

record_expiry: This is the time in minutes after which the records in the spool directory are deleted by the daemon.
spool_max_size: This specifies the maximum size of the spool directory. When the size of the spool directory reaches this limit, new telemetry records are dropped by the daemon.
spool_process_time: This specifies the time interval in seconds that the daemon waits for before checking the spool directory for records. The daemon picks up the records in the order of modification date and tries to send the record to the server. It sends a maximum of 10 records at a time. If it was able to send a record successfully, it deletes the record from the spool. If the daemon finds a record older than the "record_expiry" time, then it deletes that record. The daemon looks at a maximum of 20 records in a single spool run loop.
rate_limit_enabled: This determines whether rate-limiting is enabled or disabled. When enabled, there is a threshold on both records sent within a window of time, and record bytes sent within a window a time.
record_burst_limit: This is the maximum amount of records allowed to be passed by the daemon within the record_window_length of time. If set to -1, the rate-limiting for record bursts is disabled.
record_window_length: This is the time, in minutes (0-59), that establishes the window length for the record_burst_limit. EX: if record_burst_window=1000 and record_window_length=15, then no more than 1000 records can be passed within any given fifteen minute window.
byte_burst_limit: This is the maximum amount of bytes that can be passed by the daemon within the byte_window_length of time. If set to -1, the rate-limiting for byte bursts is disabled.
byte_window_length: This is the time, in minutes (0-59), that establishes the window length for the byte_burst_limit.
rate_limit_strategy: This is the strategy chosen once the rate-limiting threshold has been reached. Currently the options are 'drop' or 'spool', with spool being the default. If spool is chosen, records will be spooled and sent at a later time.
record_retention_enabled: When this key is enabled (true) the daemon saves a copy of the payload on disk from all valid records. To avoid the excessive use of disk space only the latest 100 records are kept. The default value for this configuration key is false.
record_server_delivery_enabled: This key controls the delivery of records to server, when enabled (default value) the record will be posted to the address in the configuration file. If this configuration key is disabled (false) records will not be spooled or posted to backend. This configuration key can be used in combination with record_retention_enabled to keep copies of telemetry records locally only.

Data reported

The data reported by the telemetry client could be understood as two main sets: metadata and a payload.

The metadata is used to report details of a machine's architecture. The following are the metadata values currently collected (Record Format Version 4):

record_format_version: version of the record, currently is 'Version 4'. This value changes when new metadata is added.
classification: this field is used to identify the type of record sent by a specific client probe; classifications use the format DOMAIN/PROBE/REST, where DOMAIN is the vendor of the probe, PROBE is the probe name, and REST is a probe-defined field to classify what is contained in the payload.
severity: this is an integer value between 1 and 4 where 1 is "low" and 4 is "critical"
machine_id: a machine identifier that is rotate every 3 days for privacy reasons.
creation_timestamp: timestamp when the record was collected.
arch: a string describing machine architecture i.e. 'x86_64'.
host_type: a string with the combination of 'System Vendor', 'Product Name', and 'Product Version' read from dmi file system.
build: OS build number.
kernel_version: Kernel version.
payload_format_version: version of the payload, currently is 'Version 1'.
system_name: the value after 'ID=' from '/etc/os-release' (or distribution provided folder)
board_name: a string read from dmi file system that combines 'Board Name' and 'Board Vendor'.
cpu_model: cpu model name extracted from '/proc/cpuinfo'.
bios_version: BIOS version.
event_id: an id to group multiple records if these were generated by a single event occurrence.

The payload as mentioned above is reported by probes. The telemetry library adds the metadata to the payload (done programatically when using library API) for more information about probes go here.

Machine id

The machine id reported by the telemetry client is rotated every 3 days for privacy reasons. If you wish to have a static machine id for testing purposes, you can opt in by creating a static machine id file named "opt-in-static-machine-id" under the directory "/etc/telemetrics/". Where "unique machine ID" is your desired static machine ID:

# mkdir -p /etc/telemetrics
# echo "unique machine ID" > /etc/telemetrics/opt-in-static-machine-id

The telemetry client reads, at most, the first 32 characters from this file uses it for the machine id. You can put a string like 'my-machine-name' in this file to easily identify your machine. Restart telemprobd for the machine id changes to take effect by running:

# systemctl restart telemprobd.service

You can switch back to the rotating machine id by deleting the override file and restarting the client. You can do a quick test to check that your machine-id has changed by running "hprobe" and verifying that a record has landed on your backend telemetrics server, with the specified machine id.

Event Id

This is a 32 character lowercase hexadecimal string i.e. '5de9de8d5f3c6a7d445d75ba01cc3322'. This header is used to group multiple records by an event id. Before this header every single record could have been thought of an event, however this is not always the case. There are "events" that trigger the creation of multiple records (i.e. updates). The event_id header was added for probes with the capability to detect events and group records based on such events. This header was added to telem-record-gen and can be specified using the -e (--event-id long form) switch.

  -e, --event-id        Event id to use in the record

Debugging locally with telemetrics-client

The function of the telemetrics-client is to handle the transport of information reported by a probe to a backend (see server key in configuration). This information is helpful for developers to debug and fix reported crashes, however developers not always have access to the backend in these case users can leverage features added for local debugging. The following is a list of steps to enable local debug:

Enabling record retention: this step configures telempostd to keep copies of telemetry records locally. To enable record retention set the value of record_retention_enabled from false to true. Optionally set record_server_delivery_enabled`` to false to keep records local only. Remember to restart the client after configuration values are updated (telemctl restart```).
Creating a record: run hprobe command to create a record for the purposes of this step by step guide. Once we have the record or records that you need to capture locally you can display the data.
Displaying record metadata: telempostd keeps metadata of any valid record, to display this data a new option to telemctl was added telemctl journal. Assuming that the last record created was the record from previous step hprobe we can use tail -n 1 to print the last created record only, i.e.

$ sudo telemctl journal | tail -n 1
$ org.clearlinux/hello/world     Mon 2018-04-02 17:48:01 UTC a19a0d41ba16788881e274b19b8a1be4 5de9de8d5f3c6a7d445d75ba01cc3322 60c014cd-4693-40f1-b334-548cd932949b

The headers for the metadata (along with other information) can be printed using the -V switch with telemctl journal command, i.e.

$ sudo telemctl journal -V | head -n 1
$ Classification             Time stamp              Record ID                    Event ID                     Boot ID

Displaying record payload: to print the content of a record payload you can use the -i (--include_record long format) option to telemctl journal command. To print the specific record you created you can use the option -r (--record_id long format) with the Record Id of the generated record, i.e.

$ sudo telemctl journal --record_id a19a0d41ba16788881e274b19b8a1be4 --include_record
$ org.clearlinux/hello/world     Mon 2018-04-02 17:48:01 UTC a19a0d41ba16788881e274b19b8a1be4 5de9de8d5f3c6a7d445d75ba01cc3322 60c014cd-4693-40f1-b334-548cd932949b
$ hello

Security Disclosures

To report a security issue or receive security advisories please follow procedures in this link.

sjas/telemetrics-client