SONiC-telemetry

Description

This repository contains implementation for the sonic system telemetry services:

• dial-in mode system telemetry server: telemetry
• dial-out mode system telemetry client dialout_client_cli

Getting Started

Prerequisites

Install go in your system https://golang.org/doc/install. Requires golang1.8+.

Standalone database-config tests

With the Go version declared in go.mod, run:

CGO_ENABLED=0 go test -tags=pure ./internal/adaptors/dbconfig

These tests read temporary JSON fixtures without SWSS, Redis, or system configuration changes. The pure provider supports the default namespace. It rejects global and INCLUDES configuration.

For formatting, CGO-disabled compilation, and race tests:

make -f pure.mk PACKAGES=internal/adaptors/dbconfig fmt-check test

Race tests require CGO and a C compiler, but not SONiC libraries. Normal builds still use SWSS and preserve DB_CONFIG_PATH and the legacy sonic_db_config APIs. In the SONiC build environment, run both provider and legacy tests:

go test -gcflags=all=-l ./internal/adaptors/dbconfig ./sonic_db_config

The existing pure and normal CI jobs include these tests.

Installing

To install dial-in mode system telemetry server, run

go get -u github.com/sonic-net/sonic-gnmi/telemetry

To install dial-out mode system telemetry client, run

go get -u github.com/sonic-net/sonic-gnmi/dialout/dialout_client_cli

There is also a test program dialout_server_cli for collecting data from dial-out mode system telemetry client. Note: it is for testing purpose only.

go get -u github.com/sonic-net/sonic-gnmi/dialout/dialout_server_cli

The binaries will be installed under $GOPATH/bin/, they may be copied to any SONiC switch and run there.

You can also build a debian package and install it:

git clone https://github.com/sonic-net/sonic-gnmi.git
pushd sonic-gnmi
dpkg-buildpackage -rfakeroot -b -us -uc
popd

Testing

Run make -f pure.mk ci-report for pure package boundary checks, compilation, race-enabled tests and CI report generation. Keep sonic-mgmt-common checked out beside this repository for the local module replacement.

To run tests or regenerate coverage separately:

make -f pure.mk test-pure        # JUnit results and raw Go coverage
make -f pure.mk report-coverage  # Existing profile to Cobertura; no test run

ci-report orders boundary checks, compilation, tests and conversion, including under parallel Make. A failed operation stops the aggregate. A new test run removes its previous reports; failed tests retain JUnit but discard partial coverage. Conversion fails on missing/invalid input or a converter error and cannot leave stale XML. Successful report paths remain test-results/junit-pure.xml, test-results/coverage-pure.txt and test-results/coverage-pure.xml.

The existing per-package test, test-coverage and coverage-xml commands still run tests. Neither ci nor ci-report clears shared Go caches. Explicit make -f pure.mk clean still clears shared Go build/test caches and per-package coverage files.

In the provisioned SONiC build environment, run make all first, then ENABLE_TRANSLIB_WRITE=y make test-integration or ENABLE_TRANSLIB_WRITE=y make test-memleak. These commands produce local reports; Azure YAML publishes them. The existing native check_gotest_junit and check_memleak_junit entrypoints remain compatible.

Running

• See SONiC gRPC telemetry for how to run dial-in mode system telemetry server
• See SONiC telemetry in dial-out mode for how to run dial-out mode system telemetry client
• See gNMI Usage Examples for gNMI client usage examples.
• See gNOI Debug actions for the Debug RPC policy and supported actions.

gNOI file read permissions

File.Get and File.Stat expose only world-readable regular files. Ancestor directories must be world-searchable; listing a directory also requires world read/search permissions. Stat listings omit private children. Missing and inaccessible paths return the same read denial.

Resolution is bounded to raw paths shorter than 4096 bytes, 40 symlinks and 64 descriptors per ancestor stack, plus a fixed number of anchor/target handles. A listing captures and validates its pinned ancestor stack once, then shares that immutable prefix across child lookups. Each child has independent traversal state and a copy of the remaining symlink budget; following links may retain one additional bounded ancestor stack. Entries exceeding these safety limits are not exposed.

The check applies regardless of caller role, including when the server runs as root. Reads use the validated descriptor; symlink targets must remain inside the host filesystem view. Extended access ACLs are conservatively denied rather than assuming mode bits alone establish public access. Put, TransferToRemote, Remove, authentication and role authorization are unchanged. No managed-image store, helper process or companion HostService change is required.

Streaming events

SONiC publishes events of interest via gNMI. Sample events could be bgp-state change, link-state change. A sample query URL is as below. gnmi_cli -client_types=gnmi -a 127.0.0.1:50051 -t EVENTS -logtostderr -insecure -v 7 -streaming_type ON_CHANGE -q all -qt s

The events URL can take the following optional parameters.
heartbeat:
[heartbeat=<N>]
    The SONiC switch publishes periodic hearbeats when there are no events to publish.
    The frequency of the heartbeat can be controlled by this parameter as beat in every N seconds.

usecache:
[usecache=true/false]
    The SONiC switch does offline cache when gNMI client is down.
    The cached events are delivered upon nest gNMI connection
    Only one client should use the cache service
    The cached events are delivered upon nest gNMI connection
    If you are running test clients, use this param to turn off cache use.

Sample URL:
gnmi_cli -client_types=gnmi -a 127.0.0.1:50051 -t EVENTS -logtostderr -insecure -v 7 -streaming_type ON_CHANGE -q all[heartbeat=5][usecache=false] -qt s

gnmi_cli is updated with following args

To receive events to a file:

Add -output_file=<file>

To filter out for specific event:

Add -expected_event=<event module:event tag>

e.g. -expected_event=sonic-events-bgp:bgp-state

To read N events and exit

Add -expected_count <N>

To run with a max timeout:

Add -streaming_timeout 10

The following cmd will run the tool until it receives 5 test events or timeout upon 10s, whichever is earlier and receives events are written into file "/tmp/ee".

gnmi_cli -client_types=gnmi -a 127.0.0.1:50051 -t EVENTS -logtostderr -insecure -v 7 -qt s -q all[heartbeat=5] -streaming_type ON_CHANGE -output_file=/tmp/ee -expected_event=":test-tag" -expected_count 5 -streaming_timeout 10

Need Help?

For general questions, setup help, or troubleshooting:

• sonicproject on Google Groups

For bug reports or feature requests, please open an Issue.

Contribution guide

See the contributors guide for information about how to contribute.

GitHub Workflow

We're following basic GitHub Flow. If you have no idea what we're talking about, check out GitHub's official guide. Note that merge is only performed by the repository maintainer.

Guide for performing commits:

• Isolate each commit to one component/bugfix/issue/feature
• Use a standard commit message format:

[component/folder touched]: Description intent of your changes

[List of changes]

Signed-off-by: Your Name your@email.com

For example:

swss-common: Stabilize the ConsumerTable

* Fixing autoreconf
* Fixing unit-tests by adding checkers and initialize the DB before start
* Adding the ability to select from multiple channels
* Health-Monitor - The idea of the patch is that if something went wrong with the notification channel,
  we will have the option to know about it (Query the LLEN table length).

  Signed-off-by: user@dev.null

• Each developer should fork this repository and add the team as a Contributor
• Push your changes to your private fork and do "pull-request" to this repository
• Use a pull request to do code review
• Use issues to keep track of what is going on