Home

Awesome

buildevents

OSS Lifecycle CircleCI

buildevents is a small binary used to help instrument builds in a build system such as Travis-CI, CircleCI, Jenkins, and so on. It is installed during the setup phase and then invoked as part of each step in order to visualize the build as a trace in Honeycomb

The trace that you get at the end represents the entire build. It has spans for each section and subsection of the build, representing groups of actual commands that are run. The duration of each span is how long that stage or specific command took to run, and includes whether or not the command succeeded.

Here's an example showing a build that ran on CircleCI. It goes through running go tests, setting up javascript dependencies, triggers js_build and poodle_test in parallel after dependencies are configured, and then continues off below the captured portion of the waterfall.

CircleCI_Build_Trace

Setup

Getting your build ready to use buildevents involves:

Installation

If you have a working go environment in your build, the easiest way to install buildevents is via go get.

go get github.com/honeycombio/buildevents/

There are also built binaries for linux and macOS hosted on Github and available under the releases tab. The following commands will download and make executable the github-hosted binary.

linux, 32-bit x86:

curl -L -o buildevents https://github.com/honeycombio/buildevents/releases/latest/download/buildevents-linux-386
chmod 755 buildevents

linux, 64-bit x86:

curl -L -o buildevents https://github.com/honeycombio/buildevents/releases/latest/download/buildevents-linux-amd64
chmod 755 buildevents

linux, arm64:

curl -L -o buildevents https://github.com/honeycombio/buildevents/releases/latest/download/buildevents-linux-arm64
chmod 755 buildevents

macOS, amd64:

curl -L -o buildevents https://github.com/honeycombio/buildevents/releases/latest/download/buildevents-darwin-amd64
chmod 755 buildevents

macOS, arm64:

curl -L -o buildevents https://github.com/honeycombio/buildevents/releases/latest/download/buildevents-darwin-arm64
chmod 755 buildevents

windows, 32-bit x86:

curl.exe -L -o buildevents.exe https://github.com/honeycombio/buildevents/releases/latest/download/buildevents-windows-386.exe

windows, 64-bit x86:

curl.exe -L -o buildevents.exe https://github.com/honeycombio/buildevents/releases/latest/download/buildevents-windows-amd64.exe

windows, arm64:

curl.exe -L -o buildevents.exe https://github.com/honeycombio/buildevents/releases/latest/download/buildevents-windows-arm64.exe

If this doesn't work for you, please let us know - we'd love to hear what would work.

<!-- TODO provide a compiled binary at some evergreen 'latest' URL -->

Environment Variables

There is one required environment variable; it will hold your Honeycomb API key (available at https://ui.honeycomb.io/account). If it is absent, events will not be sent to Honeycomb. Set BUILDEVENT_APIKEY to hold your API key.

There are several other optional enviornment variables that will adjust the behavior of buildevents:

Trace Identifier

The buildevents script needs a unique ID to join together all of the steps and commands with the build. This is the Trace ID. It must be unique within the Honeycomb dataset holding traces. An excellent choice is the Build ID, since it is both unique (even when re-running builds, you will often get a new Build ID) and is also a primary value that the build system uses to identify the build.

The Build ID may already be available in the environment for your build:

Use

Now that buildevents is installed and configured, actually generating spans to send to Honeycomb involves invoking buildevents in various places throughout your build config.

buildevents is invoked with one of three modes, build, step, and cmd.

build

Though listed first, running buildevents in build mode should actually be the last command that your build runs so that it can record the total running duration for the build. It does this by having the time the build started as one of the arguments passed in.

The output of buildevents in build will be a link to the trace within Honeycomb. Take this URL and use it in the notifications your CI system emits to easily jump to the Honeycomb trace for a build. If the API Key used in this run is not valid, no output will be emitted.

Note that CircleCI uses an alternate method of creating the root span, so the build command should not be used. Use the watch command instead.

For the build step, you must first record the time the build started.

The actual invocation of buildevents build should be as close to the last thing that the build does as possible.

Travis-CI example:

env:
  global:
    - BUILD_START=$(date +%s)

...

after_failure:
  - traceURL=$(buildevents build $TRAVIS_BUILD_ID $BUILD_START failure)
  - echo "Honeycomb Trace: $traceURL"
after_success:
  - traceURL=$(buildevents build $TRAVIS_BUILD_ID $BUILD_START success)
  - echo "Honeycomb Trace: $traceURL"

what it generates

Given this command:

buildevents $HOST -k $API_KEY build htjebmye $BUILD_STARTTIME success

The event that arrives at Honeycomb (which has no trace.parent_id since it is the root of a trace) might look like:

{
    "Timestamp": "2022-05-24T01:49:13Z",
    "command_name": "build",
    "duration_ms": 4981,
    "meta.version": "dev",
    "name": "build htjebmye",
    "service.name": "build",
    "service_name": "build",
    "source": "buildevents",
    "status": "success",
    "trace.span_id": "htjebmye",
    "trace.trace_id": "htjebmye"
}

watch

CirclecI requires use of the CircleCI API to detect when workflows start and stop. There is no facility to always run a job after all others, so what works using the Travis-CI after_failure will not work on CircleCI. However, the CircleCI API exposes when the current workflow has started, and can be used intsead.

The watch command polls the CircleCI API and waits until all jobs have finished (either succeeded, failed, or are blocked). It then reports the final status of the build with the appropriate timers. watch should be invoked in a job all on its own, dependent on only the setup job, with only the Trace ID to use. After some time, watch will timeout waiting for the build to finish and fail. The timeout default is 10 minutes and can be overridden by setting BUILDEVENT_TIMEOUT

Using the watch command requires a personal (not project) CircleCI API token. You can provide this token to buildevents via the BUILDEVENT_CIRCLE_API_TOKEN environment variable. You can get a personal API token from https://circleci.com/account/api. For more detail on tokens, please see the CircleCI API Tokens documentation

The watch command will emit a link to the finished trace to the job output in Honeycomb when the build is complete.

jobs:
  send_trace:
    steps:
      - run: buildevents watch $CIRCLE_WORKFLOW_ID

step

The step mode is the outer wrapper that joins a collection of individual cmds together in to a block. Like the build command, it should be run at the end of the collection of cmds and needs a start time collected at the beginning. In addition to the trace identifier, it needs a step identifier that will also be passed to all the cmds that are part of this step in order to tie them together in to a block. Because the step identifier must be available to all commands, both it and the start time should be generated at the beginning of the step and recorded. The step identifier must be unique within the trace (but does not need to be globally unique). To avoid being distracting, we use a hash of the step name as the identifier.

Travis-CI exmaple:

before_script:
  - STEP_START=$(date +%s)
  - STEP_SPAN_ID=$(echo before_script | sum | cut -f 1 -d \ )
  - ... do stuff
  - buildevents step $TRAVIS_BUILD_ID $STEP_SPAN_ID $STEP_START before_script

CircleCI example:

jobs:
  go_test:
    steps:
      - run: echo "STEP_START=$(date +%s)" >> $BASH_ENV
      - run: echo "STEP_SPAN_ID=$(echo go_test | sum | cut -f 1 -d \ )" >> $BASH_ENV
      - run: ... do stuff
      - run:
          name: finishing span for the job
          command: $GOPATH/bin/buildevents step $CIRCLE_WORKFLOW_ID $STEP_SPAN_ID $STEP_START go_test
          when: always   # ensures the span is always sent, even when something in the job fails

what it generates

Given this command:

buildevents $HOST -k $API_KEY step htjebmye building_htjebmye $STEP_STARTTIME building

The event that arrives at Honeycomb might look like:

{
    "Timestamp": "2022-05-24T01:49:14Z",
    "command_name": "step",
    "duration_ms": 3064,
    "meta.version": "dev",
    "name": "building",
    "service.name": "step",
    "service_name": "step",
    "source": "buildevents",
    "trace.parent_id": "htjebmye",
    "trace.span_id": "building_htjebmye",
    "trace.trace_id": "htjebmye"
}

cmd

Running buildevents cmd will run the given command, time it, and include the status of the command (success or failure). buildevents passes through both STDOUT and STDERR from the process it wraps, and exits with the same exit code as the wrapped process. The actual command to run is separated from the buildevents arguments by a double hyphen --.

This is the most frequent line you'll see in your config file; anything of consequence should generate a span.

Travis-CI example:

script:
  - buildevents cmd $TRAVIS_BUILD_ID $STEP_SPAN_ID go-test -- go test -timeout 2m -mod vendor ./...

CircleCI example:

jobs:
  go_test:
    steps:
      - run: $GOPATH/bin/buildevents cmd $TRAVIS_BUILD_ID $STEP_SPAN_ID go-test -- go test -timeout 2m -mod vendor ./...

what it generates

Given this command:

buildevents $HOST -k $API_KEY cmd htjebmye building_htjebmye compile -- sleep 1

The event that arrives at Honeycomb might look like:

{
    "Timestamp": "2022-05-24T01:49:14.653182Z",
    "cmd": "\"sleep\" \"1\"",
    "command_name": "cmd",
    "duration_ms": 1008,
    "meta.version": "dev",
    "name": "compile",
    "service.name": "cmd",
    "service_name": "cmd",
    "source": "buildevents",
    "status": "success",
    "trace.parent_id": "building_htjebmye",
    "trace.span_id": "6facde6ac6a95e704b9ec1c837270578",
    "trace.trace_id": "htjebmye"
}

Attaching more traces from your build and test process

Every command running through buildevents cmd will receive a HONEYCOMB_TRACE environment variable that contains a marshalled trace propagation context. This can be used to connect more spans to this trace.

Ruby Beeline example:

# at the very start of the command
# establish a command-level span, linking to the buildevent
process_span = Honeycomb.start_span(name: File.basename($PROGRAM_NAME), serialized_trace: ENV['HONEYCOMB_TRACE'])
Honeycomb.add_field_to_trace('process.full_name', $PROGRAM_NAME)

# if you're not passing sensitive information through CLI args, enable this for more insights.
#Honeycomb.add_field_to_trace('process.args', ARGV)

# override the HONEYCOMB_TRACE for sub-processes
ENV['HONEYCOMB_TRACE'] = process_span.to_trace_header

# ensure that the process_span is sent before the process terminates
at_exit do
  if $ERROR_INFO&.is_a?(SystemExit)
    process_span.add_field('process.exit_code', $ERROR_INFO.status)
  elsif $ERROR_INFO
    process_span.add_field('process.exit_code', $ERROR_INFO.class.name)
  else
    process_span.add_field('process.exit_code', 'unknown')
  end
  process_span.send
end

Putting it all together

We've covered each of the three modes in which buildevents is invoked and shown abbreviated examples for each one. Now it's time to look at an entire config to see how they interact: installation, running a build, and finally reporting the whole thing.

In both of these examples, the BUILDEVENT_APIKEY should be set in the protected environment variable section of the CI config so that your API key is not checked in to your source.

Travis-CI example:

env:
  global:
    - BUILD_START=$(date +%s)

install:
  - STEP_START=$(date +%s)
  - STEP_SPAN_ID=$(echo install | sum | cut -f 1 -d \ )
  - curl -L -o buildevents https://github.com/honeycombio/buildevents/releases/latest/download/buildevents-linux-amd64
  - chmod 755 buildevents
  - # ... any other setup necessary for your build
  - ./buildevents step $TRAVIS_BUILD_ID $STEP_SPAN_ID $STEP_START install

script:
  - STEP_START=$(date +%s)
  - STEP_SPAN_ID=$(echo script | sum | cut -f 1 -d \ )
  - ./buildevents cmd $TRAVIS_BUILD_ID $STEP_SPAN_ID go-tests -- go test ./...
  - ./buildevents cmd $TRAVIS_BUILD_ID $STEP_SPAN_ID js-tests -- yarn test
  - ./buildevents step $TRAVIS_BUILD_ID $STEP_SPAN_ID $STEP_START script

after_failure:
  - ./buildevents build $TRAVIS_BUILD_ID $BUILD_START failure

after_success:
  - STEP_START=$(date +%s)
  - STEP_SPAN_ID=$(echo after_success | sum | cut -f 1 -d \ )
  - ./buildevents cmd $TRAVIS_BUILD_ID $STEP_SPAN_ID build -- go install ./...
  - # ... tar up artifacts, upload them, etc.
  - ./buildevents step $TRAVIS_BUILD_ID $STEP_SPAN_ID $STEP_START after_success
  - ./buildevents build $TRAVIS_BUILD_ID $BUILD_START success

CircleCI example:

version: 2.1

# factored out start/finish_job_span commands here so we don't have every one of our build jobs duplicating them
commands:
  with_job_span:
    parameters:
      steps:
        type: steps
    steps:
      - attach_workspace:
          at: buildevents
      - run:
          name: starting span for job
          command: |
            echo "STEP_START=$(date +%s)" >> $BASH_ENV
            echo "STEP_SPAN_ID=$(echo $CIRCLE_JOB | sum | cut -f 1 -d \ )" >> $BASH_ENV
      - run: echo "PATH=$PATH:buildevents/bin/" >> $BASH_ENV
      - steps: << parameters.steps >>
      - run:
          name: finishing span for job
          command: buildevents step $CIRCLE_WORKFLOW_ID $STEP_SPAN_ID $STEP_START $CIRCLE_JOB
          when: always

jobs:
  setup:
    steps:
      - run: |
          mkdir -p buildevents/bin
          date +%s > buildevents/build_start
      - run: go get github.com/honeycombio/buildevents
      - run: cp $GOPATH/bin/buildevents buildevents/bin/
      - persist_to_workspace:
          root: buildevents
          paths:
            - build_start
            - bin/buildevents
  send_trace:
    steps:
      - attach_workspace:
          at: buildevents
      - run: buildevents watch $CIRCLE_WORKFLOW_ID
  test:
    steps:
      - with_job_span:
          steps:
            - run: buildevents cmd $CIRCLE_WORKFLOW_ID $STEP_SPAN_ID go-tests -- go test ./...
            - run: buildevents cmd $CIRCLE_WORKFLOW_ID $STEP_SPAN_ID js-tests -- yarn test
  build:
    steps:
      - with_job_span:
          steps:
            - run: mkdir artifacts
            - run: buildevents cmd $CIRCLE_WORKFLOW_ID $STEP_SPAN_ID build -- go install ./...
            - run: # publish your build artifacts

workflows:
  test-and-build:
    jobs:
      - setup
      - send_trace:
          requires:
            - setup
      - test:
          requires:
            - setup
      - build:
          requires:
            - test

GitLab CI example:

# Not a huge fan of YAML anchors, but it's the easiest way to
# extend the scripts in jobs where you need other before_script and after_script
.default_before_script: &default_before_script
  - STEP_START=$(date +%s)
  - STEP_SPAN_ID=$(echo $CI_JOB_NAME | sum | cut -f 1 -d \ )
  - echo "export STEP_START=$STEP_START" >> buildevents/env
  - echo "export STEP_SPAN_ID=$STEP_SPAN_ID" >> buildevents/env
  - echo "export PATH=\"$PATH:buildevents/bin/\"" >> buildevents/env
  - source buildevents/env
  - cat buildevents/env

.default_after_script: &default_after_script
  - source buildevents/env
  - cat buildevents/env
  - buildevents step $CI_PIPELINE_ID $STEP_SPAN_ID $STEP_START $CI_JOB_NAME

default:
  image: golang:latest
  before_script:
    - *default_before_script
  after_script:
    - *default_after_script

stages:
  # .pre and .post are guaranteed to be first and last run jobs
  # https://docs.gitlab.com/ee/ci/yaml/README.html#pre-and-post
  - .pre
  - build
  - test
  - .post

setup:
  before_script:
    - mkdir -p buildevents/bin/
    - *default_before_script
  script:
    - curl -L -o main https://github.com/honeycombio/buildevents/releases/latest/download/buildevents-linux-amd64
    - chmod 755 main
    - mv main buildevents/bin/buildevents
    - export BUILD_START=$(date +%s)
    - echo "export BUILD_START=$(date +%s)" >> buildevents/env
  artifacts:
    paths:
      - buildevents
  stage: .pre

go_build:
  script:
    - buildevents cmd $CI_PIPELINE_ID $STEP_SPAN_ID build -- go install ./...
  stage: build

go_test:
  script:
    - buildevents cmd $CI_PIPELINE_ID $STEP_SPAN_ID build -- go test ./...
  stage: test

send_success_trace:
  script:
    - "traceURL=$(buildevents build $CI_PIPELINE_ID $BUILD_START success)"
    - "echo \"Honeycomb Trace: $traceURL\""
  stage: .post
  rules:
    - when: on_success

send_failure_trace:
  script:
    - "traceURL=$(buildevents build $CI_PIPELINE_ID $BUILD_START failure)"
    - "echo \"Honeycomb Trace: $traceURL\""
  stage: .post
  rules:
    - when: on_failure

Positional argument reference

All the arguments to the various buildevents modes are listed above, but for convenience, here is a summary of the modes and the arguments that each requires.

The first argument is the running mode for this invocation of buildevents: build, watch, step, or cmd The remaining arguments differ depending on the mode.

arguments for the build mode:

  1. build_id this is used as both the trace ID and to generate a URL to link back to the build
  2. start_time used to calculate the total duration of the build
  3. status should be success or failure and indicates whether the overall build succeeeded or failed

arguments for the watch mode:

  1. build_id this is used as the trace ID

arguments for the step mode:

  1. build_id this is used as both the trace ID and to generate a URL to link back to the build
  2. step_id buildevents expects a build to contain steps, and each step to have commands. The step ID is used to help construct this tree
  3. start_time used to calculate the total duration of running this step in the build
  4. name the last argument is the name for this step or command, used in the Honeycomb UI

arguments for the cmd mode:

  1. build_id this is used as both the trace ID and to generate a URL to link back to the build
  2. step_id buildevents expects a build to contain steps, and each step to have commands. The step ID is used to help construct this tree
  3. name the name for this command, used in the Honeycomb UI
  4. -- double hyphen indicates the rest of the line will be the command to run

Note

name is most useful if it is a low-cardinality value, usually something like the name of a step in your process. Using a low-cardinality value makes it valuable to do things like GROUP BY name in your queries.

Differences between Classic and non-Classic environments

For "Honeycomb Classic", buildevents works almost the same as it always has. It has added service.name in addition to service_name; both fields have the same value.

In a non-Classic environment, there are several differences: