add opentelemery helper scripts (#119)

Author: msarahan

tests/test_rapids-get-telemetry-trace-id.py

@@ -0,0 +1,49 @@
+import os.path
+import subprocess
+
+TOOLS_DIR = os.path.join(os.path.dirname(os.path.realpath(__file__)), "..", "tools")
+
+def test_rapids_compute_trace_id():
+    result = subprocess.run(
+        os.path.join(TOOLS_DIR, "rapids-get-telemetry-trace-id"),
+        env={
+            "GITHUB_REPOSITORY": "rapidsai/gha-tools",
+            "GITHUB_RUN_ID": "1123123",
+            "RUN_ATTEMPT": "1"
+        },
+        text=True,
+        capture_output=True,
+    )
+    assert result.stdout.strip() == "22ab4ec60f37f446b4a95917e86660df"
+    assert result.stderr == ""
+    assert result.returncode == 0
+
+def test_rapids_get_traceparent():
+    result = subprocess.run(
+        [os.path.join(TOOLS_DIR, "rapids-get-telemetry-traceparent"), "my_job"],
+        env={
+            "GITHUB_REPOSITORY": "rapidsai/gha-tools",
+            "GITHUB_RUN_ID": "1123123",
+            "RUN_ATTEMPT": "1"
+        },
+        text=True,
+        capture_output=True,
+    )
+    assert result.stdout.strip() == "00-22ab4ec60f37f446b4a95917e86660df-5f57388b5b07a3e8-01"
+    assert result.stderr == ""
+    assert result.returncode == 0
+
+def test_rapids_get_traceparent_with_step():
+    result = subprocess.run(
+        [os.path.join(TOOLS_DIR, "rapids-get-telemetry-traceparent"), "my_job", "my step"],
+        env={
+            "GITHUB_REPOSITORY": "rapidsai/gha-tools",
+            "GITHUB_RUN_ID": "1123123",
+            "RUN_ATTEMPT": "1"
+        },
+        text=True,
+        capture_output=True,
+    )
+    assert result.stdout.strip() == "00-22ab4ec60f37f446b4a95917e86660df-a6e5bc57fad91889-01"
+    assert result.stderr == ""
+    assert result.returncode == 0

tools/rapids-conda-retry

@@ -67,7 +67,7 @@ condaCmd=${RAPIDS_CONDA_EXE:=conda}
 #    needToRetry: 1 if the command should be retried, 0 if it should not be
 function runConda {
     # shellcheck disable=SC2086
-    ${condaCmd} ${args} 2>&1| tee "${outfile}"
+    rapids-otel-wrap ${condaCmd} ${args} 2>&1| tee "${outfile}"
     exitcode=$?
     needToRetry=0
     needToClean=0

tools/rapids-get-pr-conda-artifact

@@ -33,4 +33,4 @@ if [[ -z "${commit}" ]]; then
     commit=$(git ls-remote https://github.com/rapidsai/"${repo}".git refs/heads/pull-request/"${pr}" | cut -c1-7)
 fi
 
-rapids-get-artifact "ci/${repo}/pull-request/${pr}/${commit}/${artifact_name}"
+rapids-otel-wrap rapids-get-artifact "ci/${repo}/pull-request/${pr}/${commit}/${artifact_name}"

tools/rapids-get-telemetry-trace-id

@@ -0,0 +1,6 @@
+#!/bin/bash
+# This is a global, per-run identifier. It is the same across all jobs and all steps within all jobs.
+# It is constant from the source repo, to shared-workflows, to shared-actions.
+
+sha="$(echo "${GITHUB_REPOSITORY}+${GITHUB_RUN_ID}+${RUN_ATTEMPT}" | sha256sum | cut -f1 -d' ')"
+echo "${sha:0:32}"

tools/rapids-get-telemetry-traceparent

@@ -0,0 +1,30 @@
+#!/bin/bash
+# This emits a TRACEPARENT, which follows the w3c trace context standard.
+# https://www.w3.org/TR/trace-context/
+#
+# This script can operate for two purposes:
+# 1. The top level of a job, whether it is the job at the source repo (e.g. rmm) level, or
+#      the matrix job level
+# 2. The steps level within a job, which uses both the job name and the step name
+#
+# The job name must always be provided as the first argument.
+# A step name MAY be provided as the second argument. If it is specified, the output corresponds to
+#     the step within the context of its job.
+
+JOB_NAME=$1
+STEP_NAME=${2:-}
+
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )
+
+TRACE_ID="$("${SCRIPT_DIR}"/rapids-get-telemetry-trace-id)"
+JOB_SPAN_ID="${TRACE_ID}-${JOB_NAME}"
+STEP_SPAN_ID="${JOB_SPAN_ID}-${STEP_NAME}"
+
+JOB_TRACEPARENT=$(echo -n "${JOB_SPAN_ID}" | sha256sum | cut -f1 -d' ')
+STEP_TRACEPARENT=$(echo -n "${STEP_SPAN_ID}" | sha256sum | cut -f1 -d' ')
+
+if [ "${STEP_NAME}" != "" ]; then
+    echo "00-${TRACE_ID}-${STEP_TRACEPARENT:0:16}-01"
+else
+    echo "00-${TRACE_ID}-${JOB_TRACEPARENT:0:16}-01"
+fi

tools/rapids-mamba-retry

@@ -46,4 +46,4 @@ for arg in "$@"; do
    fi
 done
 
-rapids-conda-retry "$@"
+rapids-otel-wrap rapids-conda-retry "${args[@]}"

tools/rapids-otel-wrap

@@ -0,0 +1,53 @@
+#!/bin/bash
+# Wraps arbitrary commands with arbitrary args. Emits an OpenTelemetry span for tracing the command
+
+SCRIPT_DIR="$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )"
+
+RAPIDS_OTEL_TRACES_EXPORTER="${RAPIDS_OTEL_TRACES_EXPORTER:-${RAPIDS_OTEL_EXPORTER:-"console"}}"
+RAPIDS_OTEL_METRICS_EXPORTER="${RAPIDS_OTEL_METRICS_EXPORTER:-${RAPIDS_OTEL_EXPORTER:-"console"}}"
+RAPIDS_OTEL_LOGS_EXPORTER="${RAPIDS_OTEL_LOGS_EXPORTER:-${RAPIDS_OTEL_EXPORTER:-"console"}}"
+OTEL_EXPORTER_OTLP_TRACES_ENDPOINT="${OTEL_EXPORTER_OTLP_TRACES_ENDPOINT:-${OTEL_EXPORTER_OTLP_ENDPOINT}/v1/traces}"
+OTEL_EXPORTER_OTLP_METRICS_ENDPOINT="${OTEL_EXPORTER_OTLP_METRICS_ENDPOINT:-${OTEL_EXPORTER_OTLP_ENDPOINT}/v1/metrics}"
+OTEL_EXPORTER_OTLP_LOGS_ENDPOINT="${OTEL_EXPORTER_OTLP_LOGS_ENDPOINT:-${OTEL_EXPORTER_OTLP_ENDPOINT}/v1/logs}"
+export TRACEPARENT="${TRACEPARENT}"
+
+if [[ $(type otel-cli >/dev/null 2>&1) -eq 0 ]] && [ "$TRACEPARENT" != "" ]; then
+    echo "Running command with OpenTelemetry instrumentation";
+
+    set -x
+    if [ "$OTEL_SERVICE_NAME" = "" ]; then
+        echo "WARNING: OTEL_SERVICE_NAME variable not provided. Traces from different steps may not be associated correctly."
+    fi
+
+    # Some commands have instrumentation. For example, conda-build has monkey-patched instrumentation
+    # that can be activated with the opentelemetry-instrument command. For these commands,
+    # we replace the command with the wrapped command, quoted as a whole for the purposes
+    # of otel-cli exec, so that flags don't get confused.
+    case "$1" in
+        conda* )
+        echo "using opentelemetry-instrument for command";
+        command="opentelemetry-instrument $*"
+        ;;
+        * )
+        command="$*"
+        ;;
+    esac
+
+    echo "TRACEPARENT prior to otel-cli exec is: \"${TRACEPARENT}\""
+    STEP_TRACEPARENT=$("${SCRIPT_DIR}/rapids-get-telemetry-traceparent" "${JOB_NAME}" "${OTEL_SERVICE_NAME}")
+
+    # otel-cli creates a span for us that bridges the traceparent from the parent process
+    # into the command we're wrapping
+    otel-cli exec \
+        --name "Run instrumented $*" \
+        --force-parent-span-id "$(cut -d'-' -f3 <<<"$STEP_TRACEPARENT")" \
+        --verbose \
+        -- "${command}"
+        RETURN_STATUS=$?
+else
+    echo "Skipping instrumentation, running \"${*}\"";
+    eval "$*"
+    RETURN_STATUS=$?
+fi
+
+exit "${RETURN_STATUS}"

tools/rapids-upload-conda-to-s3

@@ -30,4 +30,4 @@ pkg_name="$(rapids-package-name "$pkg_type")"
 # Where conda build artifacts are output
 path_to_tar_up="${RAPIDS_CONDA_BLD_OUTPUT_DIR}"
 
-rapids-upload-to-s3 "${pkg_name}" "${path_to_tar_up}"
+rapids-otel-wrap rapids-upload-to-s3 "${pkg_name}" "${path_to_tar_up}"

tools/rapids-upload-wheels-to-s3

@@ -20,4 +20,4 @@ if [ "${CI:-false}" = "false" ]; then
   exit 0
 fi
 
-rapids-upload-to-s3 "${pkg_name}" "$@"
+rapids-otel-wrap rapids-upload-to-s3 "${pkg_name}" "$@"