#81 template traceparent into exec command args (#308)

Modifies the default behavior of `otel-cli exec`  so that command-line arguments to the child process are modified to inject the traceparent anywhere `{{traceparent}}` appears in them. This makes patterns like this work:

```shell
export TRACEPARENT=00-5b28ee2bf1f1796cca65bef26fb90c35-9d699b596c3dced3-01
./otel-cli exec --endpoint localhost:4317 -- curl --verbose -H 'traceparent: {{traceparent}}' https://github.com
```

This can be disabled with a new extra argument: `otel-cli exec --tp-disable-inject`.

## Squashed Commits

* add support to template traceparent into exec command args

```
$ go build && TRACEPARENT=00-5b28ee2bf1f1796cca65bef26fb90c35-9d699b596c3dced3-01 ./otel-cli exec --endpoint localhost:4317 -- curl --verbose -H 'traceparent: {{tr
aceparent}}' http://tobert.org
*   Trying 45.33.3.11:80...
* Connected to tobert.org (45.33.3.11) port 80 (#0)
> GET / HTTP/1.1
> Host: tobert.org
> User-Agent: curl/7.81.0
> Accept: */*
> traceparent: 00-5b28ee2bf1f1796cca65bef26fb90c35-433d7e8620e9ccea-01
```

* update CHANGELOG.md

* add an example of injecting traceparent into exec args

* add a test for exec tp injection

* remove unneeded regular expression on cmd output

* add --tp-disable-inject to otel-cli exec along with tests

* group injection tests in a suite

Not a big deal, but makes a touch more sense.

CHANGELOG.md

@@ -1,3 +1,13 @@
+## [0.4.3] - 2024-03-11
+
+Add injection of `{{traceparent}}` to `otel-cli exec` as default behavior, along with
+the `otel-cli exec --tp-disable-inject` to turn it off (old behavior).
+
+### Added
+
+- `otel-cli exec echo {{traceparent}}` is now supported to pass traceparent to child process
+- `otel-cli exec --tp-disable-inject` will disable this new default behavior
+
 ## [0.4.2] - 2023-12-01
 
 The Docker container now builds off `alpine:latest` instead of `scratch`. This

README.md

@@ -70,6 +70,11 @@ otel-cli exec --kind producer "otel-cli exec --kind consumer sleep 1"
 # used by span and exec. use --tp-ignore-env to ignore it even when present
 export TRACEPARENT=00-0af7651916cd43dd8448eb211c80319c-b7ad6b7169203331-01
 
+# you can pass the traceparent to a child via arguments as well
+# {{traceparent}} in any of the command's arguments will be replaced with the traceparent string
+otel-cli exec --name "curl api" -- \
+   curl -H 'traceparent: {{traceparent}}' https://myapi.com/v1/coolstuff
+
 # create a span with a custom start/end time using either RFC3339,
 # same with the nanosecond extension, or Unix epoch, with/without nanos
 otel-cli span --start 2021-03-24T07:28:05.12345Z --end 2021-03-24T07:30:08.0001Z

data_for_test.go

@@ -902,7 +902,7 @@ var suites = []FixtureSuite{
 	// otel-cli exec runs otel-cli exec
 	{
 		{
-			Name: "otel-cli span exec (nested)",
+			Name: "otel-cli exec (nested)",
 			Config: FixtureConfig{
 				CliArgs: []string{
 					"exec", "--name", "outer", "--endpoint", "{{endpoint}}", "--fail", "--verbose", "--",
@@ -915,6 +915,40 @@ var suites = []FixtureSuite{
 			},
 		},
 	},
+	// otel-cli exec echo "{{traceparent}}" and otel-cli exec --tp-disable-inject
+	{
+		{
+			Name: "otel-cli exec with arg injection injects the traceparent",
+			Config: FixtureConfig{
+				CliArgs: []string{
+					"exec", "--endpoint", "{{endpoint}}",
+					"--force-trace-id", "e39280f2980af3a8600ae98c74f2dabf", "--force-span-id", "023eee2731392b4d",
+					"--",
+					"echo", "{{traceparent}}"},
+			},
+			Expect: Results{
+				Config:    otelcli.DefaultConfig().WithEndpoint("{{endpoint}}"),
+				CliOutput: "00-e39280f2980af3a8600ae98c74f2dabf-023eee2731392b4d-01\n",
+				SpanCount: 1,
+			},
+		},
+		{
+			Name: "otel-cli exec --tp-disable-inject returns the {{traceparent}} tag unmodified",
+			Config: FixtureConfig{
+				CliArgs: []string{
+					"exec", "--endpoint", "{{endpoint}}",
+					"--force-trace-id", "e39280f2980af3a8600ae98c74f2dabf", "--force-span-id", "023eee2731392b4d",
+					"--tp-disable-inject",
+					"--",
+					"echo", "{{traceparent}}"},
+			},
+			Expect: Results{
+				Config:    otelcli.DefaultConfig().WithEndpoint("{{endpoint}}"),
+				CliOutput: "{{traceparent}}\n",
+				SpanCount: 1,
+			},
+		},
+	},
 	// validate OTEL_EXPORTER_OTLP_PROTOCOL / --protocol
 	{
 		// --protocol

otelcli/config.go

@@ -56,6 +56,7 @@ func DefaultConfig() Config {
 		BackgroundWait:               false,
 		BackgroundSkipParentPidCheck: false,
 		ExecCommandTimeout:           "",
+		ExecTpDisableInject:          false,
 		StatusCanaryCount:            1,
 		StatusCanaryInterval:         "",
 		SpanStartTime:                "now",
@@ -109,7 +110,8 @@ type Config struct {
 	BackgroundWait               bool   `json:"background_wait" env:""`
 	BackgroundSkipParentPidCheck bool   `json:"background_skip_parent_pid_check"`
 
-	ExecCommandTimeout string `json:"exec_command_timeout" env:"OTEL_CLI_EXEC_CMD_TIMEOUT"`
+	ExecCommandTimeout  string `json:"exec_command_timeout" env:"OTEL_CLI_EXEC_CMD_TIMEOUT"`
+	ExecTpDisableInject bool   `json:"exec_tp_disable_inject" env:"OTEL_CLI_EXEC_TP_DISALBE_INJECT"`
 
 	StatusCanaryCount    int    `json:"status_canary_count"`
 	StatusCanaryInterval string `json:"status_canary_interval"`
@@ -232,6 +234,7 @@ func (c Config) ToStringMap() map[string]string {
 		"background_wait":             strconv.FormatBool(c.BackgroundWait),
 		"background_skip_pid_check":   strconv.FormatBool(c.BackgroundSkipParentPidCheck),
 		"exec_command_timeout":        c.ExecCommandTimeout,
+		"exec_tp_disable_inject":      strconv.FormatBool(c.ExecTpDisableInject),
 		"span_start_time":             c.SpanStartTime,
 		"span_end_time":               c.SpanEndTime,
 		"event_name":                  c.EventName,

otelcli/exec.go

@@ -12,6 +12,7 @@ import (
 	"time"
 
 	"github.com/equinix-labs/otel-cli/otlpclient"
+	"github.com/equinix-labs/otel-cli/w3c/traceparent"
 	"github.com/spf13/cobra"
 	tracev1 "go.opentelemetry.io/proto/otlp/trace/v1"
 )
@@ -47,33 +48,70 @@ otel-cli exec -s "outer span" 'otel-cli exec -s "inner span" sleep 1'`,
 		"timeout for the child process, when 0 otel-cli will wait forever",
 	)
 
+	cmd.Flags().BoolVar(
+		&config.ExecTpDisableInject,
+		"tp-disable-inject",
+		defaults.ExecTpDisableInject,
+		"disable automatically replacing {{traceparent}} with a traceparent",
+	)
+
 	return &cmd
 }
 
 func doExec(cmd *cobra.Command, args []string) {
 	ctx := cmd.Context()
 	config := getConfig(ctx)
+	span := config.NewProtobufSpan()
 
 	// put the command in the attributes, before creating the span so it gets picked up
 	config.Attributes["command"] = args[0]
 	config.Attributes["arguments"] = ""
 
 	// no deadline if there is no command timeout set
 	cancelCtxDeadline := func() {}
+	// fork the context for the command so its deadline doesn't impact the otlpclient ctx
 	cmdCtx := ctx
 	cmdTimeout := config.ParseExecCommandTimeout()
 	if cmdTimeout > 0 {
 		cmdCtx, cancelCtxDeadline = context.WithDeadline(ctx, time.Now().Add(cmdTimeout))
 	}
 
+	// pass the existing env but add the latest TRACEPARENT carrier so e.g.
+	// otel-cli exec 'otel-cli exec sleep 1' will relate the spans automatically
+	childEnv := []string{}
+
+	// set the traceparent to the current span to be available to the child process
+	var tp traceparent.Traceparent
+	if config.GetIsRecording() {
+		tp = otlpclient.TraceparentFromProtobufSpan(span, config.GetIsRecording())
+		childEnv = append(childEnv, fmt.Sprintf("TRACEPARENT=%s", tp.Encode()))
+		// when not recording, and a traceparent is available, pass it through
+	} else if !config.TraceparentIgnoreEnv {
+		tp := config.LoadTraceparent()
+		if tp.Initialized {
+			childEnv = append(childEnv, fmt.Sprintf("TRACEPARENT=%s", tp.Encode()))
+		}
+	}
+
 	var child *exec.Cmd
 	if len(args) > 1 {
-		// CSV-join the arguments to send as an attribute
 		buf := bytes.NewBuffer([]byte{})
-		csv.NewWriter(buf).WriteAll([][]string{args[1:]})
+		tpArgs := make([]string, len(args)-1)
+
+		if config.ExecTpDisableInject {
+			copy(tpArgs, args[1:])
+		} else {
+			// loop over the args replacing {{traceparent}} with the current tp
+			for i, arg := range args[1:] {
+				tpArgs[i] = strings.Replace(arg, "{{traceparent}}", tp.Encode(), -1)
+			}
+		}
+
+		// CSV-join the arguments to send as an attribute
+		csv.NewWriter(buf).WriteAll([][]string{tpArgs})
 		config.Attributes["arguments"] = buf.String()
 
-		child = exec.CommandContext(cmdCtx, args[0], args[1:]...)
+		child = exec.CommandContext(cmdCtx, args[0], tpArgs...)
 	} else {
 		child = exec.CommandContext(cmdCtx, args[0])
 	}
@@ -83,30 +121,13 @@ func doExec(cmd *cobra.Command, args []string) {
 	child.Stdout = os.Stdout
 	child.Stderr = os.Stderr
 
-	// pass the existing env but add the latest TRACEPARENT carrier so e.g.
-	// otel-cli exec 'otel-cli exec sleep 1' will relate the spans automatically
-	child.Env = []string{}
-
 	// grab everything BUT the TRACEPARENT envvar
 	for _, env := range os.Environ() {
 		if !strings.HasPrefix(env, "TRACEPARENT=") {
-			child.Env = append(child.Env, env)
-		}
-	}
-
-	span := config.NewProtobufSpan()
-
-	// set the traceparent to the current span to be available to the child process
-	if config.GetIsRecording() {
-		tp := otlpclient.TraceparentFromProtobufSpan(span, config.GetIsRecording())
-		child.Env = append(child.Env, fmt.Sprintf("TRACEPARENT=%s", tp.Encode()))
-		// when not recording, and a traceparent is available, pass it through
-	} else if !config.TraceparentIgnoreEnv {
-		tp := config.LoadTraceparent()
-		if tp.Initialized {
-			child.Env = append(child.Env, fmt.Sprintf("TRACEPARENT=%s", tp.Encode()))
+			childEnv = append(childEnv, env)
 		}
 	}
+	child.Env = childEnv
 
 	// ctrl-c (sigint) is forwarded to the child process
 	signals := make(chan os.Signal, 10)
@@ -119,6 +140,7 @@ func doExec(cmd *cobra.Command, args []string) {
 		close(signalsDone)
 	}()
 
+	span.StartTimeUnixNano = uint64(time.Now().UnixNano())
 	if err := child.Run(); err != nil {
 		span.Status = &tracev1.Status{
 			Message: fmt.Sprintf("exec command failed: %s", err),