feat(gateway): trace context header support (#256)

* feat(examples): wrap handler with OTel propagation * feat: add tracing sub-package based on Kubo * docs: update tracing according to comments * docs(tracing): clarify that kubo is an example Context: ipfs-inactive/bifrost-gateway#68 https://www.w3.org/TR/trace-context/ --------- Co-authored-by: Marcin Rataj <[email protected]>
ipfs · Apr 11, 2023 · 5d6c73c · 5d6c73c
1 parent 999d939
commit 5d6c73c
Show file tree

Hide file tree

Showing 14 changed files with 636 additions and 28 deletions.
diff --git a/docs/tracing.md b/docs/tracing.md
@@ -0,0 +1,142 @@
+# Tracing
+
+Tracing across the stack follows, as much as possible, the [Open Telemetry]
+specifications. Configuration environment variables are specified in the
+[OpenTelemetry Environment Variable Specification].
+
+We use the [opentelemtry-go] package, which currently does not have default support
+for the `OTEL_TRACES_EXPORTER` environment variables. Therefore, we provide some
+helper functions under [`boxo/tracing`](../tracing/) to support these.
+
+In this document, we document the quirks of our custom support for the `OTEL_TRACES_EXPORTER`,
+as well as examples on how to use tracing, create traceable headers, and how
+to use the Jaeger UI. The [Gateway examples](../examples/gateway/) fully support Tracing.
+
+- [Environment Variables](#environment-variables)
+  - [`OTEL_TRACES_EXPORTER`](#otel_traces_exporter)
+  - [`OTLP Exporter`](#otlp-exporter)
+  - [`Jaeger Exporter`](#jaeger-exporter)
+  - [`Zipkin Exporter`](#zipkin-exporter)
+  - [`File Exporter`](#file-exporter)
+  - [`OTEL_PROPAGATORS`](#otel_propagators)
+- [Using Jaeger UI](#using-jaeger-ui)
+- [Generate `traceparent` Header](#generate-traceparent-header)
+
+## Environment Variables
+
+For advanced configurations, such as ratio-based sampling, please see also the
+[OpenTelemetry Environment Variable Specification].
+
+### `OTEL_TRACES_EXPORTER`
+
+Specifies the exporters to use as a comma-separated string. Each exporter has a
+set of additional environment variables used to configure it. The following values
+are supported:
+
+- `otlp`
+- `jaeger`
+- `zipkin`
+- `stdout`
+- `file` -- appends traces to a JSON file on the filesystem
+
+Default: `""` (no exporters)
+
+### `OTLP Exporter`
+
+Unless specified in this section, the OTLP exporter uses the environment variables
+documented in [OpenTelemetry Protocol Exporter].
+
+#### `OTEL_EXPORTER_OTLP_PROTOCOL`
+Specifies the OTLP protocol to use, which is one of:
+
+- `grpc`
+- `http/protobuf`
+
+Default: `"grpc"`
+
+### `Jaeger Exporter`
+
+See [Jaeger Exporter](https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/sdk-environment-variables.md#jaeger-exporter).
+
+### `Zipkin Exporter`
+
+See [Zipkin Exporter](https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/sdk-environment-variables.md#zipkin-exporter).
+
+### `File Exporter`
+
+#### `OTEL_EXPORTER_FILE_PATH`
+
+Specifies the filesystem path for the JSON file.
+
+Default: `"$PWD/traces.json"`
+
+### `OTEL_PROPAGATORS`
+
+See [General SDK Configuration](https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/sdk-environment-variables.md#general-sdk-configuration).
+
+## Using Jaeger UI
+
+One can use the `jaegertracing/all-in-one` Docker image to run a full Jaeger stack
+and configure the Kubo daemon, or gateway examples, to publish traces to it. Here, in an
+ephemeral container:
+
+```console
+$ docker run --rm -it --name jaeger \
+    -e COLLECTOR_ZIPKIN_HOST_PORT=:9411 \
+    -p 5775:5775/udp \
+    -p 6831:6831/udp \
+    -p 6832:6832/udp \
+    -p 5778:5778 \
+    -p 16686:16686 \
+    -p 14268:14268 \
+    -p 14269:14269 \
+    -p 14250:14250 \
+    -p 9411:9411 \
+    jaegertracing/all-in-one
+```
+
+Then, in other terminal, start the app that uses `boxo/tracing` internally (e.g., a Kubo daemon), with Jaeger exporter enabled:
+
+```
+$ OTEL_TRACES_EXPORTER=jaeger ipfs daemon
+```
+
+Finally, the [Jaeger UI] is available at http://localhost:16686.
+
+## Generate `traceparent` Header
+
+If you want to trace a specific request and want to have its tracing ID, you can
+generate a `Traceparent` header. According to the [Trace Context] specification,
+the header is formed as follows:
+
+> ```
+> version-format   = trace-id "-" parent-id "-" trace-flags
+> trace-id         = 32HEXDIGLC  ; 16 bytes array identifier. All zeroes forbidden
+> parent-id        = 16HEXDIGLC  ; 8 bytes array identifier. All zeroes forbidden
+> trace-flags      = 2HEXDIGLC   ; 8 bit flags. Currently, only one bit is used. See below for details
+> ```
+
+To generate a valid `Traceparent` header value, the following script can be used:
+
+```bash
+version="00" # fixed in spec at 00
+trace_id="$(cat /dev/urandom | tr -dc 'a-f0-9' | fold -w 32 | head -n 1)"
+parent_id="00$(cat /dev/urandom | tr -dc 'a-f0-9' | fold -w 14 | head -n 1)"
+trace_flag="01"   # sampled
+traceparent="$version-$trace_id-$parent_id-$trace_flag"
+echo $traceparent
+```
+
+**NOTE**: the `tr` command behaves differently on macOS. You may want to install
+the GNU `tr` (`gtr`) and use it instead.
+
+Then, the value can be passed onto the request with `curl -H "Traceparent: $traceparent" URL`.
+If using Jaeger, you can now search by the trace with ID `$trace_id` and see
+the complete trace of this request.
+
+[Open Telemetry]: https://opentelemetry.io/
+[opentelemetry-go]: https://github.com/open-telemetry/opentelemetry-go
+[Trace Context]: https://www.w3.org/TR/trace-context
+[OpenTelemetry Environment Variable Specification]: https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/sdk-environment-variables.md
+[OpenTelemetry Protocol Exporter]: https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/protocol/exporter.md
+[Jaeger UI]: https://github.com/jaegertracing/jaeger-ui
diff --git a/examples/gateway/car/main.go b/examples/gateway/car/main.go
@@ -1,6 +1,7 @@
 package main
 
 import (
+	"context"
 	"flag"
 	"io"
 	"log"
@@ -17,16 +18,29 @@ import (
 )
 
 func main() {
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
 	carFilePtr := flag.String("c", "", "path to CAR file to back this gateway from")
 	port := flag.Int("p", 8040, "port to run this gateway from")
 	flag.Parse()
 
+	// Setups up tracing. This is optional and only required if the implementer
+	// wants to be able to enable tracing.
+	tp, err := common.SetupTracing(ctx, "CAR Gateway Example")
+	if err != nil {
+		log.Fatal(err)
+	}
+	defer (func() { _ = tp.Shutdown(ctx) })()
+
+	// Sets up a block service based on the CAR file.
 	blockService, roots, f, err := newBlockServiceFromCAR(*carFilePtr)
 	if err != nil {
 		log.Fatal(err)
 	}
 	defer f.Close()
 
+	// Creates the gateway API with the block service.
 	gwAPI, err := gateway.NewBlocksGateway(blockService)
 	if err != nil {
 		log.Fatal(err)

diff --git a/examples/gateway/common/handler.go b/examples/gateway/common/handler.go
@@ -6,6 +6,7 @@ import (
 	"github.com/ipfs/boxo/gateway"
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/client_golang/prometheus/promhttp"
+	"go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp"
 )
 
 func NewHandler(gwAPI gateway.IPFSBackend) http.Handler {
@@ -58,10 +59,16 @@ func NewHandler(gwAPI gateway.IPFSBackend) http.Handler {
 	var handler http.Handler
 	handler = gateway.WithHostname(mux, gwAPI, publicGateways, noDNSLink)
 
-	// Finally, wrap with the withConnect middleware. This is required since we use
+	// Then, wrap with the withConnect middleware. This is required since we use
 	// http.ServeMux which does not support CONNECT by default.
 	handler = withConnect(handler)
 
+	// Finally, wrap with the otelhttp handler. This will allow the tracing system
+	// to work and for correct propagation of tracing headers. This step is optional
+	// and only required if you want to use tracing. Note that OTel must be correctly
+	// setup in order for this to have an effect.
+	handler = otelhttp.NewHandler(handler, "Gateway.Request")
+
 	return handler
 }
 

diff --git a/examples/gateway/common/tracing.go b/examples/gateway/common/tracing.go
@@ -0,0 +1,61 @@
+package common
+
+import (
+	"context"
+
+	"github.com/ipfs/boxo/tracing"
+	"go.opentelemetry.io/contrib/propagators/autoprop"
+	"go.opentelemetry.io/otel"
+	"go.opentelemetry.io/otel/sdk/resource"
+	"go.opentelemetry.io/otel/sdk/trace"
+	semconv "go.opentelemetry.io/otel/semconv/v1.4.0"
+)
+
+// SetupTracing sets up the tracing based on the OTEL_* environment variables,
+// and the provided service name. It returns a trace.TracerProvider.
+func SetupTracing(ctx context.Context, serviceName string) (*trace.TracerProvider, error) {
+	tp, err := NewTracerProvider(ctx, serviceName)
+	if err != nil {
+		return nil, err
+	}
+
+	// Sets the default trace provider for this process. If this is not done, tracing
+	// will not be enabled. Please note that this will apply to the entire process
+	// as it is set as the default tracer, as per OTel recommendations.
+	otel.SetTracerProvider(tp)
+
+	// Configures the default propagators used by the Open Telemetry library. By
+	// using autoprop.NewTextMapPropagator, we ensure the value of the environmental
+	// variable OTEL_PROPAGATORS is respected, if set. By default, Trace Context
+	// and Baggage are used. More details on:
+	// https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/sdk-environment-variables.md
+	otel.SetTextMapPropagator(autoprop.NewTextMapPropagator())
+
+	return tp, nil
+}
+
+// NewTracerProvider creates and configures a TracerProvider.
+func NewTracerProvider(ctx context.Context, serviceName string) (*trace.TracerProvider, error) {
+	exporters, err := tracing.NewSpanExporters(ctx)
+	if err != nil {
+		return nil, err
+	}
+
+	options := []trace.TracerProviderOption{}
+
+	for _, exporter := range exporters {
+		options = append(options, trace.WithBatcher(exporter))
+	}
+
+	r, err := resource.Merge(
+		resource.Default(),
+		resource.NewSchemaless(
+			semconv.ServiceNameKey.String(serviceName),
+		),
+	)
+	if err != nil {
+		return nil, err
+	}
+	options = append(options, trace.WithResource(r))
+	return trace.NewTracerProvider(options...), nil
+}
diff --git a/examples/gateway/proxy/blockstore.go b/examples/gateway/proxy/blockstore.go
@@ -5,11 +5,11 @@ import (
 	"fmt"
 	"io"
 	"net/http"
-	"net/url"
 
 	"github.com/ipfs/boxo/exchange"
 	blocks "github.com/ipfs/go-block-format"
 	"github.com/ipfs/go-cid"
+	"go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp"
 )
 
 type proxyExchange struct {
@@ -19,7 +19,9 @@ type proxyExchange struct {
 
 func newProxyExchange(gatewayURL string, client *http.Client) exchange.Interface {
 	if client == nil {
-		client = http.DefaultClient
+		client = &http.Client{
+			Transport: otelhttp.NewTransport(http.DefaultTransport),
+		}
 	}
 
 	return &proxyExchange{
@@ -29,17 +31,13 @@ func newProxyExchange(gatewayURL string, client *http.Client) exchange.Interface
 }
 
 func (e *proxyExchange) fetch(ctx context.Context, c cid.Cid) (blocks.Block, error) {
-	u, err := url.Parse(fmt.Sprintf("%s/ipfs/%s?format=raw", e.gatewayURL, c))
+	urlStr := fmt.Sprintf("%s/ipfs/%s?format=raw", e.gatewayURL, c)
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, urlStr, nil)
 	if err != nil {
 		return nil, err
 	}
-	resp, err := e.httpClient.Do(&http.Request{
-		Method: http.MethodGet,
-		URL:    u,
-		Header: http.Header{
-			"Accept": []string{"application/vnd.ipld.raw"},
-		},
-	})
+	req.Header.Set("Accept", "application/vnd.ipld.raw")
+	resp, err := e.httpClient.Do(req)
 	if err != nil {
 		return nil, err
 	}

diff --git a/examples/gateway/proxy/main.go b/examples/gateway/proxy/main.go
@@ -1,6 +1,7 @@
 package main
 
 import (
+	"context"
 	"flag"
 	"log"
 	"net/http"
@@ -15,10 +16,21 @@ import (
 )
 
 func main() {
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
 	gatewayUrlPtr := flag.String("g", "", "gateway to proxy to")
 	port := flag.Int("p", 8040, "port to run this gateway from")
 	flag.Parse()
 
+	// Setups up tracing. This is optional and only required if the implementer
+	// wants to be able to enable tracing.
+	tp, err := common.SetupTracing(ctx, "CAR Gateway Example")
+	if err != nil {
+		log.Fatal(err)
+	}
+	defer (func() { _ = tp.Shutdown(ctx) })()
+
 	// Sets up a blockstore to hold the blocks we request from the gateway
 	// Note: in a production environment you would likely want to choose a more efficient datastore implementation
 	// as well as one that has a way of pruning storage so as not to hold data in memory indefinitely.