Define ignoreURLs generic use case and weigh different technical approaches

[surbhiia]

Use case

• Ignore URLs matching a pattern from being traced by HTTP instrumentations.
• If URL is ignored on the client side, server should also get the knowledge that this was ignored on client.

Techincal Approaches

1. Sampler (Exists)

val urlBlocklistSampler = object : Sampler {
    override fun shouldSample(
        parentContext: Context,
        traceId: String,
        name: String,
        spanKind: SpanKind,
        attributes: Attributes,
        parentLinks: List<LinkData>,
    ): SamplingResult {
        val url = attributes.get(AttributeKey.stringKey("url.full"))
        if (url != null && shouldIgnoreUrl(url)) { //shouldIgnoreUrl function uses ignoreURL pattern config provided by customer
            return SamplingResult.drop()
        }
        return Sampler.parentBased(Sampler.alwaysOn()).shouldSample(
            parentContext, traceId, name, spanKind, attributes, parentLinks,
        )
    }
    override fun getDescription(): String = "UrlBlocklistSampler"
}

// When building Opentelemetry Tracer provider, add the sampler
SdkTracerProvider.builder().setSampler(urlBlocklistSampler)

2. Instrumentation level configuration (Needs Implementation)

Upgrade the TracingInterceptor and ConnectionErrorInterceptor

2.1 ConnectionErrorInterceptor

@Override
public Response intercept(Chain chain) throws IOException {
    Request request = chain.request();
   
    //Add this if statement to support ignoreURLs use case
    if (shouldIgnoreUrl(request.url())) {
        return chain.proceed(request);
    }

    Context parentContext = Context.current();
    Response response = null;
    Throwable error = null;
    Instant startTime = Instant.now();
    try {
        response = chain.proceed(request);
        return response;
    } catch (Throwable t) {
        error = t;
        throw t;
    } finally {
        if (HttpClientRequestResendCount.get(parentContext) == 0) {
            if (instrumenter.shouldStart(parentContext, chain)) {
                InstrumenterUtil.startAndEnd(
                    instrumenter, parentContext, chain, response, error, startTime, Instant.now());
            }
        }
    }
}

2.2 TracingInterceptor

@Override
public Response intercept(Chain chain) throws IOException {
    Request request = chain.request();
    Context parentContext = Context.current();

   //Add this if statement to support ignoreURLs use case
    if (shouldIgnoreUrl(request.url())) { 
       <Generate a span id and propagate w3c traceparent context to server with sample flag as dropped even if url is ignored>    
        return chain.proceed(injected);
    }

    if (!instrumenter.shouldStart(parentContext, chain)) {
        return chain.proceed(chain.request());
    }
    Context context = instrumenter.start(parentContext, chain);
    request = injectContextToRequest(request, context);
    Response response;
    try (Scope ignored = context.makeCurrent()) {
        response = chain.proceed(request);
    } catch (Throwable t) {
        instrumenter.end(context, chain, null, t);
        throw t;
    }
    instrumenter.end(context, chain, response, null);
    return response;
}

Note - Using suppression context key is not an option as that would mean that all requests originating within that context independent of the URL will be ignored and that is not something we want.

Open Questions

• Is use case complete? Is there more or less things we need to consider?
• What approach is preferred and why?

[LikeTheSalad]

It seems like the sampler option is the simplest and global, which I think is ideal. We're going to need to access the URL from the sampler, though, to make the decision, which I'm guessing should always be available, but just in case, I wanted to mention that the OTel Java SDK allows spans to get attributes added after the span starts, and last time I checked, samplers cannot see the post-start span attributes. Again, it's probably not something to worry too much about, because I expect instrumentations to always set the HTTP spans urls before starting them, but still, I mention it in case someone sees something I might've missed here.

[breedx-splk]

The description suggests that the sampler already exists. Is that right @surbhiia ?