Add http client/server example #131
Conversation
|
I added a note that this requires the fix from #118, which I'm working on next. |
examples/http/client.rb
Outdated
| exporter = Examples::Exporters::Console.new | ||
| processor = SDK::Trace::Export::SimpleSpanProcessor.new(exporter) | ||
| tracer = OpenTelemetry.tracer_factory.tracer('http.client', 'semver:1.0') | ||
| tracer.add_span_processor(processor) |
There was a problem hiding this comment.
I think that our client/server examples should demonstrate how to handle context propagation at process boundaries. So, the server example should show the use of extract to extract a span context and parent a span in process. The client example should use inject to inject a span context into the http headers of the outbound request.
For the server example, I think it'd make sense to demonstrate it using a Rack middleware as it's a realistic example of where a user will be at a request boundary.
For the client example, we can stick with the manual request with Faraday, but add a step to inject the current span's context into the outbound request.
|
It's possible that we could de-grapify the example from: #132 and use it here. |
|
I completely agree with Matt. A simple example with Rack middleware to extract + creation of a child span in the server, and root span + Faraday + inject in the client is 💯 |
|
@mwear @fbogsany thanks for the feedback. I'll get the direct code feedback in here today/this weekend. As far as the context propagation - I mentioned that in the PR description, I would LOVE to have that in there but it isn't really clear how ours works. I guess it's the 🐔 + 🥚 problem of having no examples while writing the examples. 😄 Any chance I could get some guidance on propagation? I think there's some usefulness in having really really simple examples. I'm not sure if that means this example should use Rack middleware or not. I realize that most people looking at our examples may be more than passingly familiar with tracing concepts but as a complete newbie I found the Go example to be very clear. Having said all that, I'm happy to take this in the direction you think it should go. |
|
@mwear @fbogsany just to add a little bit of background to my question about context propagation: The docs for It looks like other libraries have propagators that know how to interact with the carrier. Does that sound right? Are we planning to create propagators for typical use cases? |
|
@elskwid Currently there is an HTTPTextFormat propagator. The API is based off of this specification. The blocks play the role of the This is a diversion from OpenTracing, where inject and extract know how to get or set the context in the carrier. Your examples have shown us where our APIs could use some improvement previously, I'm curious if you can point out the differences you see with other languages here. As far as distributed context goes, we don't yet have an implementation for the SDK. This is something that is very likely to change in light of context propagation OTEP. |
examples/http/server.rb
Outdated
| span.set_attribute('http.method', env['REQUEST_METHOD']) | ||
| span.set_attribute('http.route', env['PATH_INFO']) | ||
| span.set_attribute('http.url', env['REQUEST_URI']) | ||
| span.add_event(name: 'handle http.server.get') |
There was a problem hiding this comment.
This is a slightly odd example. A more realistic one might be the size of the response body (after the @app.call).
There was a problem hiding this comment.
@fbogsany as an event? Okay, I was reading the name event literally as in, this thing happened at this time.
There was a problem hiding this comment.
That's true, but for this particular operation, that's implicitly the start of the span, so the event isn't that interesting.
I like this blog post as an example of events in a trace (look at the pictures rather than the text): https://medium.com/google-cloud/tracing-google-cloud-8f0c8ba8181c. There, the events are informative little snippets about things that occurred during the span. Some of them may be numeric values (like 146 bytes sent and 40 bytes received) while others are things like DNSStart and DNSDone to record when DNS lookup happened. The latter could conceivably be a span itself, but doesn't have to be.
There was a problem hiding this comment.
I guess my larger point here is that events should typically be used to add colour to the span, and this specific one doesn't really do that.
There was a problem hiding this comment.
@fbogsany thank you for all the context. I removed the events to clarify the example and keep it relatively simple.
Adds simple client and server examples using a console exporter to demonstrate instrumentation.
Overview
Adds a simple http client/server example using a Sinatra server and Faraday client. The example is modeled after the Go http example.
Note: the console exporter won't be called until we merge in the
in_spanfix from #118Details
examplestop-level directoryhttpexample directory that contains the client, server, and a small helper scriptex-httpdocker-compose server to simplify setup/running of the exampleDepends on:
This PR depends on #144 to function but it can be merged in any order.
Finishes
#130 - Add simple HTTP client/server example