Update http Semconv to v1.20.0

[MadVikingGod]

This updates all http instrumentation (anything using internal/shared/semconvutil) to v1.20.0.

In addition, I did an audit of the currently recommended semantic conventions for HTTP and put a comment of what is included and what is not to hopefully be easier to reference on the next update.

This doesn't go to 1.21 as that included a new set of logic around the OTEL_SEMCONV_STABILITY_OPT_IN environment variable. That will require a release using 1.20.0 for this to work correctly.

[codecov]

Codecov Report

Merging #4320 (59ecc7a) into main (f7ca3ec) will increase coverage by 0.2%.
The diff coverage is 94.0%.

Additional details and impacted files

@@           Coverage Diff           @@
##            main   #4320     +/-   ##
=======================================
+ Coverage   80.8%   81.1%   +0.2%     
=======================================
  Files        150     150             
  Lines      10374   10759    +385     
=======================================
+ Hits        8388    8731    +343     
- Misses      1844    1872     +28     
- Partials     142     156     +14     

Files	Coverage Δ
...estful/otelrestful/internal/semconvutil/netconv.go	100.0% <100.0%> (ø)
...hub.com/emicklei/go-restful/otelrestful/restful.go	100.0% <ø> (ø)
...ation/github.com/gin-gonic/gin/otelgin/gintrace.go	83.1% <ø> (ø)
...-gonic/gin/otelgin/internal/semconvutil/netconv.go	100.0% <100.0%> (ø)
...orilla/mux/otelmux/internal/semconvutil/netconv.go	100.0% <100.0%> (ø)
...trumentation/github.com/gorilla/mux/otelmux/mux.go	89.2% <ø> (ø)
...entation/github.com/labstack/echo/otelecho/echo.go	100.0% <ø> (ø)
...tack/echo/otelecho/internal/semconvutil/netconv.go	100.0% <100.0%> (ø)
...ron.v1/otelmacaron/internal/semconvutil/netconv.go	100.0% <100.0%> (ø)
...ntation/gopkg.in/macaron.v1/otelmacaron/macaron.go	91.4% <ø> (ø)
... and 12 more

[dmathieu]

Nice! :)
How about a changelog entry?

[MrAlias]

From SIG meeting:

• @MadVikingGod ran a semconv checker against this and found things that need to be changed or documented. He plans to update with that info/changes.

[MadVikingGod]

So they aren't lost, this was the results of running the example from the sig meeting:

2023/10/26 12:54:07 INFO missing attributes type=trace section=span scope.name=go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp name=http.server.root attributes="[http.client_ip net.sock.host.addr net.sock.host.port http.route]"
2023/10/26 12:54:07 INFO extra attributes type=trace section=span scope.name=go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp name=http.server.root attributes="[net.sock.peer.port user_agent.original net.protocol.version http.method net.sock.peer.addr http.wrote_bytes http.status_code]"
2023/10/26 12:54:07 INFO missing attributes type=trace section=span scope.name=go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp name=http.client.GET. attributes=[http.resend_count]
2023/10/26 12:54:07 INFO extra attributes type=trace section=span scope.name=go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp name=http.client.GET. attributes="[http.status_code http.response_content_length http.method]"

From this The server is missing:

• http.client_ip - Recommended
• net.sock.host.addr - Opt-In
• net.sock.host.port - Conditionally Required: If defined for the address family and if different than net.host.port and if net.sock.host.addr is set. In other cases, it is still recommended to set this.
• http.route - Conditionally Required: If and only if it's available

And the Client is missing:

• http.resend_count - Recommended: if and only if request was retried.

Note these may be optional or not accessible.

[MadVikingGod]

I think we are compliant:

• http.client_ip - Is recorded from X-Forwarded-For, but we didn't have a proxy so it is absent in the test.
• net.sock.host.addr - It is Opt-in, we don't have the config option for it, and I don't know if we can access this data from within the handler.
• net.sock.host.port - Requires net.sock.host.addr, and isn't included for the same reasons.
• http.route - This is stored at the router/mux level and not accessible to the handler.
• http.resend_count - Resends must happen in the application code above the instrumented client. A client request will only ever send once.

[pellared]

I think that handling of net.protocol.name and net.protocol.version should be changed.

References:

• https://github.com/open-telemetry/opentelemetry-specification/blob/v1.20.0/specification/trace/semantic_conventions/http.md
• https://github.com/open-telemetry/opentelemetry-specification/blob/v1.20.0/specification/metrics/semantic_conventions/http-metrics.md

Besides I only left some nit comments regarding the comments.

[pellared]

@MadVikingGod Any update on this PR?