integrate and implement open telemetry tracing

[achen2289]

Test plan -

• Set Open Telemetry tracer and ensured query state changes are tracked as well as query completes execution successfully.
• Ensured spans of same query are grouped under same parent span / traceId and any prior context, if provided, is properly propagated
• Ensured baggage propagation is supported as well as set as tags of new spans
• Tested trace exporting (export to Datadog agent)
• Configured without open telemetry as well as with SimpleTracer / NoopTracer to ensure backwards compatibility
• Tested edge cases with system config not matching plugin config

== RELEASE NOTES ==

General Changes
* Introduced a new plugin type for tracing. The legacy tracer module can be replaced by the new plugin for loading customized tracing infrastructure.

Open Telemetry Changes
* Introduced a new Open Telemetry tracer implementation. Users are able to enable the tracer by installing the ``presto-open-telemetry`` plugin and updating the application configuration (``config.properties``). Open Telemetry tracer can take in propagated context (only B3 specification currently supported) and baggage (W3C specification) headers, if provided, and inject into new traces / spans. Traces can be exported to any specified backend with the ``OTEL_EXPORTER_OTLP_ENDPOINT`` environment variable.

[linux-foundation-easycla]

The committers listed above are authorized under a signed CLA.

• ✅ login: achen2289 / name: Alex Chen (21888a73662611aaee0fd34fc6ee20aa48646f34, f1099bde939e1aa1cc1568d93faa9f33a8285ace, 1eab0578bfbad5c047744cf68c9ad5e4fdd828d5, 9c47baad2aaeeca5a41fe418ccfce1400932cfc2)

[cliandy]

@highker could we get a first pass on this?

[tanjialiang]

Reviewed the first commit

[tanjialiang]

Do we need to take care of any unfinished Span?

[tanjialiang]

Can you help me understanding the closing logic of opentelemetry here? Thank you

[achen2289]

I am looking into this

[achen2289]

@tanjialiang Addressed by ending any unended spans at the time of ending trace.

[highker]

high-level comments only

[achen2289]

Hey @highker, I want to try to get my changes pushed through by the end of next week if possible. If there's anything in this PR that still needs further discussion or you still want revised, I'm happy to remove it from this PR and address it as a new one for the future if that works.

[highker]

Please squash all commits into one.

I have one major comment in the end. Please check that one first.

[highker]

Let's remove these headers to presto-open-telemetry. Check my other comment

[highker]

Can we use the header instead of session property to control the flags?

[highker]

Let's refactor the code a bit so we don't have to use session property or the header explicitly. Check my other comment

[highker]

Move this to presto-open-telemetry module. Check my other comment

[highker]

nit

@Nullable
private TracerProvider tracerProvider;

[highker]

check non null

[highker]

directly return instead of creating openTelemetry

[highker]

[Major] this is my major comment.

Similar to many other "handle" classes in our codebase, let's create a handle interface in com.facebook.presto.spi.tracing:

public interface TracerHandle
{
}

Change the signature of getNewTracer to

Tracer getNewTracer(TracerHandle handle);

Create a new method of TracerProvider:

/**
 * The method returns a function to take a set of HTTP headers and generate the tracer handle
 */
Function<Set<String>, TracerHandle> getHandleGenerator();

In your OpenTelemetry implementation, you should have

public class OpenTelemetryTracerHandle
        implements TracerHandle
{
    private final String traceToken;
    private final String contextPropagator;
    private final String propagatedContext;
    private final String baggage;

    ... 
}

In your OpenTelemetryTracerProvider, you should have

@Override
public Tracer getNewTracer(TracerHandle handle)
{
    OpenTelemetryTracerHandle tracerHandle = (TracerHandle) handle;
    return new OpenTelemetryTracer(tracerHandle.getXXX, ...);
}

In your getHandleGenerator implementation, move all the logics you have in presto-main here so we can hide all those headers and specialized extraction methods.

[highker]

add javadoc

[highker]

add javadoc

[achen2289]

@highker Trace handle implementation done

[highker]

awesome change! All nits. Otherwise, good to go

[highker]

I know it's a bit anti pattern; but in order to keep the logic so we don't pollute the traceToken below. Check my comment below.

[highker]

Add a new branch here: else if (trimEmptyToNull(servletRequest.getHeader(PRESTO_TRACE_TOKEN)) != null) or something like that. And inside the body, we can either create a SimpleTracerProvider and get the token or directly assign traceToken from PRESTO_TRACE_TOKEN

[achen2289]

updated trace token to replicate similar logic as before

[highker]

Use immutableMap builder

[highker]

nit: it's not critical but good to follow:

we don't usually name a function or var with word "Map"; instead, we use getRequestHeaders. Same for the variable headerMap --> headers. And everything else in the PR.

[highker]

nit: requestHeaders

[highker]

remove else; it's redundant

[highker]

remove else; it's redundant

[highker]

remove else; it's redundant

[highker]

this is not a generic error; check my other comment

[highker]

two options:

• you can create your own error code class within the module with a brand new base code; check any other implementation of ErrorCodeSupplier; or
• you can just throw runtime error or put generic code to be a catch-all to save the effort, which is not recommended but fine

[achen2289]

@highker Addressed nits, and made one further change for open telemetry baggage propagation. Put it in new commit so it's easier to see and will squash if it looks good.

[highker]

Thanks. Please squash all commits into one.

[highker]

more nits; otherwise LGTM

[highker]

nit: ImmutableMap<String, String> -> Map<String, String>

[highker]

nit: Optional.of(tunnelTraceId)

it's not null already

[highker]

nit: ImmutableMap<String, String> -> Map<String, String>

[highker]

comment not addressed? https://github.com/prestodb/presto/pull/18534/files#r1035527320

[highker]

We need a base so we don't collide with other error codes. Find a unique one closest to an existing one. I assume it's starting with 0x05... something

[highker]

I'm not expert in open telemetry but are we sure the logic is correct by modifying static variables?

[achen2289]

I believe it should be fine. If we have a different context propagator, we need to rebuild the open telemetry instance for all tracers that may exist.

[highker]

One last nit

[highker]

Can we delete this class from this file?

[achen2289]

Done

[achen2289]

@highker Thanks for the helpful reviews and merge!