Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Support experimental declarative configuration #12265

Merged
merged 8 commits into from
Oct 16, 2024

Conversation

jack-berg
Copy link
Member

@jack-berg jack-berg commented Sep 16, 2024

When declarative config is used, AutoConfigureUtil.getConfig(AutoConfiguredOpenTelemetrySdk) returns null.

This adjusts all calls to this method to be able to accommodate the null response.

Introduces StructuredConfigPropertiesBridge, which implements ConfigProperties when structured config is used and reads agent / instrumentation configuration from the YAML source. See comment for more details.

@jack-berg jack-berg requested a review from a team September 16, 2024 20:49
@jack-berg jack-berg marked this pull request as draft September 16, 2024 20:49
@github-actions github-actions bot added the test native This label can be applied to PRs to trigger them to run native tests label Sep 16, 2024
import java.util.Map;
import org.junit.jupiter.api.Test;

class StructuredConfigPropertiesBridgeTest {
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

StructuredConfigPropertiesBridge is my proposal for how we keep a sane configuration story in the agent / instrumentation despite the SDK having either flat sys prop / env var or structured YAML configuration options.

The idea is the implement the ConfigProperties interface using a backing StructuredConfigProperties instance, and have special logic to determine how the [standard java agent](https://opentelemetry.io/docs/zero-code/java/agent/disable/ properties map to a structured representation.

  • Only properties starting with otel.instrumentation. resolve
  • otel.instrumentation. refers to the node at .instrumentation.java in the declarative config data model
  • To resolve a property, we take the portion after otel.instrumentation and break it up into segments. For each segment (N-1) we walk down the StructuredConfigProperties tree, and read the final segment from the resolve node.

So given the following YAML, reading configProperties.getBoolean("otel.instrumentation.common.default-enabled") resolves true:

instrumentation:
  java:
    common:
      default-enabled: true

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

See open-telemetry/opentelemetry-java-examples#492 for a demonstration

(Note, requires building the javaagent locally, including this PR from opentelemetry-java)

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

long-term, would we bridge in the other direction? i.e. replace ConfigProperties usage with StructuredConfigProperties everywhere in the Java agent, and have a bridge from ConfigProperties -> StructuredConfigProperties?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That's certainly an option. I think we ought to talk through options and repercussions as a group.

One thing to note is that the access patterns for StructuredConfigProperties and ConfigProperties are quite different.

For example, accessing otel.instrumentation.common.default-enabled is something like:

  • ConfigProperties: config.getBoolean("otel.instrumentation.common.default-enabled")
  • StructuredConfigProperties (assuming you're starting at the .instrumentation.java node) config.getStructured("config").getBoolean("default-enabled")
    • needs additional null handling each time you walk down a node
    • diverges further for each level of nesting

Also need to remember that there's now a specification for an instrumentation config API (and java PR here open-telemetry/opentelemetry-java#6549). The java agent works around the historic lack of availability of such a thing, and we'll need to think about how / when / (if?) to cut over. There's lots of instrumentation that reads config - is it feasible for a one time cutover to the new APIs, or do we need to plan for some sort of iterative approach?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

needs additional null handling each time you walk down a node

what do you think of optional return types that support subsequent accessors without null check?

diverges further for each level of nesting

I didn't follow what you mean?

we'll need to think about how / when / (if?) to cut over. There's lots of instrumentation that reads config - is it feasible for a one time cutover to the new APIs, or do we need to plan for some sort of iterative approach?

if we do some initial prototyping and like it, we could do a one-time cutover in the 3.0.0 release

Copy link
Member Author

@jack-berg jack-berg Sep 23, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

diverges further for each level of nesting
I didn't follow what you mean?

Just that the difference in syntax grows the more levels of nesting you have.

what do you think of optional return types that support subsequent accessors without null check?

Well we need some sort of syntactic sugar to not it make so painful, that's for sure.

The problem with returning Optional is that it creates a tri-state (optional-present, optional-empty, and null). While we can trust that the core implementation of StructuredConfigProperties never returns null, other user-provided interfaces might not be so well behaved.

Could also define a "get or default" overload of the form: getStructured(String name, StructuredConfigProperties defaultProperties)

If we combine this with a EMPTY_CONFIG constant, walking the tree becomes:

config.getStructured("common", EMPTY_CONFIG).getBoolean("default-enabled", true);

Not bad.

if we do some initial prototyping and like it, we could do a one-time cutover in the 3.0.0 release

That's true, but I doubt that the instrumentation config API will be stable in time for 3.0.0 which I assume is happening at the end of the year. So then we'll need to consider whether we're comfortable depending on an experimental API, or should wait till 4.0.0 for a cutover.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see these as feasible variants

  1. Optional:
config.getStructured("common")
  .flatMap(c -> c.getStructured("foo")
  .flatMap(c -> c.getBoolean("default-enabled", true));
  1. getStructured never returns null (but maybe has an isPresent method) - I like that one
config.getStructured("common").getStructured("foo").getBoolean("default-enabled", true);

It's true that custom implementations may still return null (for any of the above) - but they could also throw exceptions.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

  1. Optional:

The SDK has no precedent for returning optional instead of null. I believe that was an intentional design decision long ago, and I'm not interested in breaking symmetry.

  1. getStructured never returns null (but maybe has an isPresent method) - I like that one

Users need a way to tell determine the when a key was present and empty vs. null, so never returning null (or some equivalent) is not an option. We could add isPresent, but that's the same as calling getString("foo") != null so no improvement in user calling ergonomics.

PR here to add getStructured default method, which I think has good ergonomics and is symmetric with existing patterns in opentelemetry-java: open-telemetry/opentelemetry-java#6759

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

So it would look like this for a real example - looks too heavy IMO:

structuredConfigProps
                .getStructured("otel", empty())
                .getStructured("instrumentation", empty())
                .getStructured("http", empty())
                .getStructured("client", empty())
                .getMap("capture-request-headers"))

Copy link
Member Author

@jack-berg jack-berg Oct 14, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There's abstractions in place to read common properties so that only a few places in code have to do deep tree walking. And no code will need to read the .getStructured("otel", empty()).getStructured("instrumentation", empty()) steps since those happen automatically.

* string_key: value
* </pre>
*/
final class StructuredConfigPropertiesBridge implements ConfigProperties {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

would make sense to instrumentation-api-incubator so that in can be used in extensions

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I wasn't sure where to include this. @open-telemetry/java-instrumentation-maintainers is this or instrumentation-api-incubator the right thing for this type of thing?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

javaagent modules can't be used outside the agent - please correct me if I'm wrong @laurit

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

the current location seems good

would make sense to instrumentation-api-incubator so that in can be used in extensions

@zeitlinger maybe what you really want is open-telemetry/opentelemetry-java#6549?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I want to use the bridge in the starter at some point here:

But maybe that's a next step.

import java.util.Map;
import org.junit.jupiter.api.Test;

class StructuredConfigPropertiesBridgeTest {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see these as feasible variants

  1. Optional:
config.getStructured("common")
  .flatMap(c -> c.getStructured("foo")
  .flatMap(c -> c.getBoolean("default-enabled", true));
  1. getStructured never returns null (but maybe has an isPresent method) - I like that one
config.getStructured("common").getStructured("foo").getBoolean("default-enabled", true);

It's true that custom implementations may still return null (for any of the above) - but they could also throw exceptions.

Copy link
Member

@trask trask left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

lgtm

@jack-berg jack-berg marked this pull request as ready for review October 14, 2024 14:29
@jack-berg jack-berg requested a review from a team as a code owner October 14, 2024 14:29
Comment on lines +92 to +100
@Nullable
@Override
public Duration getDuration(String propertyName) {
Long millis = getPropertyValue(propertyName, StructuredConfigProperties::getLong);
if (millis == null) {
return null;
}
return Duration.ofMillis(millis);
}
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

is this intentionally scoping down and not handling durations with units like DefaultConfigProperties?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes. All durations in the declarative config schema are integer types assumed to be ms. There is no ability to specify the 1s, 1h, 1ms at this time.

@trask trask merged commit 780cdf4 into open-telemetry:main Oct 16, 2024
56 checks passed
@trask trask changed the title Support declarative configuration Support experimental declarative configuration Oct 18, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
test native This label can be applied to PRs to trigger them to run native tests
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants