feat(mongodb): add OpenTelemetry tracing and metrics

[prathamesh-sonpatki]

Summary

• Adds MongoDB instrumentation using event.CommandMonitor from mongo-driver v1, replacing the deprecated otelmongo contrib package
• Provides NewClient(Config) and Instrument(*options.ClientOptions) APIs consistent with existing Redis/Kafka integration patterns
• Includes OTel semantic convention attributes, operation metrics (count, errors, duration histogram), command filtering, and monitor chaining

Details

Tracing: Spans are created per MongoDB command with names following OTel DB semconv ("insert users", "find orders"). Housekeeping commands (hello, isMaster, ping, saslStart, etc.) are filtered out. Spans carry db.system, db.name, db.operation, db.mongodb.collection, server.address, and server.port attributes.

Metrics: Three instruments — db.mongodb.operations (counter), db.mongodb.errors (counter), db.mongodb.operation.duration (histogram in ms).

Context propagation: Spans correctly parent under application spans via the driver's context forwarding.

Monitor chaining: If the caller has already set a CommandMonitor on their options, both monitors are called.

Test plan

• TestMongoDB_InsertFind — verifies insert/find spans with correct attributes
• TestMongoDB_UpdateDelete — verifies update/delete span attributes
• TestMongoDB_ContextPropagation — verifies child spans share parent trace ID
• TestMongoDB_Aggregate — verifies aggregation pipeline spans
• TestMongoDB_SkippedCommands — verifies housekeeping commands produce no spans
• TestMongoDB_DuplicateKeyStillTraced — verifies both insert spans exist even when one returns a duplicate key error
• TestMongoDB_InstrumentAPI — verifies Instrument() API path produces traced spans

All 7 integration tests pass against a mongo:7 testcontainer.

Resolves: FDE-48

🤖 Generated with Claude Code

[prathamesh-sonpatki]

Code review

Found 1 issue:

1. connectWithMonitor silently swallows instrumentation errors — the newMonitor error is logged but never returned to the caller. A caller of NewClient or Instrument receives a connected client with nil error even when metric instrument creation failed, giving no programmatic indication that instrumentation is degraded. The Redis integration (integrations/redis/redis.go) returns both the client and the instrumentation error, letting callers decide how to handle it. The MongoDB integration should follow the same pattern.

go-agent/integrations/mongodb/mongodb.go

Lines 171 to 193 in 9b058b7

	func connectWithMonitor(opts *options.ClientOptions, baseAttrs []attribute.KeyValue) (*mongo.Client, error) {
	m, err := newMonitor(baseAttrs)
	if err != nil {
	log.Printf("[Last9 Agent] Warning: partial MongoDB monitor setup: %v", err)
	}
	// If monitor creation failed entirely, connect without instrumentation.
	if m != nil {
	cmdMonitor := m.commandMonitor()
	// Chain with any existing monitor the caller may have set.
	if opts.Monitor != nil {
	cmdMonitor = chainMonitors(opts.Monitor, cmdMonitor)
	}
	opts.SetMonitor(cmdMonitor)
	}
	client, err := mongo.Connect(context.Background(), opts)
	if err != nil {
	return nil, fmt.Errorf("mongodb: failed to connect: %w", err)
	}
	return client, nil
	}

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}

[ashishl9]

LGTM!

Excellent MongoDB instrumentation implementation:

• Replaces deprecated otelmongo with native CommandMonitor approach
• Complete observability: spans, metrics (operations, errors, duration), proper OTel semconv
• Smart housekeeping command filtering reduces noise
• Proper context propagation with correct span nesting
• Flexible APIs: NewClient() and Instrument() patterns
• Monitor chaining respects existing configurations
• 7 comprehensive integration tests covering CRUD, aggregation, errors

Well-executed following established Redis/Kafka integration patterns.

[ashishl9]

@prathamesh-sonpatki PR #16 has been merged. This PR needs to be updated with the latest main branch. Please update the branch and I'll merge it.