query-orchestration/internal/test/config.go

// Package test provides configuration utilities for test infrastructure.
// This file centralizes timeout configuration for CI/CD stability.
package test

import (
	"fmt"
	"log/slog"
	"net/http"
	"os"
	"time"
)

// GetHTTPTimeout returns the HTTP client timeout for API calls.
// Default is 30 seconds, configurable via E2E_HTTP_TIMEOUT environment variable.
// In CI environments, this timeout ensures that stalled HTTP calls don't block
// indefinitely, which was a root cause of test hangs.
func GetHTTPTimeout() time.Duration {
	if t := os.Getenv("E2E_HTTP_TIMEOUT"); t != "" {
		if d, err := time.ParseDuration(t); err == nil {
			return d
		}
	}
	return 30 * time.Second
}

// GetPollingTimeout returns the timeout for polling operations like waiting
// for client status or mock endpoint verification.
// Default is 60 seconds locally, 180 seconds in CI (detected via CI or
// BITBUCKET_PIPELINE_UUID environment variables).
// Configurable via E2E_WAIT_TIMEOUT environment variable.
func GetPollingTimeout() time.Duration {
	if t := os.Getenv("E2E_WAIT_TIMEOUT"); t != "" {
		if d, err := time.ParseDuration(t); err == nil {
			return d
		}
	}
	if isCI() {
		return 180 * time.Second // 3 minutes in CI
	}
	return 60 * time.Second // 1 minute locally
}

// GetPollInterval returns the interval between polling attempts.
// Default is 500ms, configurable via E2E_POLL_INTERVAL environment variable.
func GetPollInterval() time.Duration {
	if t := os.Getenv("E2E_POLL_INTERVAL"); t != "" {
		if d, err := time.ParseDuration(t); err == nil {
			return d
		}
	}
	return 500 * time.Millisecond
}

// GetHTTPClient returns an HTTP client with the configured timeout.
// This should be used instead of &http.Client{} to ensure all HTTP operations
// have bounded timeouts and won't hang indefinitely.
func GetHTTPClient() *http.Client {
	return &http.Client{
		Timeout: GetHTTPTimeout(),
	}
}

// isCI returns true if running in a CI environment.
// Detects CI via the CI or BITBUCKET_PIPELINE_UUID environment variables.
func isCI() bool {
	return os.Getenv("CI") != "" || os.Getenv("BITBUCKET_PIPELINE_UUID") != ""
}

// WaitForExternalReadiness waits for a service to be reachable from outside Docker.
// This addresses the race condition where a container's internal health check passes
// but the external port mapping isn't yet ready. This is particularly important in
// Docker-in-Docker (DinD) environments like Bitbucket Pipelines.
//
// Parameters:
//   - uri: The base URI of the service (e.g., "http://localhost:32768")
//   - healthPath: The health check endpoint path (e.g., "/health" or "/liveness/probe")
//   - timeout: Maximum time to wait for the service to become reachable
//
// Returns an error if the service is not reachable within the timeout period.
func WaitForExternalReadiness(uri string, healthPath string, timeout time.Duration) error {
	client := &http.Client{Timeout: 2 * time.Second}
	deadline := time.Now().Add(timeout)

	fullURL := fmt.Sprintf("%s%s", uri, healthPath)
	slog.Info("Waiting for external readiness", "url", fullURL, "timeout", timeout)

	var lastErr error
	attempts := 0
	for time.Now().Before(deadline) {
		attempts++
		resp, err := client.Get(fullURL)
		if err == nil && resp.StatusCode == http.StatusOK {
			resp.Body.Close()
			slog.Info("Service is externally reachable", "url", fullURL, "attempts", attempts)
			return nil
		}
		if err != nil {
			lastErr = err
		} else {
			lastErr = fmt.Errorf("unexpected status code: %d", resp.StatusCode)
			resp.Body.Close()
		}
		time.Sleep(500 * time.Millisecond)
	}

	return fmt.Errorf("service not reachable from test client after %v (%d attempts): %w", timeout, attempts, lastErr)
}

// WaitForAPIRoutes waits for API routes to be fully registered.
// This addresses a race condition where the health endpoint returns 200 but other
// API routes haven't been registered yet, causing "404 page not found" errors.
// This is particularly important in slow CI environments where route registration
// may take longer after the health endpoint becomes available.
//
// Parameters:
//   - uri: The base URI of the service (e.g., "http://localhost:32768")
//   - timeout: Maximum time to wait for routes to be registered
//
// Returns an error if routes are not available within the timeout period.
func WaitForAPIRoutes(uri string, timeout time.Duration) error {
	client := &http.Client{Timeout: 2 * time.Second}
	deadline := time.Now().Add(timeout)

	// Check /clients endpoint - should return a response when routes are ready
	// A "404 page not found" means routes aren't registered yet
	// Note: /query endpoint was removed as part of query functionality removal
	routeURL := fmt.Sprintf("%s/clients", uri)
	slog.Info("Waiting for API routes to be registered", "url", routeURL, "timeout", timeout)

	var lastErr error
	attempts := 0
	for time.Now().Before(deadline) {
		attempts++
		resp, err := client.Get(routeURL)
		if err == nil {
			// Any response other than connection error means routes are registered
			// 200 = success, 401 = auth required (but route exists), etc.
			// Only "404 page not found" indicates routes not yet registered
			if resp.StatusCode != http.StatusNotFound {
				resp.Body.Close()
				slog.Info("API routes are registered", "url", routeURL, "attempts", attempts, "status", resp.StatusCode)
				return nil
			}
			lastErr = fmt.Errorf("routes not registered yet: %d", resp.StatusCode)
			resp.Body.Close()
		} else {
			lastErr = err
		}
		time.Sleep(500 * time.Millisecond)
	}

	return fmt.Errorf("API routes not registered after %v (%d attempts): %w", timeout, attempts, lastErr)
}