Double the timeout.

README.md

@@ -24,9 +24,11 @@ Therefore, I present you the simple client to which you can copy & paste your gr
 ## Features
 
 * Executing GraphQL queries as they are, without types declaration
-* HTTP2 support!
+* HTTP/2 support for HTTPS endpoints (automatic)
+* HTTP/1.1 for HTTP endpoints (compatible with all servers)
+* Automatic retry with configurable error patterns
 * Support for additional headers
-* Query cache
+* Query cache with compression
 
 ## Usage example
 
@@ -37,15 +39,113 @@ Therefore, I present you the simple client to which you can copy & paste your gr
 * `GRAPHQL_CACHE_TTL` -  Cache TTL in seconds for SELECT type of queries. Default: `5`
 * `GRAPHQL_OUTPUT` - Output format. Default: `string`, available: `byte`, `string`, `mapstring`
 * `LOG_LEVEL` - Logging level. Default: `info` available: `debug`, `info`, `warn`, `error`
-* `GRAPHQL_RETRIES_ENABLE` - Should retries be enabled? Default: `true`
-* `GRAPHQL_RETRIES_NUMBER` - Number of retries: Default: `1`
+* `GRAPHQL_RETRIES_ENABLE` - Should retries be enabled? Default: `false`
+* `GRAPHQL_RETRIES_NUMBER` - Number of retries: Default: `3`
 * `GRAPHQL_RETRIES_DELAY` - Delay in retries in milliseconds. Default: `250`
+* `GRAPHQL_RETRIES_PATTERNS` - Comma-separated error patterns that trigger retries. Default: `postgres,connection,timeout,transaction,could not,temporarily unavailable,deadlock`
+* `GRAPHQL_POOL_WARMUP_ENABLED` - Enable intelligent connection pool warmup. Default: `false`
+* `GRAPHQL_POOL_SIZE` - Number of connections to pre-create and maintain. Default: `5`
+* `GRAPHQL_POOL_WARMUP_QUERY` - Query used for warming up connections. Default: `query{__typename}`
+* `GRAPHQL_POOL_HEALTH_INTERVAL` - Interval in seconds for pool health checks. Default: `30`
 
 ### Modifiers on the fly
 
 * `gql.SetEndpoint('your-endpoint-url')` - modifies endpoint, without the need to set the environment variable
 * `gql.SetOutput('byte')` - modifies output format, without the need to set the environment variable
 
+### Retries
+
+The library supports automatic retries for both HTTP-level errors and GraphQL errors that match configurable patterns.
+
+**Retry behavior:**
+- **HTTP errors** (status codes outside 200-299): Always retried if retries are enabled
+- **GraphQL errors**: Only retried if the error message contains one of the configured patterns
+- **Non-matching errors**: Fail immediately without retrying (e.g., validation errors, field not found)
+
+**Configuring retryable patterns:**
+
+By default, errors containing these patterns will trigger retries:
+- `postgres` - Database connection/transaction errors
+- `connection` - Network connection issues
+- `timeout` - Request timeout errors
+- `transaction` - Database transaction conflicts
+- `could not` - Generic transient failures
+- `temporarily unavailable` - Service unavailability
+- `deadlock` - Database deadlocks
+
+You can customize these patterns using the `GRAPHQL_RETRIES_PATTERNS` environment variable:
+
+```bash
+export GRAPHQL_RETRIES_PATTERNS="postgres,connection,timeout,custom-error-pattern"
+```
+
+**Example:** To handle flaky Hasura database connections, enable retries:
+
+```bash
+export GRAPHQL_RETRIES_ENABLE=true
+export GRAPHQL_RETRIES_NUMBER=3
+export GRAPHQL_RETRIES_DELAY=250
+```
+
+This will retry GraphQL queries up to 3 times when postgres transaction errors occur, with exponential backoff starting at 250ms.
+
+### Intelligent Connection Pooling
+
+The library supports intelligent connection pooling with pre-warmed connections and automatic health monitoring. This significantly improves performance during traffic spikes by keeping hot standby connections ready.
+
+**How it works:**
+- **Pre-warming**: Connections are created and warmed up during initialization
+- **Health monitoring**: Background process periodically checks connection health
+- **Auto-refresh**: Automatically replaces stale or broken connections
+- **Hot standbys**: Ready connections eliminate cold-start latency
+
+**Configuration:**
+
+Enable connection pool warmup for production environments with variable traffic:
+
+```bash
+export GRAPHQL_POOL_WARMUP_ENABLED=true
+export GRAPHQL_POOL_SIZE=5                    # Number of hot standby connections
+export GRAPHQL_POOL_HEALTH_INTERVAL=30        # Health check every 30 seconds
+export GRAPHQL_POOL_WARMUP_QUERY="query{__typename}"  # Lightweight health check query
+```
+
+**Benefits:**
+- ⚡ **Instant response** - Connections already established during traffic spikes
+- 🔄 **Auto-healing** - Detects and replaces broken connections automatically
+- 📊 **Predictable performance** - Eliminates connection establishment latency
+- 🛡️ **Resilience** - Maintains pool health even during network instability
+
+**Recommended settings by use case:**
+
+```bash
+# High-traffic API (e.g., production web app)
+GRAPHQL_POOL_WARMUP_ENABLED=true
+GRAPHQL_POOL_SIZE=10
+GRAPHQL_POOL_HEALTH_INTERVAL=15
+
+# Medium-traffic service (e.g., background jobs)
+GRAPHQL_POOL_WARMUP_ENABLED=true
+GRAPHQL_POOL_SIZE=5
+GRAPHQL_POOL_HEALTH_INTERVAL=30
+
+# Low-traffic/development
+GRAPHQL_POOL_WARMUP_ENABLED=false  # Default: on-demand connections
+```
+
+**Graceful shutdown:**
+
+When shutting down your application, stop the pool monitor gracefully:
+
+```go
+func main() {
+    gql := graphql.NewConnection()
+    defer gql.StopPoolMonitor()  // Ensure clean shutdown
+
+    // Your application logic
+}
+```
+
 ### Cache
 
 You have two options to enable the cache:

execute_query.go

@@ -3,11 +3,11 @@ package gql
 import (
 	"bytes"
 	"compress/gzip"
-	"crypto/tls"
 	"errors"
 	"fmt"
 	"io"
 	"net/http"
+	"strings"
 	"time"
 	"unicode/utf8"
 
@@ -32,23 +32,49 @@ func max(a, b int) int {
 	return b
 }
 
-var (
-	// Shared HTTP transport with optimized settings
-	defaultTransport = &http.Transport{
-		MaxIdleConns:        100,
-		MaxIdleConnsPerHost: 100,
-		IdleConnTimeout:     90 * time.Second,
-		DisableCompression:  true, // Disable automatic request compression to prevent "trailing garbage" errors
-		ForceAttemptHTTP2:   true,
-		TLSClientConfig:     &tls.Config{InsecureSkipVerify: true}, // Skip TLS verification for test environments
+// isRetryableError checks if GraphQL errors contain patterns that indicate a transient failure
+// that should be retried (e.g., database connection issues, timeouts, transaction conflicts)
+func isRetryableError(errors []struct {
+	Message interface{} `json:"message"`
+}, patterns []string) bool {
+	if len(errors) == 0 || len(patterns) == 0 {
+		return false
 	}
 
-	// Shared HTTP client with timeouts
-	defaultClient = &http.Client{
-		Transport: defaultTransport,
-		Timeout:   30 * time.Second,
+	for _, err := range errors {
+		// Try to extract error message as string
+		var msgStr string
+		switch msg := err.Message.(type) {
+		case string:
+			msgStr = msg
+		case map[string]interface{}:
+			// Handle nested error structures (e.g., {"message": "...", "extensions": {...}})
+			if nestedMsg, ok := msg["message"].(string); ok {
+				msgStr = nestedMsg
+			}
+		default:
+			// If we can't convert to string, try fmt.Sprint as fallback
+			msgStr = fmt.Sprint(msg)
+		}
+
+		if msgStr == "" {
+			continue
+		}
+
+		// Check if error message contains any of the configured retry patterns (case-insensitive)
+		msgLower := strings.ToLower(msgStr)
+		for _, pattern := range patterns {
+			if strings.Contains(msgLower, strings.ToLower(pattern)) {
+				return true
+			}
+		}
 	}
-)
+
+	return false
+}
+
+// Note: defaultClient removed - all clients should be created via NewConnection()
+// which properly configures HTTP/2 transport with correct settings
 
 func (qe *QueryExecutor) executeQuery() ([]byte, error) {
 	// Reuse buffer from pool to avoid allocations
@@ -131,12 +157,16 @@ func (qe *QueryExecutor) executeQuery() ([]byte, error) {
 				},
 			})
 
-			// Use default client if custom client is not set
-			client := qe.client
-			if client == nil {
-				client = defaultClient
+			// Client should always be set via NewConnection(), but check for safety
+			if qe.client == nil {
+				qe.Logger.Error(&libpack_logger.LogMessage{
+					Message: "HTTP client not initialized - use NewConnection() to create client",
+					Pairs:   nil,
+				})
+				return fmt.Errorf("HTTP client not initialized")
 			}
-			httpResponse, err := client.Do(httpRequest)
+
+			httpResponse, err := qe.client.Do(httpRequest)
 			if err != nil {
 				return err
 			}
@@ -331,6 +361,38 @@ func (qe *QueryExecutor) executeQuery() ([]byte, error) {
 				return fmt.Errorf("error unmarshalling HTTP response: %w", err)
 			}
 
+			// Check for GraphQL errors in the response
+			if len(queryResult.Errors) > 0 {
+				// Determine if this is a retryable error based on configured patterns
+				if isRetryableError(queryResult.Errors, qe.retries_patterns) {
+					// Log as warning since we'll retry
+					qe.Logger.Warning(&libpack_logger.LogMessage{
+						Message: "Retryable GraphQL error detected",
+						Pairs:   map[string]interface{}{"errors": queryResult.Errors},
+					})
+					// Return error to trigger retry mechanism
+					return fmt.Errorf("retryable error executing query: %s", queryResult.Errors)
+				}
+
+				// Non-retryable error - log and fail immediately without retrying
+				qe.Logger.Error(&libpack_logger.LogMessage{
+					Message: "Non-retryable GraphQL error",
+					Pairs:   map[string]interface{}{"errors": queryResult.Errors},
+				})
+				// Use retry.Unrecoverable to skip remaining retry attempts
+				return retry.Unrecoverable(fmt.Errorf("error executing query: %s", queryResult.Errors))
+			}
+
+			// Check for null data in the response
+			if queryResult.Data == nil {
+				qe.Logger.Error(&libpack_logger.LogMessage{
+					Message: "GraphQL query returned no data",
+					Pairs:   map[string]interface{}{"error": "data field is null"},
+				})
+				// Null data is typically not retryable (indicates query structure issue)
+				return retry.Unrecoverable(errors.New("error executing query: no data"))
+			}
+
 			return nil
 		},
 		retry.OnRetry(func(n uint, err error) {
@@ -346,33 +408,18 @@ func (qe *QueryExecutor) executeQuery() ([]byte, error) {
 		retry.LastErrorOnly(true),
 	)
 	if err != nil {
-		qe.Logger.Debug(&libpack_logger.LogMessage{
-			Message: "Error executing HTTP request",
+		qe.Logger.Error(&libpack_logger.LogMessage{
+			Message: "Query execution failed after retries",
 			Pairs:   map[string]interface{}{"error": err.Error()},
 		})
 		return nil, err
 	}
 
-	if len(queryResult.Errors) > 0 {
-		qe.Logger.Debug(&libpack_logger.LogMessage{
-			Message: "Error executing query",
-			Pairs:   map[string]interface{}{"error": queryResult.Errors},
-		})
-		return nil, fmt.Errorf("error executing query: %s", queryResult.Errors)
-	}
-
-	if queryResult.Data == nil {
-		qe.Logger.Debug(&libpack_logger.LogMessage{
-			Message: "Error executing query",
-			Pairs:   map[string]interface{}{"error": "no data"},
-		})
-		return nil, errors.New("error executing query: no data")
-	}
-
+	// At this point, we have successfully executed the query and validated the response
 	// Marshal queryResult.Data
 	jsonData, err := json.Marshal(queryResult.Data)
 	if err != nil {
-		qe.Logger.Debug(&libpack_logger.LogMessage{
+		qe.Logger.Error(&libpack_logger.LogMessage{
 			Message: "Error marshalling query result",
 			Pairs:   map[string]interface{}{"error": err.Error(), "data": queryResult.Data},
 		})

execute_query_test.go

@@ -185,7 +185,7 @@ func (suite *Tests) TestQueryExecutor_executeQuery() {
 		assert.Equal(3, attempts) // Should have retried twice before succeeding
 	})
 
-	suite.T().Run("should use default client when none provided", func(t *testing.T) {
+	suite.T().Run("should return error when client not initialized", func(t *testing.T) {
 		client := CreateTestClient()
 
 		qe := &QueryExecutor{
@@ -196,12 +196,13 @@ func (suite *Tests) TestQueryExecutor_executeQuery() {
 			Retries:    false,
 		}
 		qe.endpoint = mockServer.URL
-		qe.client = nil // No client provided
+		qe.client = nil // No client provided (should always use NewConnection())
 		qe.cache = client.cache
 
 		result, err := qe.executeQuery()
-		assert.NoError(err)
-		assert.NotNil(result)
+		assert.Error(err)
+		assert.Nil(result)
+		assert.Contains(err.Error(), "HTTP client not initialized")
 	})
 
 	suite.T().Run("should set default Content-Type header", func(t *testing.T) {

http.go

@@ -11,60 +11,72 @@ import (
 )
 
 func (b *BaseClient) createHttpClient() (http_client *http.Client) {
-	// Optimized connection pool settings for better performance
-	httpTransport := &http.Transport{
-		MaxIdleConns:          100,              // Reduced from 512
-		MaxConnsPerHost:       50,               // Reduced from 512
-		MaxIdleConnsPerHost:   10,               // Reduced from 512
-		IdleConnTimeout:       30 * time.Second, // Increased for better reuse
-		ResponseHeaderTimeout: 10 * time.Second, // Reduced for faster timeouts
-		DisableKeepAlives:     false,
-		DisableCompression:    true, // Disable automatic request compression to prevent "trailing garbage" errors
-		WriteBufferSize:       4096, // Optimize buffer size
-		ReadBufferSize:        4096, // Optimize buffer size
-	}
-
 	if strings.HasPrefix(b.endpoint, "http://") {
+		// Use HTTP/1.1 for http:// endpoints
+		// Note: h2c (HTTP/2 Cleartext) is rarely supported by servers
+		httpTransport := &http.Transport{
+			MaxIdleConns:          100,
+			MaxConnsPerHost:       50,
+			MaxIdleConnsPerHost:   10,
+			IdleConnTimeout:       60 * time.Second,
+			ResponseHeaderTimeout: 20 * time.Second,
+			DisableKeepAlives:     false,
+			DisableCompression:    true, // Disable automatic compression to prevent trailing garbage errors
+			WriteBufferSize:       4096,
+			ReadBufferSize:        4096,
+			// Additional settings for connection health
+			ExpectContinueTimeout: 2 * time.Second,  // Timeout for 100-Continue
+			TLSHandshakeTimeout:   20 * time.Second, // Timeout for TLS handshake
+		}
+
 		http_client = &http.Client{
-			Timeout:   30 * time.Second, // Increased for better reliability
+			Timeout:   60 * time.Second,
 			Transport: httpTransport,
 			CheckRedirect: func(req *http.Request, via []*http.Request) error {
 				return http.ErrUseLastResponse
 			},
 		}
+
 		b.Logger.Debug(&libpack_logger.LogMessage{
-			Message: "Using HTTP/1.1 over http",
+			Message: "Using HTTP/1.1 for http:// endpoint",
 			Pairs:   nil,
 		})
 	} else if strings.HasPrefix(b.endpoint, "https://") {
-		tlsClientConfig := &tls.Config{}
-		if strings.HasPrefix(b.endpoint, "https://") {
-			tlsClientConfig.InsecureSkipVerify = true
+		// Use HTTP/2 for https:// endpoints
+		tlsClientConfig := &tls.Config{
+			InsecureSkipVerify: true, // TODO: Make this configurable via environment variable
 		}
+
 		http2Transport := &http2.Transport{
 			AllowHTTP:          true,
 			TLSClientConfig:    tlsClientConfig,
-			ReadIdleTimeout:    30 * time.Second, // Increased for better reuse
-			PingTimeout:        10 * time.Second, // Reduced for faster detection
-			WriteByteTimeout:   10 * time.Second, // Add write timeout
-			DisableCompression: true,             // Disable automatic request compression to prevent "trailing garbage" errors
+			ReadIdleTimeout:    60 * time.Second, // Close idle connections after 60s
+			PingTimeout:        20 * time.Second, // Detect dead connections with PING
+			WriteByteTimeout:   20 * time.Second, // Timeout for write operations
+			DisableCompression: true,             // Disable automatic compression to prevent trailing garbage errors
+			// StrictMaxConcurrentStreams forces the HTTP/2 connection to wait for SETTINGS frame
+			// This helps with connection health detection
+			StrictMaxConcurrentStreams: true,
 		}
+
 		http_client = &http.Client{
-			Timeout:   30 * time.Second, // Increased for better reliability
+			Timeout:   60 * time.Second,
 			Transport: http2Transport,
 			CheckRedirect: func(req *http.Request, via []*http.Request) error {
 				return http.ErrUseLastResponse
 			},
 		}
+
 		b.Logger.Debug(&libpack_logger.LogMessage{
-			Message: "Using HTTP/2 over https",
+			Message: "Using HTTP/2 for https:// endpoint",
 			Pairs:   nil,
 		})
 	} else {
 		b.Logger.Critical(&libpack_logger.LogMessage{
-			Message: "Invalid endpoint - neither http or https",
+			Message: "Invalid endpoint - must start with http:// or https://",
 		})
 		return nil
 	}
+
 	return http_client
 }

http_test.go

@@ -8,6 +8,8 @@ import (
 	"golang.org/x/net/http2"
 )
 
+// Note: http.Transport import is kept for http.ErrUseLastResponse usage in redirect tests
+
 func (suite *Tests) TestBaseClient_createHttpClient() {
 	suite.T().Run("should create HTTP/1.1 client for http endpoints", func(t *testing.T) {
 		client := NewConnection()
@@ -15,17 +17,17 @@ func (suite *Tests) TestBaseClient_createHttpClient() {
 
 		httpClient := client.createHttpClient()
 		assert.NotNil(httpClient)
-		assert.Equal(30*time.Second, httpClient.Timeout)
+		assert.Equal(60*time.Second, httpClient.Timeout)
 		assert.NotNil(httpClient.Transport)
 
-		// Check that it's using regular HTTP transport
+		// Check that it's using HTTP/1.1 transport (h2c is rarely supported by servers)
 		transport, ok := httpClient.Transport.(*http.Transport)
 		assert.True(ok)
 		assert.Equal(100, transport.MaxIdleConns)
 		assert.Equal(50, transport.MaxConnsPerHost)
 		assert.Equal(10, transport.MaxIdleConnsPerHost)
-		assert.Equal(30*time.Second, transport.IdleConnTimeout)
-		assert.Equal(10*time.Second, transport.ResponseHeaderTimeout)
+		assert.Equal(60*time.Second, transport.IdleConnTimeout)
+		assert.Equal(20*time.Second, transport.ResponseHeaderTimeout)
 		assert.False(transport.DisableKeepAlives)
 		assert.True(transport.DisableCompression) // Request compression disabled to prevent "trailing garbage" errors
 		assert.Equal(4096, transport.WriteBufferSize)
@@ -38,7 +40,7 @@ func (suite *Tests) TestBaseClient_createHttpClient() {
 
 		httpClient := client.createHttpClient()
 		assert.NotNil(httpClient)
-		assert.Equal(30*time.Second, httpClient.Timeout)
+		assert.Equal(60*time.Second, httpClient.Timeout)
 		assert.NotNil(httpClient.Transport)
 
 		// Check that it's using HTTP/2 transport
@@ -47,9 +49,9 @@ func (suite *Tests) TestBaseClient_createHttpClient() {
 		assert.True(transport.AllowHTTP)
 		assert.NotNil(transport.TLSClientConfig)
 		assert.True(transport.TLSClientConfig.InsecureSkipVerify)
-		assert.Equal(30*time.Second, transport.ReadIdleTimeout)
-		assert.Equal(10*time.Second, transport.PingTimeout)
-		assert.Equal(10*time.Second, transport.WriteByteTimeout)
+		assert.Equal(60*time.Second, transport.ReadIdleTimeout)
+		assert.Equal(20*time.Second, transport.PingTimeout)
+		assert.Equal(20*time.Second, transport.WriteByteTimeout)
 	})
 
 	suite.T().Run("should handle invalid endpoints", func(t *testing.T) {
@@ -92,19 +94,19 @@ func (suite *Tests) TestBaseClient_createHttpClient() {
 }
 
 func (suite *Tests) TestBaseClient_createHttpClient_transportSettings() {
-	suite.T().Run("should configure HTTP transport correctly", func(t *testing.T) {
+	suite.T().Run("should configure HTTP/1.1 transport correctly for http://", func(t *testing.T) {
 		client := NewConnection()
 		client.SetEndpoint("http://example.com/graphql")
 
 		httpClient := client.createHttpClient()
 		transport := httpClient.Transport.(*http.Transport)
 
-		// Verify all transport settings
+		// Verify all HTTP/1.1 transport settings
 		assert.Equal(100, transport.MaxIdleConns)
 		assert.Equal(50, transport.MaxConnsPerHost)
 		assert.Equal(10, transport.MaxIdleConnsPerHost)
-		assert.Equal(30*time.Second, transport.IdleConnTimeout)
-		assert.Equal(10*time.Second, transport.ResponseHeaderTimeout)
+		assert.Equal(60*time.Second, transport.IdleConnTimeout)
+		assert.Equal(20*time.Second, transport.ResponseHeaderTimeout)
 		assert.False(transport.DisableKeepAlives)
 		assert.True(transport.DisableCompression) // Request compression disabled to prevent "trailing garbage" errors
 		assert.Equal(4096, transport.WriteBufferSize)
@@ -122,9 +124,9 @@ func (suite *Tests) TestBaseClient_createHttpClient_transportSettings() {
 		assert.True(transport.AllowHTTP)
 		assert.NotNil(transport.TLSClientConfig)
 		assert.True(transport.TLSClientConfig.InsecureSkipVerify)
-		assert.Equal(30*time.Second, transport.ReadIdleTimeout)
-		assert.Equal(10*time.Second, transport.PingTimeout)
-		assert.Equal(10*time.Second, transport.WriteByteTimeout)
+		assert.Equal(60*time.Second, transport.ReadIdleTimeout)
+		assert.Equal(20*time.Second, transport.PingTimeout)
+		assert.Equal(20*time.Second, transport.WriteByteTimeout)
 	})
 }
 

main.go

@@ -2,6 +2,7 @@ package gql
 
 import (
 	"net/http"
+	"strings"
 	"time"
 
 	"github.com/gookit/goutil/envutil"
@@ -9,6 +10,22 @@ import (
 	logging "github.com/lukaszraczylo/go-simple-graphql/logging"
 )
 
+// parseRetryPatterns parses a comma-separated string of retry patterns into a slice
+func parseRetryPatterns(patterns string) []string {
+	if patterns == "" {
+		return []string{}
+	}
+	parts := strings.Split(patterns, ",")
+	result := make([]string, 0, len(parts))
+	for _, p := range parts {
+		trimmed := strings.TrimSpace(p)
+		if trimmed != "" {
+			result = append(result, trimmed)
+		}
+	}
+	return result
+}
+
 func NewConnection() (b *BaseClient) {
 	// Read LOG_LEVEL environment variable
 	logLevelStr := envutil.Getenv("LOG_LEVEL", "info")
@@ -30,23 +47,39 @@ func NewConnection() (b *BaseClient) {
 	})
 
 	b = &BaseClient{
-		endpoint:       envutil.Getenv("GRAPHQL_ENDPOINT", "https://api.github.com/graphql"),
-		responseType:   envutil.Getenv("GRAPHQL_OUTPUT", "string"),
-		Logger:         logger,
-		cache:          cache.New(time.Duration(envutil.GetInt("GRAPHQL_CACHE_TTL", 5)) * time.Second),
-		cache_global:   envutil.GetBool("GRAPHQL_CACHE_ENABLED", false),
-		retries_enable: envutil.GetBool("GRAPHQL_RETRIES_ENABLE", false),
-		retries_delay:  time.Duration(envutil.GetInt("GRAPHQL_RETRIES_DELAY", 250) * int(time.Millisecond)),
-		retries_number: envutil.GetInt("GRAPHQL_RETRIES_NUMBER", 3),
-		minify_queries: envutil.GetBool("GRAPHQL_MINIFY_QUERIES", true), // Default: enabled for production efficiency
+		endpoint:             envutil.Getenv("GRAPHQL_ENDPOINT", "https://api.github.com/graphql"),
+		responseType:         envutil.Getenv("GRAPHQL_OUTPUT", "string"),
+		Logger:               logger,
+		cache:                cache.New(time.Duration(envutil.GetInt("GRAPHQL_CACHE_TTL", 5)) * time.Second),
+		cache_global:         envutil.GetBool("GRAPHQL_CACHE_ENABLED", false),
+		retries_enable:       envutil.GetBool("GRAPHQL_RETRIES_ENABLE", false),
+		retries_delay:        time.Duration(envutil.GetInt("GRAPHQL_RETRIES_DELAY", 250) * int(time.Millisecond)),
+		retries_number:       envutil.GetInt("GRAPHQL_RETRIES_NUMBER", 3),
+		retries_patterns:     parseRetryPatterns(envutil.Getenv("GRAPHQL_RETRIES_PATTERNS", "postgres,connection,timeout,transaction,could not,temporarily unavailable,deadlock")),
+		minify_queries:       envutil.GetBool("GRAPHQL_MINIFY_QUERIES", true), // Default: enabled for production efficiency
+		pool_warmup_enabled:  envutil.GetBool("GRAPHQL_POOL_WARMUP_ENABLED", false),
+		pool_size:            envutil.GetInt("GRAPHQL_POOL_SIZE", 5),
+		pool_warmup_query:    envutil.Getenv("GRAPHQL_POOL_WARMUP_QUERY", "query{__typename}"),
+		pool_health_interval: time.Duration(envutil.GetInt("GRAPHQL_POOL_HEALTH_INTERVAL", 30)) * time.Second,
+		pool_stop:            make(chan bool, 1),
 	}
 	b.client = b.createHttpClient()
 	b.Logger.Debug(&logging.LogMessage{
 		Message: "Created new GraphQL client connection",
 		Pairs: map[string]interface{}{
-			"values": b,
+			"endpoint":          b.endpoint,
+			"pool_warmup":       b.pool_warmup_enabled,
+			"pool_size":         b.pool_size,
+			"pool_health_check": b.pool_health_interval,
 		},
 	})
+
+	// Initialize connection pool if warmup is enabled
+	if b.pool_warmup_enabled {
+		b.warmupConnectionPool()
+		b.startPoolHealthMonitor()
+	}
+
 	return b
 }
 

pool.go

@@ -0,0 +1,250 @@
+package gql
+
+import (
+	"context"
+	"sync"
+	"time"
+
+	libpack_logger "github.com/lukaszraczylo/go-simple-graphql/logging"
+)
+
+// warmupConnectionPool pre-creates connections by executing warmup queries
+// This establishes connections in the pool before actual traffic arrives
+func (b *BaseClient) warmupConnectionPool() {
+	b.Logger.Info(&libpack_logger.LogMessage{
+		Message: "Warming up connection pool",
+		Pairs: map[string]interface{}{
+			"pool_size":    b.pool_size,
+			"warmup_query": b.pool_warmup_query,
+			"endpoint":     b.endpoint,
+		},
+	})
+
+	startTime := time.Now()
+	var wg sync.WaitGroup
+	successCount := 0
+	failCount := 0
+	var mu sync.Mutex
+
+	// Pre-create connections concurrently
+	for i := 0; i < b.pool_size; i++ {
+		wg.Add(1)
+		go func(index int) {
+			defer wg.Done()
+
+			// Execute warmup query to establish connection
+			_, err := b.Query(b.pool_warmup_query, nil, nil)
+
+			mu.Lock()
+			if err != nil {
+				failCount++
+				b.Logger.Warning(&libpack_logger.LogMessage{
+					Message: "Failed to warm up connection",
+					Pairs: map[string]interface{}{
+						"connection_index": index,
+						"error":            err.Error(),
+					},
+				})
+			} else {
+				successCount++
+				b.Logger.Debug(&libpack_logger.LogMessage{
+					Message: "Connection warmed up successfully",
+					Pairs: map[string]interface{}{
+						"connection_index": index,
+					},
+				})
+			}
+			mu.Unlock()
+		}(i)
+	}
+
+	wg.Wait()
+	duration := time.Since(startTime)
+
+	b.Logger.Info(&libpack_logger.LogMessage{
+		Message: "Connection pool warmup completed",
+		Pairs: map[string]interface{}{
+			"successful":  successCount,
+			"failed":      failCount,
+			"total":       b.pool_size,
+			"duration_ms": duration.Milliseconds(),
+		},
+	})
+}
+
+// startPoolHealthMonitor starts a background goroutine that periodically
+// checks and maintains pool health by sending health check queries
+func (b *BaseClient) startPoolHealthMonitor() {
+	b.Logger.Info(&libpack_logger.LogMessage{
+		Message: "Starting connection pool health monitor",
+		Pairs: map[string]interface{}{
+			"check_interval": b.pool_health_interval,
+			"pool_size":      b.pool_size,
+		},
+	})
+
+	go func() {
+		ticker := time.NewTicker(b.pool_health_interval)
+		defer ticker.Stop()
+
+		for {
+			select {
+			case <-ticker.C:
+				b.performPoolHealthCheck()
+			case <-b.pool_stop:
+				b.Logger.Info(&libpack_logger.LogMessage{
+					Message: "Stopping connection pool health monitor",
+					Pairs:   nil,
+				})
+				return
+			}
+		}
+	}()
+}
+
+// performPoolHealthCheck executes health check queries to maintain pool connections
+// This keeps connections alive and detects/replaces dead connections
+func (b *BaseClient) performPoolHealthCheck() {
+	b.Logger.Debug(&libpack_logger.LogMessage{
+		Message: "Performing pool health check",
+		Pairs: map[string]interface{}{
+			"pool_size": b.pool_size,
+		},
+	})
+
+	startTime := time.Now()
+	var wg sync.WaitGroup
+	healthyCount := 0
+	unhealthyCount := 0
+	var mu sync.Mutex
+
+	// Check a subset of connections to ensure pool health
+	// We check pool_size/2 to balance between thoroughness and overhead
+	checkCount := max(1, b.pool_size/2)
+
+	for i := 0; i < checkCount; i++ {
+		wg.Add(1)
+		go func(index int) {
+			defer wg.Done()
+
+			// Create a context with timeout for health check
+			ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+			defer cancel()
+
+			// Use a channel to handle timeout
+			done := make(chan error, 1)
+			go func() {
+				_, err := b.Query(b.pool_warmup_query, nil, nil)
+				done <- err
+			}()
+
+			select {
+			case err := <-done:
+				mu.Lock()
+				if err != nil {
+					unhealthyCount++
+					b.Logger.Warning(&libpack_logger.LogMessage{
+						Message: "Pool health check failed",
+						Pairs: map[string]interface{}{
+							"check_index": index,
+							"error":       err.Error(),
+						},
+					})
+				} else {
+					healthyCount++
+				}
+				mu.Unlock()
+			case <-ctx.Done():
+				mu.Lock()
+				unhealthyCount++
+				b.Logger.Warning(&libpack_logger.LogMessage{
+					Message: "Pool health check timeout",
+					Pairs: map[string]interface{}{
+						"check_index": index,
+					},
+				})
+				mu.Unlock()
+			}
+		}(i)
+	}
+
+	wg.Wait()
+	duration := time.Since(startTime)
+
+	b.Logger.Debug(&libpack_logger.LogMessage{
+		Message: "Pool health check completed",
+		Pairs: map[string]interface{}{
+			"healthy":     healthyCount,
+			"unhealthy":   unhealthyCount,
+			"checked":     checkCount,
+			"duration_ms": duration.Milliseconds(),
+		},
+	})
+
+	// If more than 50% of checks failed, trigger a pool refresh
+	if unhealthyCount > checkCount/2 {
+		b.Logger.Warning(&libpack_logger.LogMessage{
+			Message: "Pool health degraded, triggering refresh",
+			Pairs: map[string]interface{}{
+				"unhealthy_ratio": float64(unhealthyCount) / float64(checkCount),
+			},
+		})
+		b.refreshConnectionPool()
+	}
+}
+
+// refreshConnectionPool attempts to refresh the connection pool
+// by executing warmup queries to replace stale connections
+func (b *BaseClient) refreshConnectionPool() {
+	b.Logger.Info(&libpack_logger.LogMessage{
+		Message: "Refreshing connection pool",
+		Pairs: map[string]interface{}{
+			"pool_size": b.pool_size,
+		},
+	})
+
+	var wg sync.WaitGroup
+	refreshCount := max(1, b.pool_size/3) // Refresh 1/3 of pool
+
+	for i := 0; i < refreshCount; i++ {
+		wg.Add(1)
+		go func(index int) {
+			defer wg.Done()
+
+			_, err := b.Query(b.pool_warmup_query, nil, nil)
+			if err != nil {
+				b.Logger.Debug(&libpack_logger.LogMessage{
+					Message: "Pool refresh connection failed",
+					Pairs: map[string]interface{}{
+						"connection_index": index,
+						"error":            err.Error(),
+					},
+				})
+			} else {
+				b.Logger.Debug(&libpack_logger.LogMessage{
+					Message: "Pool connection refreshed",
+					Pairs: map[string]interface{}{
+						"connection_index": index,
+					},
+				})
+			}
+		}(i)
+	}
+
+	wg.Wait()
+}
+
+// StopPoolMonitor gracefully stops the pool health monitor
+// Call this when shutting down the application
+// Uses sync.Once to ensure it only executes once, preventing race conditions
+func (b *BaseClient) StopPoolMonitor() {
+	if b.pool_warmup_enabled {
+		b.pool_stop_once.Do(func() {
+			b.Logger.Info(&libpack_logger.LogMessage{
+				Message: "Stopping pool monitor",
+				Pairs:   nil,
+			})
+			b.pool_stop <- true
+		})
+	}
+}

pool_test.go

@@ -0,0 +1,135 @@
+package gql
+
+import (
+	"os"
+	"testing"
+	"time"
+)
+
+func (suite *Tests) TestPoolWarmup() {
+	suite.T().Run("should create client with pool warmup disabled by default", func(t *testing.T) {
+		// Save original env vars
+		originalPoolEnabled := os.Getenv("GRAPHQL_POOL_WARMUP_ENABLED")
+		originalEndpoint := os.Getenv("GRAPHQL_ENDPOINT")
+		defer func() {
+			os.Setenv("GRAPHQL_POOL_WARMUP_ENABLED", originalPoolEnabled)
+			os.Setenv("GRAPHQL_ENDPOINT", originalEndpoint)
+		}()
+
+		os.Setenv("GRAPHQL_POOL_WARMUP_ENABLED", "false")
+		os.Setenv("GRAPHQL_ENDPOINT", mockServer.URL)
+
+		client := NewConnection()
+		defer client.StopPoolMonitor()
+
+		assert.False(client.pool_warmup_enabled)
+		assert.Equal(5, client.pool_size) // Default size
+	})
+
+	suite.T().Run("should warmup pool when enabled", func(t *testing.T) {
+		originalPoolEnabled := os.Getenv("GRAPHQL_POOL_WARMUP_ENABLED")
+		originalPoolSize := os.Getenv("GRAPHQL_POOL_SIZE")
+		originalEndpoint := os.Getenv("GRAPHQL_ENDPOINT")
+		defer func() {
+			os.Setenv("GRAPHQL_POOL_WARMUP_ENABLED", originalPoolEnabled)
+			os.Setenv("GRAPHQL_POOL_SIZE", originalPoolSize)
+			os.Setenv("GRAPHQL_ENDPOINT", originalEndpoint)
+		}()
+
+		os.Setenv("GRAPHQL_POOL_WARMUP_ENABLED", "true")
+		os.Setenv("GRAPHQL_POOL_SIZE", "3")
+		os.Setenv("GRAPHQL_ENDPOINT", mockServer.URL)
+
+		startTime := time.Now()
+		client := NewConnection()
+		defer client.StopPoolMonitor()
+
+		duration := time.Since(startTime)
+
+		assert.True(client.pool_warmup_enabled)
+		assert.Equal(3, client.pool_size)
+		// Warmup should complete relatively quickly for 3 connections
+		assert.Less(duration.Seconds(), 5.0, "Pool warmup took too long")
+	})
+
+	suite.T().Run("should use custom warmup query", func(t *testing.T) {
+		originalPoolEnabled := os.Getenv("GRAPHQL_POOL_WARMUP_ENABLED")
+		originalQuery := os.Getenv("GRAPHQL_POOL_WARMUP_QUERY")
+		originalEndpoint := os.Getenv("GRAPHQL_ENDPOINT")
+		defer func() {
+			os.Setenv("GRAPHQL_POOL_WARMUP_ENABLED", originalPoolEnabled)
+			os.Setenv("GRAPHQL_POOL_WARMUP_QUERY", originalQuery)
+			os.Setenv("GRAPHQL_ENDPOINT", originalEndpoint)
+		}()
+
+		os.Setenv("GRAPHQL_POOL_WARMUP_ENABLED", "false") // Don't actually warmup, just test configuration
+		os.Setenv("GRAPHQL_POOL_WARMUP_QUERY", "query{__schema{queryType{name}}}")
+		os.Setenv("GRAPHQL_ENDPOINT", mockServer.URL)
+
+		client := NewConnection()
+		defer client.StopPoolMonitor()
+
+		assert.Equal("query{__schema{queryType{name}}}", client.pool_warmup_query)
+	})
+}
+
+func (suite *Tests) TestPoolHealthMonitor() {
+	suite.T().Run("should stop pool monitor gracefully", func(t *testing.T) {
+		client := CreateTestClient()
+		client.pool_warmup_enabled = false
+
+		// Should not panic
+		client.StopPoolMonitor()
+
+		assert.True(true, "Handled stop gracefully when monitor not running")
+	})
+
+	suite.T().Run("should handle multiple stop calls", func(t *testing.T) {
+		originalPoolEnabled := os.Getenv("GRAPHQL_POOL_WARMUP_ENABLED")
+		originalPoolSize := os.Getenv("GRAPHQL_POOL_SIZE")
+		originalEndpoint := os.Getenv("GRAPHQL_ENDPOINT")
+		defer func() {
+			os.Setenv("GRAPHQL_POOL_WARMUP_ENABLED", originalPoolEnabled)
+			os.Setenv("GRAPHQL_POOL_SIZE", originalPoolSize)
+			os.Setenv("GRAPHQL_ENDPOINT", originalEndpoint)
+		}()
+
+		os.Setenv("GRAPHQL_POOL_WARMUP_ENABLED", "true")
+		os.Setenv("GRAPHQL_POOL_SIZE", "2")
+		os.Setenv("GRAPHQL_ENDPOINT", mockServer.URL)
+
+		client := NewConnection()
+
+		// Call stop multiple times - should not panic
+		client.StopPoolMonitor()
+		client.StopPoolMonitor()
+		client.StopPoolMonitor()
+
+		assert.True(true, "Handled multiple stop calls gracefully")
+	})
+}
+
+func (suite *Tests) TestPoolHealthCheck() {
+	suite.T().Run("should perform health check without panic", func(t *testing.T) {
+		client := CreateTestClient()
+		client.pool_warmup_enabled = true
+		client.pool_size = 2
+		client.pool_warmup_query = "query{__typename}"
+
+		// Perform a health check - should not panic
+		client.performPoolHealthCheck()
+
+		assert.True(true, "Health check completed")
+	})
+
+	suite.T().Run("should refresh pool without panic", func(t *testing.T) {
+		client := CreateTestClient()
+		client.pool_size = 2
+		client.pool_warmup_query = "query{__typename}"
+
+		// Perform pool refresh - should not panic
+		client.refreshConnectionPool()
+
+		assert.True(true, "Pool refresh completed")
+	})
+}

request_compression_test.go

@@ -11,6 +11,7 @@ import (
 
 	assertions "github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/suite"
+	"golang.org/x/net/http2"
 )
 
 type RequestCompressionTestSuite struct {
@@ -74,17 +75,23 @@ func (suite *RequestCompressionTestSuite) TestRequestCompression() {
 				httpClient := client.createHttpClient()
 				assert.NotNil(httpClient, "HTTP client should be created")
 
-				// Verify transport settings
+				// Verify transport settings based on protocol
 				if strings.HasPrefix(endpoint, "http://") {
-					// HTTP/1.1 transport
+					// HTTP/1.1 transport for http:// (h2c rarely supported)
 					transport, ok := httpClient.Transport.(*http.Transport)
-					assert.True(ok, "Should be http.Transport for HTTP endpoints")
+					assert.True(ok, "Should be http.Transport for http:// endpoints")
 					if ok {
-						assert.True(transport.DisableCompression, "DisableCompression should be true for HTTP transport")
+						assert.True(transport.DisableCompression, "DisableCompression should be true to prevent trailing garbage")
 					}
 				} else {
-					// HTTP/2 transport - this is where the issue might be
-					// We need to ensure HTTP/2 transport also has compression disabled
+					// HTTP/2 transport for https://
+					transport, ok := httpClient.Transport.(*http2.Transport)
+					assert.True(ok, "Should be http2.Transport for https:// endpoints")
+					if ok {
+						assert.True(transport.DisableCompression, "DisableCompression should be true to prevent trailing garbage")
+						assert.True(transport.AllowHTTP, "AllowHTTP should be true")
+						assert.NotNil(transport.TLSClientConfig, "TLSClientConfig should be set for https:// endpoints")
+					}
 				}
 
 				qe := &QueryExecutor{

structs.go

@@ -2,6 +2,7 @@ package gql
 
 import (
 	"net/http"
+	"sync"
 	"time"
 
 	cache "github.com/lukaszraczylo/go-simple-graphql/cache"
@@ -9,17 +10,24 @@ import (
 )
 
 type BaseClient struct {
-	cache          *cache.Cache
-	Logger         *logging.Logger
-	client         *http.Client
-	endpoint       string
-	responseType   string
-	retries_delay  time.Duration
-	retries_number int
-	MaxGoRoutines  int
-	cache_global   bool
-	retries_enable bool
-	minify_queries bool // Enable GraphQL query minification (default: true)
+	cache                *cache.Cache
+	Logger               *logging.Logger
+	client               *http.Client
+	endpoint             string
+	responseType         string
+	retries_delay        time.Duration
+	retries_number       int
+	retries_patterns     []string      // Error patterns that should trigger retries
+	pool_size            int           // Number of connections to pre-warm
+	pool_warmup_enabled  bool          // Enable connection pool warmup
+	pool_warmup_query    string        // Query to use for warming up connections
+	pool_health_interval time.Duration // Interval for pool health checks
+	pool_stop            chan bool     // Channel to stop pool monitor
+	pool_stop_once       sync.Once     // Ensures StopPoolMonitor is called only once
+	MaxGoRoutines        int
+	cache_global         bool
+	retries_enable       bool
+	minify_queries       bool // Enable GraphQL query minification (default: true)
 }
 
 type Query struct {