aiqa-client 0.3.7__tar.gz → 0.4.1__tar.gz

{aiqa_client-0.3.7/aiqa_client.egg-info → aiqa_client-0.4.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: aiqa-client
-Version: 0.3.7
+Version: 0.4.1
 Summary: OpenTelemetry-based Python client for tracing functions and sending traces to the AIQA server
 Author-email: AIQA <info@aiqa.dev>
 License: MIT
@@ -56,6 +56,20 @@ pip install -r requirements.txt
 pip install -e .
 ```
 
+### Development Setup
+
+For development, install with dev dependencies to run tests:
+
+```bash
+pip install -e ".[dev]"
+```
+
+Then run the unit tests:
+
+```bash
+pytest
+```
+
 See [TESTING.md](TESTING.md) for detailed testing instructions.
 
 ## Setup

{aiqa_client-0.3.7 → aiqa_client-0.4.1}/README.md

@@ -19,6 +19,20 @@ pip install -r requirements.txt
 pip install -e .
 ```
 
+### Development Setup
+
+For development, install with dev dependencies to run tests:
+
+```bash
+pip install -e ".[dev]"
+```
+
+Then run the unit tests:
+
+```bash
+pytest
+```
+
 See [TESTING.md](TESTING.md) for detailed testing instructions.
 
 ## Setup

{aiqa_client-0.3.7 → aiqa_client-0.4.1}/aiqa/__init__.py

@@ -1,22 +1,29 @@
 """
 Python client for AIQA server - OpenTelemetry tracing decorators.
 
-IMPORTANT: Before using any AIQA functionality, you must call get_aiqa_client() to initialize
-the client and load environment variables (AIQA_SERVER_URL, AIQA_API_KEY, AIQA_COMPONENT_TAG, etc.).
+Initialization is automatic - you don't need to call get_aiqa_client() explicitly.
+The client initializes automatically when WithTracing is first used.
+
+Set environment variables:
+    AIQA_SERVER_URL: URL of the AIQA server
+    AIQA_API_KEY: API key for authentication
+    AIQA_COMPONENT_TAG: Optional component identifier
+    AIQA_STARTUP_DELAY_SECONDS: Optional delay before first flush (default: 10s)
 
 Example:
     from dotenv import load_dotenv
-    from aiqa import get_aiqa_client, WithTracing
+    from aiqa import WithTracing
     
     # Load environment variables from .env file (if using one)
     load_dotenv()
     
-    # Initialize client (must be called before using WithTracing or other functions)
-    get_aiqa_client()
-    
+    # No explicit initialization needed - it happens automatically when used
     @WithTracing
     def my_function():
         return "Hello, AIQA!"
+    
+    # Call the function - initialization happens on first use
+    result = my_function()
 """
 
 from .tracing import (
@@ -36,8 +43,7 @@ from .tracing import (
 )
 from .client import get_aiqa_client
 from .experiment_runner import ExperimentRunner
-
-__version__ = "0.3.7"
+from .constants import VERSION
 
 __all__ = [
     "WithTracing",
@@ -55,6 +61,6 @@ __all__ = [
     "set_conversation_id",
     "set_component_tag",
     "get_span",
-    "__version__",
+    "VERSION",
 ]
 

{aiqa_client-0.3.7 → aiqa_client-0.4.1}/aiqa/aiqa_exporter.py

@@ -9,12 +9,12 @@ import logging
 import threading
 import time
 import io
+import asyncio
 from typing import List, Dict, Any, Optional
 from opentelemetry.sdk.trace import ReadableSpan
 from opentelemetry.sdk.trace.export import SpanExporter, SpanExportResult
 
-from .constants import AIQA_TRACER_NAME
-from . import __version__
+from .constants import AIQA_TRACER_NAME, VERSION
 
 logger = logging.getLogger("AIQA")
 
@@ -32,6 +32,7 @@ class AIQASpanExporter(SpanExporter):
         flush_interval_seconds: float = 5.0,
         max_batch_size_bytes: int = 5 * 1024 * 1024,  # 5MB default
         max_buffer_spans: int = 10000,  # Maximum spans to buffer (prevents unbounded growth)
+        startup_delay_seconds: Optional[float] = None,
     ):
         """
         Initialize the AIQA span exporter.
@@ -41,24 +42,44 @@ class AIQASpanExporter(SpanExporter):
             api_key: API key for authentication (defaults to AIQA_API_KEY env var)
             flush_interval_seconds: How often to flush spans to the server
             max_batch_size_bytes: Maximum size of a single batch in bytes (default: 5mb)
+            max_buffer_spans: Maximum spans to buffer (prevents unbounded growth)
+            startup_delay_seconds: Delay before starting auto-flush (default: 10s, or AIQA_STARTUP_DELAY_SECONDS env var)
         """
         self._server_url = server_url
         self._api_key = api_key
         self.flush_interval_ms = flush_interval_seconds * 1000
         self.max_batch_size_bytes = max_batch_size_bytes
         self.max_buffer_spans = max_buffer_spans
+        
+        # Get startup delay from parameter or environment variable (default: 10s)
+        if startup_delay_seconds is None:
+            env_delay = os.getenv("AIQA_STARTUP_DELAY_SECONDS")
+            if env_delay:
+                try:
+                    startup_delay_seconds = float(env_delay)
+                except ValueError:
+                    logger.warning(f"Invalid AIQA_STARTUP_DELAY_SECONDS value '{env_delay}', using default 10.0")
+                    startup_delay_seconds = 10.0
+            else:
+                startup_delay_seconds = 10.0
+        self.startup_delay_seconds = startup_delay_seconds
+        
         self.buffer: List[Dict[str, Any]] = []
         self.buffer_span_keys: set = set()  # Track (traceId, spanId) tuples to prevent duplicates (Python 3.8 compatible)
         self.buffer_lock = threading.Lock()
         self.flush_lock = threading.Lock()
+        # shutdown_requested is only set once (in shutdown()) and read many times
+        # No lock needed: worst case is reading stale False, which is acceptable
         self.shutdown_requested = False
         self.flush_timer: Optional[threading.Thread] = None
+        self._auto_flush_started = False
+        self._auto_flush_lock = threading.Lock()  # Lock for lazy thread creation
         
         logger.info(
             f"Initializing AIQASpanExporter: server_url={self.server_url or 'not set'}, "
-            f"flush_interval={flush_interval_seconds}s"
+            f"flush_interval={flush_interval_seconds}s, startup_delay={startup_delay_seconds}s"
         )
-        self._start_auto_flush()
+        # Don't start thread immediately - start lazily on first export to avoid startup issues
 
     @property
     def server_url(self) -> str:         
@@ -89,6 +110,11 @@ class AIQASpanExporter(SpanExporter):
             pass
         
         logger.debug(f"AIQA export() called with {len(spans)} spans")
+        
+        # Lazy initialization: start auto-flush thread on first export
+        # This avoids thread creation during initialization, which can cause issues in ECS deployments
+        self._ensure_auto_flush_started()
+        
         # Serialize and add to buffer, deduplicating by (traceId, spanId)
         with self.buffer_lock:
             serialized_spans = []
@@ -198,7 +224,7 @@ class AIQASpanExporter(SpanExporter):
         """
         return {
             "name": AIQA_TRACER_NAME,
-            "version": __version__,
+            "version": VERSION,
         }
 
     def _time_to_tuple(self, nanoseconds: int) -> tuple:
@@ -325,6 +351,8 @@ class AIQASpanExporter(SpanExporter):
         """
         Flush buffered spans to the server. Thread-safe: ensures only one flush operation runs at a time.
         Atomically extracts spans to prevent race conditions with concurrent export() calls.
+        
+        Lock ordering: flush_lock -> buffer_lock (must be consistent to avoid deadlocks)
         """
         logger.debug("flush() called - attempting to acquire flush lock")
         with self.flush_lock:
@@ -347,49 +375,88 @@ class AIQASpanExporter(SpanExporter):
                 self._remove_span_keys_from_tracking(spans_to_flush)
                 return
 
-            logger.info(f"flush() sending {len(spans_to_flush)} span(s) to server")
-            try:
-                await self._send_spans(spans_to_flush)
-                logger.info(f"flush() successfully sent {len(spans_to_flush)} span(s) to server")
-                # Spans already removed from buffer during extraction
-                # Now clear their keys from tracking set to free memory
-                self._remove_span_keys_from_tracking(spans_to_flush)
-            except RuntimeError as error:
-                if self._is_interpreter_shutdown_error(error):
-                    if self.shutdown_requested:
-                        logger.debug(f"flush() skipped due to interpreter shutdown: {error}")
-                        # Put spans back for retry with sync send during shutdown
-                        self._prepend_spans_to_buffer(spans_to_flush)
-                    else:
-                        logger.warning(f"flush() interrupted by interpreter shutdown: {error}")
-                        # Put spans back for retry
-                        self._prepend_spans_to_buffer(spans_to_flush)
-                    raise
-                logger.error(f"Error flushing spans to server: {error}")
-                # Put spans back for retry
+        # Release flush_lock before I/O to avoid blocking other flush attempts
+        # Spans are already extracted, so concurrent exports won't interfere
+        logger.info(f"flush() sending {len(spans_to_flush)} span(s) to server")
+        try:
+            await self._send_spans(spans_to_flush)
+            logger.info(f"flush() successfully sent {len(spans_to_flush)} span(s) to server")
+            # Spans already removed from buffer during extraction
+            # Now clear their keys from tracking set to free memory
+            self._remove_span_keys_from_tracking(spans_to_flush)
+        except RuntimeError as error:
+            if self._is_interpreter_shutdown_error(error):
+                if self.shutdown_requested:
+                    logger.debug(f"flush() skipped due to interpreter shutdown: {error}")
+                else:
+                    logger.warning(f"flush() interrupted by interpreter shutdown: {error}")
+                # Put spans back for retry with sync send during shutdown
                 self._prepend_spans_to_buffer(spans_to_flush)
                 raise
-            except Exception as error:
-                logger.error(f"Error flushing spans to server: {error}")
-                # Put spans back for retry
-                self._prepend_spans_to_buffer(spans_to_flush)
-                if self.shutdown_requested:
-                    raise
+            logger.error(f"Error flushing spans to server: {error}")
+            # Put spans back for retry
+            self._prepend_spans_to_buffer(spans_to_flush)
+            raise
+        except Exception as error:
+            logger.error(f"Error flushing spans to server: {error}")
+            # Put spans back for retry
+            self._prepend_spans_to_buffer(spans_to_flush)
+            if self.shutdown_requested:
+                raise
 
-    def _start_auto_flush(self) -> None:
-        """Start the auto-flush timer."""
-        if self.shutdown_requested:
-            logger.warning("_start_auto_flush() called but shutdown already requested")
+    def _ensure_auto_flush_started(self) -> None:
+        """Ensure auto-flush thread is started (lazy initialization). Thread-safe."""
+        # Fast path: check without lock first
+        if self._auto_flush_started or self.shutdown_requested:
             return
-
-        logger.info(f"Starting auto-flush thread with interval {self.flush_interval_ms / 1000.0}s")
-
-        def flush_worker():
-            import asyncio
-            logger.debug("Auto-flush worker thread started")
+        
+        # Slow path: acquire lock and double-check
+        with self._auto_flush_lock:
+            if self._auto_flush_started or self.shutdown_requested:
+                return
+            
+            try:
+                self._start_auto_flush()
+                self._auto_flush_started = True
+            except Exception as e:
+                logger.error(f"Failed to start auto-flush thread: {e}", exc_info=True)
+                # Don't raise - allow spans to be buffered even if auto-flush fails
+                # They can still be flushed manually or on shutdown
+
+    def _flush_worker(self) -> None:
+        """Worker function for auto-flush thread. Runs in a separate thread with its own event loop."""
+        import asyncio
+        logger.debug("Auto-flush worker thread started")
+        
+        # Wait for startup delay before beginning flush operations
+        # This gives the container/application time to stabilize, which helps avoid startup issues (seen with AWS ECS, Dec 2025).
+        if self.startup_delay_seconds > 0:
+            logger.info(f"Auto-flush waiting {self.startup_delay_seconds}s before first flush (startup delay)")
+            # Sleep in small increments to allow for early shutdown
+            sleep_interval = 0.5
+            remaining_delay = self.startup_delay_seconds
+            while remaining_delay > 0 and not self.shutdown_requested:
+                sleep_time = min(sleep_interval, remaining_delay)
+                time.sleep(sleep_time)
+                remaining_delay -= sleep_time
+            
+            if self.shutdown_requested:
+                logger.debug("Auto-flush startup delay interrupted by shutdown")
+                return
+            
+            logger.info("Auto-flush startup delay complete, beginning flush operations")
+        
+        # Create event loop in this thread (isolated from main thread's event loop)
+        # This prevents interference with the main application's event loop
+        try:
             loop = asyncio.new_event_loop()
             asyncio.set_event_loop(loop)
-            
+        except Exception as e:
+            logger.error(f"Failed to create event loop for auto-flush thread: {e}", exc_info=True)
+            return
+        
+        # Ensure event loop is always closed, even if an exception occurs
+        try:
             cycle_count = 0
             while not self.shutdown_requested:
                 cycle_count += 1
@@ -397,27 +464,39 @@ class AIQASpanExporter(SpanExporter):
                 try:
                     loop.run_until_complete(self.flush())
                     logger.debug(f"Auto-flush cycle #{cycle_count} completed, sleeping {self.flush_interval_ms / 1000.0}s")
-                    time.sleep(self.flush_interval_ms / 1000.0)
                 except Exception as e:
                     logger.error(f"Error in auto-flush cycle #{cycle_count}: {e}")
                     logger.debug(f"Auto-flush cycle #{cycle_count} error handled, sleeping {self.flush_interval_ms / 1000.0}s")
+                
+                # Sleep after each cycle (including errors) to avoid tight loops
+                if not self.shutdown_requested:
                     time.sleep(self.flush_interval_ms / 1000.0)
             
             logger.info(f"Auto-flush worker thread stopping (shutdown requested). Completed {cycle_count} cycles.")
-            
             # Don't do final flush here - shutdown() will handle it with synchronous send
             # This avoids event loop shutdown issues
             logger.debug("Auto-flush thread skipping final flush (will be handled by shutdown() with sync send)")
-            
-            # Close the event loop
+        finally:
+            # Always close the event loop, even if an exception occurs
             try:
                 if not loop.is_closed():
                     loop.close()
                 logger.debug("Auto-flush worker thread event loop closed")
             except Exception:
                 pass  # Ignore errors during cleanup
+    
+    def _start_auto_flush(self) -> None:
+        """Start the auto-flush timer with startup delay."""
+        if self.shutdown_requested:
+            logger.warning("_start_auto_flush() called but shutdown already requested")
+            return
+
+        logger.info(
+            f"Starting auto-flush thread with interval {self.flush_interval_ms / 1000.0}s, "
+            f"startup delay {self.startup_delay_seconds}s"
+        )
 
-        flush_thread = threading.Thread(target=flush_worker, daemon=True, name="AIQA-AutoFlush")
+        flush_thread = threading.Thread(target=self._flush_worker, daemon=True, name="AIQA-AutoFlush")
         flush_thread.start()
         self.flush_timer = flush_thread
         logger.info(f"Auto-flush thread started: {flush_thread.name} (daemon={flush_thread.daemon})")
@@ -439,8 +518,10 @@ class AIQASpanExporter(SpanExporter):
         else:
             logger.debug("_send_spans() no API key provided")
 
+        # Use timeout to prevent hanging on unreachable servers
+        timeout = aiohttp.ClientTimeout(total=30.0, connect=10.0)
         errors = []
-        async with aiohttp.ClientSession() as session:
+        async with aiohttp.ClientSession(timeout=timeout) as session:
             for batch_idx, batch in enumerate(batches):
                 try:
                     logger.debug(f"_send_spans() sending batch {batch_idx + 1}/{len(batches)} with {len(batch)} spans to {url}")
@@ -458,6 +539,12 @@ class AIQASpanExporter(SpanExporter):
                             # Continue with other batches even if one fails
                             continue
                         logger.debug(f"_send_spans() batch {batch_idx + 1} successfully sent {len(batch)} spans")
+                except (aiohttp.ClientError, asyncio.TimeoutError) as e:
+                    # Network errors and timeouts - log but don't fail completely
+                    error_msg = f"Network error in batch {batch_idx + 1}: {type(e).__name__}: {e}"
+                    logger.warning(f"_send_spans() {error_msg} - will retry on next flush")
+                    errors.append((batch_idx + 1, error_msg))
+                    # Continue with other batches
                 except RuntimeError as e:
                     if self._is_interpreter_shutdown_error(e):
                         if self.shutdown_requested:
@@ -476,6 +563,7 @@ class AIQASpanExporter(SpanExporter):
                     # Continue with other batches
         
         # If any batches failed, raise an exception with details
+        # Spans will be restored to buffer for retry on next flush
         if errors:
             error_summary = "; ".join([f"batch {idx}: {msg}" for idx, msg in errors])
             raise Exception(f"Failed to send some spans: {error_summary}")
@@ -537,7 +625,8 @@ class AIQASpanExporter(SpanExporter):
             logger.info(f"shutdown() buffer contains {buffer_size} span(s) before shutdown")
 
         # Wait for flush thread to finish (it will do final flush)
-        if self.flush_timer and self.flush_timer.is_alive():
+        # Only wait if thread was actually started
+        if self._auto_flush_started and self.flush_timer and self.flush_timer.is_alive():
             logger.info("shutdown() waiting for auto-flush thread to complete (timeout=10s)")
             self.flush_timer.join(timeout=10.0)
             if self.flush_timer.is_alive():

{aiqa_client-0.3.7 → aiqa_client-0.4.1}/aiqa/client.py

@@ -7,9 +7,6 @@ from opentelemetry import trace
 from opentelemetry.sdk.trace import TracerProvider
 from opentelemetry.sdk.trace.export import BatchSpanProcessor
 
-if TYPE_CHECKING:
-    from .aiqa_exporter import AIQASpanExporter
-
 logger = logging.getLogger("AIQA")
 
 # Compatibility import for TraceIdRatioBased sampler
@@ -46,7 +43,7 @@ class AIQAClient:
         if cls._instance is None:
             cls._instance = super().__new__(cls)
             cls._instance._provider: Optional[TracerProvider] = None
-            cls._instance._exporter: Optional[AIQASpanExporter] = None
+            cls._instance._exporter = None # reduce circular import issues by not importing for typecheck here
             cls._instance._enabled: bool = True
             cls._instance._initialized: bool = False
         return cls._instance
@@ -62,12 +59,12 @@ class AIQAClient:
         self._provider = value
     
     @property
-    def exporter(self) -> Optional[AIQASpanExporter]:
+    def exporter(self) -> Optional[Any]:
         """Get the span exporter."""
         return self._exporter
     
     @exporter.setter
-    def exporter(self, value: Optional[AIQASpanExporter]) -> None:
+    def exporter(self, value: Optional[Any]) -> None:
         """Set the span exporter."""
         self._exporter = value
     
@@ -132,9 +129,14 @@ def get_aiqa_client() -> AIQAClient:
     """
     Initialize and return the AIQA client singleton.
     
-    This function must be called before using any AIQA tracing functionality to ensure
-    that environment variables (such as AIQA_SERVER_URL, AIQA_API_KEY, AIQA_COMPONENT_TAG)
-    are properly loaded and the tracing system is initialized.
+    This function is called automatically when WithTracing is first used, so you typically
+    don't need to call it explicitly. However, you can call it manually if you want to:
+    - Check if tracing is enabled (client.enabled)
+    - Initialize before the first @WithTracing usage
+    - Access the client object for advanced usage
+    
+    The function loads environment variables (AIQA_SERVER_URL, AIQA_API_KEY, AIQA_COMPONENT_TAG)
+    and initializes the tracing system.
     
     The client object manages the tracing system state. Tracing is done by the WithTracing 
     decorator. Experiments are run by the ExperimentRunner class.
@@ -145,12 +147,14 @@ def get_aiqa_client() -> AIQAClient:
     Example:
         from aiqa import get_aiqa_client, WithTracing
         
-        # Initialize client (loads env vars)
+        # Optional: Initialize explicitly (usually not needed)
         client = get_aiqa_client()
+        if client.enabled:
+            print("Tracing is enabled")
         
         @WithTracing
         def my_function():
-            pass
+            pass  # Initialization happens automatically here if not done above
     """
     global client
     try:
@@ -252,10 +256,10 @@ def get_aiqa_tracer() -> trace.Tracer:
     """
     try:
         # Import here to avoid circular import
-        from . import __version__
+        from . import VERSION
         # Compatibility: version parameter may not be supported in older OpenTelemetry versions
         # Try with version parameter (newer OpenTelemetry versions)
-        return trace.get_tracer(AIQA_TRACER_NAME, version=__version__)
+        return trace.get_tracer(AIQA_TRACER_NAME, version=VERSION)
     except Exception as e:
         # Log issue but still return a tracer
         logger.info(f"Issue getting AIQA tracer with version: {e}, using fallback")

{aiqa_client-0.3.7 → aiqa_client-0.4.1}/aiqa/constants.py

@@ -3,3 +3,4 @@ Constants used across the AIQA client package.
 """
 
 AIQA_TRACER_NAME = "aiqa-tracer"
+VERSION = "0.4.1" # automatically updated by set-version-json.sh

aiqa_client-0.4.1/aiqa/test_startup_reliability.py

@@ -0,0 +1,249 @@
+"""
+Test startup reliability - simulates ECS deployment scenarios where rapid initialization
+and network issues could cause deployment failures.
+
+These tests verify that:
+1. Exporter initialization doesn't block or create threads immediately
+2. Thread creation is lazy (only on first export)
+3. Network failures during startup don't cause hangs
+4. Multiple rapid initializations don't cause issues
+"""
+
+import os
+import time
+import threading
+import pytest
+from unittest.mock import patch, MagicMock
+from opentelemetry.sdk.trace import TracerProvider
+from opentelemetry.sdk.trace.export import BatchSpanProcessor
+
+from aiqa.client import get_aiqa_client, AIQAClient
+from aiqa.aiqa_exporter import AIQASpanExporter
+
+
+class TestStartupReliability:
+    """Tests for startup reliability in ECS-like scenarios."""
+
+    def test_exporter_initialization_does_not_create_thread_immediately(self):
+        """Verify that creating an exporter doesn't immediately start a thread."""
+        with patch.dict(
+            os.environ,
+            {
+                "AIQA_SERVER_URL": "http://localhost:3000",
+                "AIQA_API_KEY": "test-api-key",
+            },
+        ):
+            exporter = AIQASpanExporter(startup_delay_seconds=0.1)
+            
+            # Thread should not be created immediately
+            assert exporter.flush_timer is None
+            assert not exporter._auto_flush_started
+            
+            # Cleanup
+            exporter.shutdown()
+
+    def test_thread_created_lazily_on_first_export(self):
+        """Verify thread is only created when first span is exported."""
+        with patch.dict(
+            os.environ,
+            {
+                "AIQA_SERVER_URL": "http://localhost:3000",
+                "AIQA_API_KEY": "test-api-key",
+            },
+        ):
+            exporter = AIQASpanExporter(startup_delay_seconds=0.1)
+            
+            # Thread should not exist yet
+            assert exporter.flush_timer is None
+            
+            # Create a mock span and export it
+            from opentelemetry.sdk.trace import ReadableSpan
+            from opentelemetry.trace import SpanContext, TraceFlags
+            
+            mock_span = MagicMock(spec=ReadableSpan)
+            mock_span.get_span_context.return_value = SpanContext(
+                trace_id=1, span_id=1, is_remote=False, trace_flags=TraceFlags(0x01)
+            )
+            mock_span.name = "test_span"
+            mock_span.kind = 1
+            mock_span.start_time = 1000000000
+            mock_span.end_time = 2000000000
+            mock_span.status.status_code = 1
+            mock_span.attributes = {}
+            mock_span.links = []
+            mock_span.events = []
+            mock_span.resource.attributes = {}
+            mock_span.parent = None
+            
+            # Export should trigger thread creation
+            result = exporter.export([mock_span])
+            
+            # Give thread a moment to start
+            time.sleep(0.2)
+            
+            # Now thread should exist
+            assert exporter._auto_flush_started
+            assert exporter.flush_timer is not None
+            assert exporter.flush_timer.is_alive()
+            
+            # Cleanup
+            exporter.shutdown()
+            if exporter.flush_timer:
+                exporter.flush_timer.join(timeout=2.0)
+
+    def test_rapid_multiple_initializations(self):
+        """Test that multiple rapid initializations don't cause issues (simulates health checks)."""
+        with patch.dict(
+            os.environ,
+            {
+                "AIQA_SERVER_URL": "http://localhost:3000",
+                "AIQA_API_KEY": "test-api-key",
+            },
+        ):
+            # Simulate rapid health check calls
+            clients = []
+            for _ in range(10):
+                client = get_aiqa_client()
+                clients.append(client)
+                time.sleep(0.01)  # Very short delay
+            
+            # All should be the same singleton
+            assert all(c is clients[0] for c in clients)
+            
+            # Should not have created multiple threads
+            if clients[0].exporter:
+                assert clients[0].exporter._auto_flush_started or clients[0].exporter.flush_timer is None
+
+    def test_initialization_with_unreachable_server(self):
+        """Test that initialization doesn't hang when server is unreachable."""
+        with patch.dict(
+            os.environ,
+            {
+                "AIQA_SERVER_URL": "http://unreachable-server:3000",
+                "AIQA_API_KEY": "test-api-key",
+            },
+        ):
+            # Should not block or raise
+            client = get_aiqa_client()
+            assert client is not None
+            assert client._initialized
+            
+            # Exporter should exist but thread shouldn't be started yet
+            if client.exporter:
+                # Thread creation is lazy, so it might not exist
+                assert client.exporter.flush_timer is None or not client.exporter._auto_flush_started
+
+    def test_startup_delay_respected(self):
+        """Verify that startup delay prevents immediate flush attempts."""
+        with patch.dict(
+            os.environ,
+            {
+                "AIQA_SERVER_URL": "http://localhost:3000",
+                "AIQA_API_KEY": "test-api-key",
+            },
+        ):
+            exporter = AIQASpanExporter(startup_delay_seconds=0.5)
+            
+            # Create and export a span to trigger thread creation
+            from opentelemetry.sdk.trace import ReadableSpan
+            from opentelemetry.trace import SpanContext, TraceFlags
+            
+            mock_span = MagicMock(spec=ReadableSpan)
+            mock_span.get_span_context.return_value = SpanContext(
+                trace_id=1, span_id=1, is_remote=False, trace_flags=TraceFlags(0x01)
+            )
+            mock_span.name = "test_span"
+            mock_span.kind = 1
+            mock_span.start_time = 1000000000
+            mock_span.end_time = 2000000000
+            mock_span.status.status_code = 1
+            mock_span.attributes = {}
+            mock_span.links = []
+            mock_span.events = []
+            mock_span.resource.attributes = {}
+            mock_span.parent = None
+            
+            exporter.export([mock_span])
+            
+            # Thread should be created
+            time.sleep(0.1)
+            assert exporter._auto_flush_started
+            
+            # But flush should not have happened yet (within delay period)
+            # We can't easily test this without mocking time, but we verify thread exists
+            assert exporter.flush_timer is not None
+            
+            # Cleanup
+            exporter.shutdown()
+            if exporter.flush_timer:
+                exporter.flush_timer.join(timeout=2.0)
+
+    def test_concurrent_initialization(self):
+        """Test concurrent initialization from multiple threads (simulates ECS health checks)."""
+        with patch.dict(
+            os.environ,
+            {
+                "AIQA_SERVER_URL": "http://localhost:3000",
+                "AIQA_API_KEY": "test-api-key",
+            },
+        ):
+            clients = []
+            errors = []
+            
+            def init_client():
+                try:
+                    client = get_aiqa_client()
+                    clients.append(client)
+                except Exception as e:
+                    errors.append(e)
+            
+            # Start multiple threads initializing simultaneously
+            threads = [threading.Thread(target=init_client) for _ in range(5)]
+            for t in threads:
+                t.start()
+            for t in threads:
+                t.join(timeout=5.0)
+            
+            # Should have no errors
+            assert len(errors) == 0
+            
+            # All should be the same singleton
+            assert len(set(id(c) for c in clients)) == 1
+
+    def test_shutdown_before_thread_starts(self):
+        """Test that shutdown works even if thread was never started."""
+        with patch.dict(
+            os.environ,
+            {
+                "AIQA_SERVER_URL": "http://localhost:3000",
+                "AIQA_API_KEY": "test-api-key",
+            },
+        ):
+            exporter = AIQASpanExporter(startup_delay_seconds=1.0)
+            
+            # Thread should not exist
+            assert exporter.flush_timer is None
+            
+            # Shutdown should work without errors
+            exporter.shutdown()
+            
+            # Should still be able to call shutdown again
+            exporter.shutdown()
+
+    def test_initialization_timeout(self):
+        """Test that initialization completes quickly even with network issues."""
+        with patch.dict(
+            os.environ,
+            {
+                "AIQA_SERVER_URL": "http://localhost:3000",
+                "AIQA_API_KEY": "test-api-key",
+            },
+        ):
+            start_time = time.time()
+            client = get_aiqa_client()
+            elapsed = time.time() - start_time
+            
+            # Initialization should be fast (< 1 second)
+            assert elapsed < 1.0
+            assert client is not None
+

{aiqa_client-0.3.7 → aiqa_client-0.4.1}/aiqa/tracing.py

@@ -590,7 +590,8 @@ def WithTracing(
         is_generator = inspect.isgeneratorfunction(fn)
         is_async_generator = inspect.isasyncgenfunction(fn) if hasattr(inspect, 'isasyncgenfunction') else False
         
-        tracer = get_aiqa_tracer()
+        # Don't get tracer here - get it lazily when function is called
+        # This ensures initialization only happens when tracing is actually used
         
         def _setup_span(span: trace.Span, input_data: Any) -> bool:
             """Setup span with input data. Returns True if span is recording."""
@@ -627,10 +628,13 @@ def WithTracing(
         def _execute_with_span_sync(executor: Callable[[], Any], input_data: Any) -> Any:
             """Execute sync function within span context, handling input/output and exceptions."""
             # Ensure tracer provider is initialized before creating spans
+            # This is called lazily when the function runs, not at decorator definition time
             client = get_aiqa_client()
             if not client.enabled:
                 return executor()
             
+            # Get tracer after initialization (lazy)
+            tracer = get_aiqa_tracer()
             with tracer.start_as_current_span(fn_name) as span:
                 if not _setup_span(span, input_data):
                     return executor()
@@ -646,10 +650,13 @@ def WithTracing(
         async def _execute_with_span_async(executor: Callable[[], Any], input_data: Any) -> Any:
             """Execute async function within span context, handling input/output and exceptions."""
             # Ensure tracer provider is initialized before creating spans
+            # This is called lazily when the function runs, not at decorator definition time
             client = get_aiqa_client()
             if not client.enabled:
                 return await executor()
             
+            # Get tracer after initialization (lazy)
+            tracer = get_aiqa_tracer()
             with tracer.start_as_current_span(fn_name) as span:
                 if not _setup_span(span, input_data):
                     return await executor()
@@ -668,10 +675,13 @@ def WithTracing(
         def _execute_generator_sync(executor: Callable[[], Any], input_data: Any) -> Any:
             """Execute sync generator function, returning a traced generator."""
             # Ensure tracer provider is initialized before creating spans
+            # This is called lazily when the function runs, not at decorator definition time
             client = get_aiqa_client()
             if not client.enabled:
                 return executor()
             
+            # Get tracer after initialization (lazy)
+            tracer = get_aiqa_tracer()
             # Create span but don't use 'with' - span will be closed by TracedGenerator
             span = tracer.start_span(fn_name)
             token = trace.context_api.attach(trace.context_api.set_span_in_context(span))
@@ -694,10 +704,13 @@ def WithTracing(
         async def _execute_generator_async(executor: Callable[[], Any], input_data: Any) -> Any:
             """Execute async generator function, returning a traced async generator."""
             # Ensure tracer provider is initialized before creating spans
+            # This is called lazily when the function runs, not at decorator definition time
             client = get_aiqa_client()
             if not client.enabled:
                 return await executor()
             
+            # Get tracer after initialization (lazy)
+            tracer = get_aiqa_tracer()
             # Create span but don't use 'with' - span will be closed by TracedAsyncGenerator
             span = tracer.start_span(fn_name)
             token = trace.context_api.attach(trace.context_api.set_span_in_context(span))
@@ -935,7 +948,8 @@ def set_component_tag(tag: str) -> None:
     This can also be set via the AIQA_COMPONENT_TAG environment variable.
     The component tag allows you to identify which component/system generated the spans.
     
-    Note: If using environment variables, ensure you call get_aiqa_client() first to initialize
+    Note: Initialization is automatic when WithTracing is first used. You can also call
+    get_aiqa_client() explicitly if needed.
     the client and load environment variables.
     
     Args:
@@ -1045,6 +1059,8 @@ def create_span_from_trace_id(
         from opentelemetry.trace import set_span_in_context
         parent_context = set_span_in_context(trace.NonRecordingSpan(parent_span_context))
         
+        # Ensure initialization before creating span
+        get_aiqa_client()
         # Start a new span in this context (it will be a child of the parent span)
         tracer = get_aiqa_tracer()
         span = tracer.start_span(span_name, context=parent_context)
@@ -1057,6 +1073,8 @@ def create_span_from_trace_id(
         return span
     except (ValueError, AttributeError) as e:
         logger.error(f"Error creating span from trace_id: {e}")
+        # Ensure initialization before creating span
+        get_aiqa_client()
         # Fallback: create a new span
         tracer = get_aiqa_tracer()
         span = tracer.start_span(span_name)

{aiqa_client-0.3.7 → aiqa_client-0.4.1/aiqa_client.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: aiqa-client
-Version: 0.3.7
+Version: 0.4.1
 Summary: OpenTelemetry-based Python client for tracing functions and sending traces to the AIQA server
 Author-email: AIQA <info@aiqa.dev>
 License: MIT
@@ -56,6 +56,20 @@ pip install -r requirements.txt
 pip install -e .
 ```
 
+### Development Setup
+
+For development, install with dev dependencies to run tests:
+
+```bash
+pip install -e ".[dev]"
+```
+
+Then run the unit tests:
+
+```bash
+pytest
+```
+
 See [TESTING.md](TESTING.md) for detailed testing instructions.
 
 ## Setup

{aiqa_client-0.3.7 → aiqa_client-0.4.1}/aiqa_client.egg-info/SOURCES.txt

@@ -11,6 +11,7 @@ aiqa/experiment_runner.py
 aiqa/object_serialiser.py
 aiqa/py.typed
 aiqa/test_experiment_runner.py
+aiqa/test_startup_reliability.py
 aiqa/test_tracing.py
 aiqa/tracing.py
 aiqa_client.egg-info/PKG-INFO

{aiqa_client-0.3.7 → aiqa_client-0.4.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "aiqa-client"
-version = "0.3.7"
+version = "0.4.1"
 description = "OpenTelemetry-based Python client for tracing functions and sending traces to the AIQA server"
 readme = "README.md"
 requires-python = ">=3.8"