Enhance the version endpoint and fix OTEL deployment

config.py

@@ -316,13 +316,6 @@ ENABLE_MALICIOUS_CONTENT_SCAN = True
 # Logging configuration - can be overridden in local config
 # Example: LOGS_DIR = "/var/log/ivatar"  # For production deployments
 
-# OpenTelemetry feature flag - can be disabled for F/LOSS deployments
-ENABLE_OPENTELEMETRY = os.environ.get("ENABLE_OPENTELEMETRY", "false").lower() in (
-    "true",
-    "1",
-    "yes",
-)
-
 # This MUST BE THE LAST!
 if os.path.isfile(os.path.join(BASE_DIR, "config_local.py")):
     from config_local import *  # noqa # flake8: noqa # NOQA # pragma: no cover

ivatar/opentelemetry_config.py

@@ -129,6 +129,9 @@ class OpenTelemetryConfig:
             )
             metrics.set_meter_provider(meter_provider)
 
+            # Start Prometheus HTTP server for metrics endpoint
+            self._start_prometheus_server(prometheus_reader, prometheus_endpoint)
+
             logger.info(
                 f"OpenTelemetry metrics configured with Prometheus endpoint: {prometheus_endpoint}"
             )
@@ -137,6 +140,33 @@ class OpenTelemetryConfig:
             logger.error(f"Failed to setup OpenTelemetry metrics: {e}")
             self.enabled = False
 
+    def _start_prometheus_server(
+        self, prometheus_reader: PrometheusMetricReader, endpoint: str
+    ) -> None:
+        """Start Prometheus HTTP server for metrics endpoint."""
+        try:
+            from prometheus_client import start_http_server, REGISTRY
+
+            # Parse endpoint to get host and port
+            if ":" in endpoint:
+                host, port = endpoint.split(":", 1)
+                port = int(port)
+            else:
+                host = "0.0.0.0"
+                port = int(endpoint)
+
+            # Register the PrometheusMetricReader collector with prometheus_client
+            REGISTRY.register(prometheus_reader._collector)
+
+            # Start HTTP server
+            start_http_server(port, addr=host)
+
+            logger.info(f"Prometheus metrics server started on {host}:{port}")
+
+        except Exception as e:
+            logger.error(f"Failed to start Prometheus metrics server: {e}")
+            self.enabled = False
+
     def setup_instrumentation(self) -> None:
         """Set up OpenTelemetry instrumentation for various libraries."""
         try:

ivatar/settings.py

@@ -311,16 +311,14 @@ DEFAULT_AUTO_FIELD = "django.db.models.BigAutoField"
 from config import *  # pylint: disable=wildcard-import,wrong-import-position,unused-wildcard-import  # noqa
 
 # OpenTelemetry setup - must be after config import
-# Only setup if feature flag is enabled
+# Always setup OpenTelemetry (instrumentation always enabled, export controlled by OTEL_EXPORT_ENABLED)
 try:
-    if getattr(globals(), "ENABLE_OPENTELEMETRY", False):
+    from ivatar.opentelemetry_config import setup_opentelemetry
-        from ivatar.opentelemetry_config import setup_opentelemetry
 
-        setup_opentelemetry()
+    setup_opentelemetry()
 
-        # Add OpenTelemetry middleware if enabled
+    # Add OpenTelemetry middleware (always enabled)
-        MIDDLEWARE.append("ivatar.opentelemetry_middleware.OpenTelemetryMiddleware")
+    MIDDLEWARE.append("ivatar.opentelemetry_middleware.OpenTelemetryMiddleware")
 except (ImportError, NameError):
     # OpenTelemetry packages not installed or configuration failed
-    # ENABLE_OPENTELEMETRY not defined (shouldn't happen but be safe)
     pass

ivatar/views.py

@@ -40,41 +40,13 @@ from .ivataraccount.models import Photo
 from .ivataraccount.models import pil_format, file_format
 from .utils import is_trusted_url, mm_ng, resize_animated_gif
 
-# Import OpenTelemetry only if feature flag is enabled
+# Import OpenTelemetry (always enabled, export controlled by OTEL_EXPORT_ENABLED)
 try:
-    from django.conf import settings
+    from .opentelemetry_middleware import trace_avatar_operation, get_avatar_metrics
 
-    if getattr(settings, "ENABLE_OPENTELEMETRY", False):
+    avatar_metrics = get_avatar_metrics()
-        from .opentelemetry_middleware import trace_avatar_operation, get_avatar_metrics
-
-        avatar_metrics = get_avatar_metrics()
-    else:
-        # Create no-op decorators and metrics when OpenTelemetry is disabled
-        def trace_avatar_operation(operation_name):
-            def decorator(func):
-                return func
-
-            return decorator
-
-        class NoOpMetrics:
-            def record_avatar_generated(self, *args, **kwargs):
-                pass
-
-            def record_cache_hit(self, *args, **kwargs):
-                pass
-
-            def record_cache_miss(self, *args, **kwargs):
-                pass
-
-            def record_external_request(self, *args, **kwargs):
-                pass
-
-            def record_file_upload(self, *args, **kwargs):
-                pass
-
-        avatar_metrics = NoOpMetrics()
 except ImportError:
-    # Django not available or settings not loaded
+    # OpenTelemetry packages not installed
     def trace_avatar_operation(operation_name):
         def decorator(func):
             return func
@@ -845,7 +817,7 @@ class StatsView(TemplateView, JsonResponse):
         return JsonResponse(retval)
 
 
-# Thread-safe version cache
+# Thread-safe version cache - cached indefinitely since container restarts on changes
 _version_cache = None
 _version_cache_lock = threading.Lock()
 
@@ -889,15 +861,29 @@ def _get_git_info_from_files():
             branch_name = "detached"
 
         # Try to get commit date from git log file (if available)
+        # Optimize: read only the last line instead of entire file
         commit_date = None
         log_file = path.join(git_dir, "logs", "HEAD")
         if path.exists(log_file):
             try:
-                with open(log_file, "r") as f:
+                with open(log_file, "rb") as f:
-                    # Read last line to get most recent commit info
+                    # Seek to end and read backwards to find last line
-                    lines = f.readlines()
+                    f.seek(0, 2)  # Seek to end
-                    if lines:
+                    file_size = f.tell()
-                        last_line = lines[-1].strip()
+
+                    # Read backwards in chunks to find the last line
+                    chunk_size = min(1024, file_size)
+                    f.seek(max(0, file_size - chunk_size))
+                    chunk = f.read().decode("utf-8", errors="ignore")
+
+                    # Find the last newline
+                    last_newline = chunk.rfind("\n")
+                    if last_newline != -1:
+                        last_line = chunk[last_newline + 1:].strip()
+                    else:
+                        last_line = chunk.strip()
+
+                    if last_line:
                         # Git log format: <old_hash> <new_hash> <author> <timestamp> <timezone> <message>
                         parts = last_line.split("\t")
                         if len(parts) >= 2:
@@ -910,7 +896,7 @@ def _get_git_info_from_files():
                                 commit_date = datetime.datetime.fromtimestamp(
                                     timestamp
                                 ).strftime("%Y-%m-%d %H:%M:%S %z")
-            except (ValueError, IndexError):
+            except (ValueError, IndexError, UnicodeDecodeError):
                 pass
 
         # Fallback: try to get date from commit object if available
@@ -942,6 +928,7 @@ def _get_git_info_from_files():
 def _get_cached_version_info():
     """
     Get cached version information, loading it if not available
+    Since containers restart on content changes, cache indefinitely
     """
     global _version_cache
 

requirements.txt

@@ -35,6 +35,7 @@ opentelemetry-instrumentation-urllib3>=0.42b0
 opentelemetry-sdk>=1.20.0
 Pillow
 pip
+prometheus-client>=0.20.0
 psycopg2-binary
 py3dns
 pydocstyle

scripts/check_deployment.py

@@ -20,6 +20,7 @@ import argparse
 import json
 import random
 import ssl
+import subprocess
 import sys
 import tempfile
 import time
@@ -57,6 +58,52 @@ def colored_print(message: str, color: str = Colors.NC) -> None:
     print(f"{color}{message}{Colors.NC}")
 
 
+def get_current_commit_hash() -> Optional[str]:
+    """Get the current commit hash from git."""
+    try:
+        result = subprocess.run(
+            ["git", "rev-parse", "HEAD"],
+            capture_output=True,
+            text=True,
+            check=True,
+        )
+        return result.stdout.strip()
+    except (subprocess.CalledProcessError, FileNotFoundError):
+        return None
+
+
+def is_commit_newer_or_equal(commit1: str, commit2: str) -> Optional[bool]:
+    """
+    Check if commit1 is newer than or equal to commit2 in git history.
+
+    Returns:
+        True if commit1 is newer or equal to commit2
+        False if commit1 is older than commit2
+        None if comparison fails
+    """
+    try:
+        # Use git merge-base to check if commit1 is reachable from commit2
+        # If commit1 is newer or equal, it should be reachable from commit2
+        subprocess.run(
+            ["git", "merge-base", "--is-ancestor", commit2, commit1],
+            capture_output=True,
+            check=True,
+        )
+        return True
+    except subprocess.CalledProcessError:
+        # If the above fails, try the reverse - check if commit2 is newer
+        try:
+            result = subprocess.run(
+                ["git", "merge-base", "--is-ancestor", commit1, commit2],
+                capture_output=True,
+                check=True,
+            )
+            return False
+        except subprocess.CalledProcessError:
+            # If both fail, we can't determine the relationship
+            return None
+
+
 def make_request(
     url: str,
     method: str = "GET",
@@ -289,14 +336,43 @@ def test_deployment(
             )
 
             # Display version information
-            commit_hash = version_info.get("commit_hash", "Unknown")
+            deployed_commit = version_info.get("commit_hash", "Unknown")
             branch = version_info.get("branch", "Unknown")
             version = version_info.get("version", "Unknown")
 
-            colored_print(f"Deployed commit: {commit_hash}", Colors.BLUE)
+            colored_print(f"Deployed commit: {deployed_commit}", Colors.BLUE)
             colored_print(f"Deployed branch: {branch}", Colors.BLUE)
             colored_print(f"Deployed version: {version}", Colors.BLUE)
 
+            # Check if we're looking for a specific version and compare
+            current_commit = get_current_commit_hash()
+            if current_commit and deployed_commit != "Unknown":
+                if deployed_commit == current_commit:
+                    colored_print(
+                        "✅ Exact version match - deployment is up to date!",
+                        Colors.GREEN,
+                    )
+                else:
+                    # Check if deployed version is newer
+                    comparison = is_commit_newer_or_equal(
+                        deployed_commit, current_commit
+                    )
+                    if comparison is True:
+                        colored_print(
+                            "ℹ️  Note: A newer version is already deployed (this is fine!)",
+                            Colors.YELLOW,
+                        )
+                    elif comparison is False:
+                        colored_print(
+                            "⚠️  Warning: Deployed version appears to be older than expected",
+                            Colors.YELLOW,
+                        )
+                    else:
+                        colored_print(
+                            "⚠️  Warning: Could not determine version relationship",
+                            Colors.YELLOW,
+                        )
+
             # Run functionality tests
             colored_print("Running basic functionality tests...", Colors.YELLOW)