sentienceapi 0.90.12__py3-none-any.whl → 0.92.2__py3-none-any.whl

sentience/cloud_tracing.py

@@ -4,16 +4,20 @@ Cloud trace sink with pre-signed URL upload.
 Implements "Local Write, Batch Upload" pattern for enterprise cloud tracing.
 """
 
+import base64
 import gzip
 import json
 import os
 import threading
 from collections.abc import Callable
+from concurrent.futures import ThreadPoolExecutor, as_completed
 from pathlib import Path
-from typing import Any, Protocol
+from typing import Any, Optional, Protocol, Union
 
 import requests
 
+from sentience.models import TraceStats
+from sentience.trace_file_manager import TraceFileManager
 from sentience.tracing import TraceSink
 
 
@@ -95,6 +99,7 @@ class CloudTraceSink(TraceSink):
         # Use persistent cache directory instead of temp file
         # This ensures traces survive process crashes
         cache_dir = Path.home() / ".sentience" / "traces" / "pending"
+        # Create directory if it doesn't exist (ensure_directory is for file paths, not dirs)
         cache_dir.mkdir(parents=True, exist_ok=True)
 
         # Persistent file (survives process crash)
@@ -103,9 +108,11 @@ class CloudTraceSink(TraceSink):
         self._closed = False
         self._upload_successful = False
 
-        # File size tracking (NEW)
+        # File size tracking
         self.trace_file_size_bytes = 0
         self.screenshot_total_size_bytes = 0
+        self.screenshot_count = 0  # Track number of screenshots extracted
+        self.index_file_size_bytes = 0  # Track index file size
 
     def emit(self, event: dict[str, Any]) -> None:
         """
@@ -119,9 +126,7 @@ class CloudTraceSink(TraceSink):
         if self._closed:
             raise RuntimeError("CloudTraceSink is closed")
 
-        json_str = json.dumps(event, ensure_ascii=False)
-        self._trace_file.write(json_str + "\n")
-        self._trace_file.flush()  # Ensure written to disk
+        TraceFileManager.write_event(self._trace_file, event)
 
     def close(
         self,
@@ -142,9 +147,25 @@ class CloudTraceSink(TraceSink):
 
         self._closed = True
 
-        # Close file first
+        # Flush and sync file to disk before closing to ensure all data is written
+        # This is critical on CI systems where file system operations may be slower
+        self._trace_file.flush()
+        try:
+            # Force OS to write buffered data to disk
+            os.fsync(self._trace_file.fileno())
+        except (OSError, AttributeError):
+            # Some file handles don't support fsync (e.g., StringIO in tests)
+            # This is fine - flush() is usually sufficient
+            pass
         self._trace_file.close()
 
+        # Ensure file exists and has content before proceeding
+        if not self._path.exists() or self._path.stat().st_size == 0:
+            # No events were emitted, nothing to upload
+            if self.logger:
+                self.logger.warning("No trace events to upload (file is empty or missing)")
+            return
+
         # Generate index after closing file
         self._generate_index()
 
@@ -165,21 +186,36 @@ class CloudTraceSink(TraceSink):
         """
         Internal upload method with progress tracking.
 
+        Extracts screenshots from trace events, uploads them separately,
+        then removes screenshot_base64 from events before uploading trace.
+
         Args:
             on_progress: Optional callback(uploaded_bytes, total_bytes) for progress updates
         """
         try:
-            # Read and compress
-            with open(self._path, "rb") as f:
+            # Step 1: Extract screenshots from trace events
+            screenshots = self._extract_screenshots_from_trace()
+            self.screenshot_count = len(screenshots)
+
+            # Step 2: Upload screenshots separately
+            if screenshots:
+                self._upload_screenshots(screenshots, on_progress)
+
+            # Step 3: Create cleaned trace file (without screenshot_base64)
+            cleaned_trace_path = self._path.with_suffix(".cleaned.jsonl")
+            self._create_cleaned_trace(cleaned_trace_path)
+
+            # Step 4: Read and compress cleaned trace
+            with open(cleaned_trace_path, "rb") as f:
                 trace_data = f.read()
 
             compressed_data = gzip.compress(trace_data)
             compressed_size = len(compressed_data)
 
-            # Measure trace file size (NEW)
+            # Measure trace file size
             self.trace_file_size_bytes = compressed_size
 
-            # Log file sizes if logger is provided (NEW)
+            # Log file sizes if logger is provided
             if self.logger:
                 self.logger.info(
                     f"Trace file size: {self.trace_file_size_bytes / 1024 / 1024:.2f} MB"
@@ -192,8 +228,9 @@ class CloudTraceSink(TraceSink):
             if on_progress:
                 on_progress(0, compressed_size)
 
-            # Upload to DigitalOcean Spaces via pre-signed URL
-            print(f"📤 [Sentience] Uploading trace to cloud ({compressed_size} bytes)...")
+            # Step 5: Upload cleaned trace to cloud
+            if self.logger:
+                self.logger.info(f"Uploading trace to cloud ({compressed_size} bytes)")
 
             response = requests.put(
                 self.upload_url,
@@ -208,6 +245,8 @@ class CloudTraceSink(TraceSink):
             if response.status_code == 200:
                 self._upload_successful = True
                 print("✅ [Sentience] Trace uploaded successfully")
+                if self.logger:
+                    self.logger.info("Trace uploaded successfully")
 
                 # Report progress: complete
                 if on_progress:
@@ -219,22 +258,28 @@ class CloudTraceSink(TraceSink):
                 # Call /v1/traces/complete to report file sizes
                 self._complete_trace()
 
-                # Delete file only on successful upload
-                if os.path.exists(self._path):
-                    try:
-                        os.remove(self._path)
-                    except Exception:
-                        pass  # Ignore cleanup errors
+                # Delete files only on successful upload
+                self._cleanup_files()
+
+                # Clean up temporary cleaned trace file
+                if cleaned_trace_path.exists():
+                    cleaned_trace_path.unlink()
             else:
                 self._upload_successful = False
                 print(f"❌ [Sentience] Upload failed: HTTP {response.status_code}")
-                print(f"   Response: {response.text}")
+                print(f"   Response: {response.text[:200]}")
                 print(f"   Local trace preserved at: {self._path}")
+                if self.logger:
+                    self.logger.error(
+                        f"Upload failed: HTTP {response.status_code}, Response: {response.text[:200]}"
+                    )
 
         except Exception as e:
             self._upload_successful = False
             print(f"❌ [Sentience] Error uploading trace: {e}")
             print(f"   Local trace preserved at: {self._path}")
+            if self.logger:
+                self.logger.error(f"Error uploading trace: {e}")
             # Don't raise - preserve trace locally even if upload fails
 
     def _generate_index(self) -> None:
@@ -242,10 +287,15 @@ class CloudTraceSink(TraceSink):
         try:
             from .trace_indexing import write_trace_index
 
-            write_trace_index(str(self._path))
+            # Use frontend format to ensure 'step' field is present (1-based)
+            # Frontend derives sequence from step.step - 1, so step must be valid
+            index_path = Path(str(self._path).replace(".jsonl", ".index.json"))
+            write_trace_index(str(self._path), str(index_path), frontend_format=True)
         except Exception as e:
             # Non-fatal: log but don't crash
             print(f"⚠️  Failed to generate trace index: {e}")
+            if self.logger:
+                self.logger.warning(f"Failed to generate trace index: {e}")
 
     def _upload_index(self) -> None:
         """
@@ -292,17 +342,35 @@ class CloudTraceSink(TraceSink):
                     self.logger.warning("No upload URL in index upload response")
                 return
 
-            # Read and compress index file
-            with open(index_path, "rb") as f:
-                index_data = f.read()
+            # Read index file and update trace_file.path to cloud storage path
+            with open(index_path, encoding="utf-8") as f:
+                index_json = json.load(f)
+
+            # Extract cloud storage path from trace upload URL
+            # upload_url format: https://...digitaloceanspaces.com/traces/{run_id}.jsonl.gz
+            # Extract path: traces/{run_id}.jsonl.gz
+            try:
+                from urllib.parse import urlparse
+
+                parsed_url = urlparse(self.upload_url)
+                # Extract path after domain (e.g., /traces/run-123.jsonl.gz -> traces/run-123.jsonl.gz)
+                cloud_trace_path = parsed_url.path.lstrip("/")
+                # Update trace_file.path in index
+                if "trace_file" in index_json and isinstance(index_json["trace_file"], dict):
+                    index_json["trace_file"]["path"] = cloud_trace_path
+            except Exception as e:
+                if self.logger:
+                    self.logger.warning(f"Failed to extract cloud path from upload URL: {e}")
 
+            # Serialize updated index to JSON
+            index_data = json.dumps(index_json, indent=2).encode("utf-8")
             compressed_index = gzip.compress(index_data)
             index_size = len(compressed_index)
+            self.index_file_size_bytes = index_size  # Track index file size
 
             if self.logger:
                 self.logger.info(f"Index file size: {index_size / 1024:.2f} KB")
-
-            print(f"📤 [Sentience] Uploading trace index ({index_size} bytes)...")
+                self.logger.info(f"Uploading trace index ({index_size} bytes)")
 
             # Upload index to cloud storage
             index_response = requests.put(
@@ -316,7 +384,8 @@ class CloudTraceSink(TraceSink):
             )
 
             if index_response.status_code == 200:
-                print("✅ [Sentience] Trace index uploaded successfully")
+                if self.logger:
+                    self.logger.info("Trace index uploaded successfully")
 
                 # Delete local index file after successful upload
                 try:
@@ -325,20 +394,108 @@ class CloudTraceSink(TraceSink):
                     pass  # Ignore cleanup errors
             else:
                 if self.logger:
-                    self.logger.warning(
-                        f"Index upload failed: HTTP {index_response.status_code}"
-                    )
-                print(f"⚠️  [Sentience] Index upload failed: HTTP {index_response.status_code}")
+                    self.logger.warning(f"Index upload failed: HTTP {index_response.status_code}")
 
         except Exception as e:
             # Non-fatal: log but don't crash
             if self.logger:
                 self.logger.warning(f"Error uploading trace index: {e}")
-            print(f"⚠️  [Sentience] Error uploading trace index: {e}")
+
+    def _infer_final_status_from_trace(
+        self, events: list[dict[str, Any]], run_end: dict[str, Any] | None
+    ) -> str:
+        """
+        Infer final status from trace events by reading the trace file.
+
+        Returns:
+            Final status: "success", "failure", "partial", or "unknown"
+        """
+        try:
+            # Read trace file to analyze events
+            with open(self._path, encoding="utf-8") as f:
+                events = []
+                for line in f:
+                    line = line.strip()
+                    if not line:
+                        continue
+                    try:
+                        event = json.loads(line)
+                        events.append(event)
+                    except json.JSONDecodeError:
+                        continue
+
+            if not events:
+                return "unknown"
+
+            # Check for run_end event with status
+            for event in reversed(events):
+                if event.get("type") == "run_end":
+                    status = event.get("data", {}).get("status")
+                    if status in ("success", "failure", "partial", "unknown"):
+                        return status
+
+            # Infer from error events
+            has_errors = any(e.get("type") == "error" for e in events)
+            if has_errors:
+                # Check if there are successful steps too (partial success)
+                step_ends = [e for e in events if e.get("type") == "step_end"]
+                if step_ends:
+                    return "partial"
+                return "failure"
+
+            # If we have step_end events and no errors, likely success
+            step_ends = [e for e in events if e.get("type") == "step_end"]
+            if step_ends:
+                return "success"
+
+            return "unknown"
+
+        except Exception:
+            # If we can't read the trace, default to unknown
+            return "unknown"
+
+    def _extract_stats_from_trace(self) -> TraceStats:
+        """
+        Extract execution statistics from trace file.
+
+        Returns:
+            TraceStats with stats fields for /v1/traces/complete
+        """
+        try:
+            # Check if file exists before reading
+            if not self._path.exists():
+                if self.logger:
+                    self.logger.warning(f"Trace file not found: {self._path}")
+                return TraceStats(
+                    total_steps=0,
+                    total_events=0,
+                    duration_ms=None,
+                    final_status="unknown",
+                    started_at=None,
+                    ended_at=None,
+                )
+
+            # Read trace file to extract stats
+            events = TraceFileManager.read_events(self._path)
+            # Use TraceFileManager to extract stats (with custom status inference)
+            return TraceFileManager.extract_stats(
+                events, infer_status_func=self._infer_final_status_from_trace
+            )
+        except Exception as e:
+            if self.logger:
+                self.logger.warning(f"Error extracting stats from trace: {e}")
+            return TraceStats(
+                total_steps=0,
+                total_events=0,
+                duration_ms=None,
+                final_status="unknown",
+                started_at=None,
+                ended_at=None,
+            )
 
     def _complete_trace(self) -> None:
         """
-        Call /v1/traces/complete to report file sizes to gateway.
+        Call /v1/traces/complete to report file sizes and stats to gateway.
 
         This is a best-effort call - failures are logged but don't affect upload success.
         """
@@ -347,15 +504,24 @@ class CloudTraceSink(TraceSink):
             return
 
         try:
+            # Extract stats from trace file
+            stats = self._extract_stats_from_trace()
+
+            # Build completion payload with stats and file size fields
+            completion_payload = {
+                **stats.model_dump(),  # Convert TraceStats to dict
+                "trace_file_size_bytes": self.trace_file_size_bytes,
+                "screenshot_total_size_bytes": self.screenshot_total_size_bytes,
+                "screenshot_count": self.screenshot_count,
+                "index_file_size_bytes": self.index_file_size_bytes,
+            }
+
             response = requests.post(
                 f"{self.api_url}/v1/traces/complete",
                 headers={"Authorization": f"Bearer {self.api_key}"},
                 json={
                     "run_id": self.run_id,
-                    "stats": {
-                        "trace_file_size_bytes": self.trace_file_size_bytes,
-                        "screenshot_total_size_bytes": self.screenshot_total_size_bytes,
-                    },
+                    "stats": completion_payload,
                 },
                 timeout=10,
             )
@@ -374,6 +540,262 @@ class CloudTraceSink(TraceSink):
             if self.logger:
                 self.logger.warning(f"Error reporting trace completion: {e}")
 
+    def _extract_screenshots_from_trace(self) -> dict[int, dict[str, Any]]:
+        """
+        Extract screenshots from trace events.
+
+        Returns:
+            dict mapping sequence number to screenshot data:
+            {seq: {"base64": str, "format": str, "step_id": str}}
+        """
+        screenshots: dict[int, dict[str, Any]] = {}
+        sequence = 0
+
+        try:
+            # Check if file exists before reading
+            if not self._path.exists():
+                if self.logger:
+                    self.logger.warning(f"Trace file not found: {self._path}")
+                return screenshots
+
+            events = TraceFileManager.read_events(self._path)
+            for event in events:
+                # Check if this is a snapshot event with screenshot
+                if event.get("type") == "snapshot":
+                    data = event.get("data", {})
+                    screenshot_base64 = data.get("screenshot_base64")
+
+                    if screenshot_base64:
+                        sequence += 1
+                        screenshots[sequence] = {
+                            "base64": screenshot_base64,
+                            "format": data.get("screenshot_format", "jpeg"),
+                            "step_id": event.get("step_id"),
+                        }
+        except Exception as e:
+            if self.logger:
+                self.logger.error(f"Error extracting screenshots: {e}")
+
+        return screenshots
+
+    def _create_cleaned_trace(self, output_path: Path) -> None:
+        """
+        Create trace file without screenshot_base64 fields.
+
+        Args:
+            output_path: Path to write cleaned trace file
+        """
+        try:
+            # Check if file exists before reading
+            if not self._path.exists():
+                if self.logger:
+                    self.logger.warning(f"Trace file not found: {self._path}")
+                # Create empty cleaned trace file
+                output_path.parent.mkdir(parents=True, exist_ok=True)
+                output_path.touch()
+                return
+
+            events = TraceFileManager.read_events(self._path)
+            with open(output_path, "w", encoding="utf-8") as outfile:
+                for event in events:
+                    # Remove screenshot_base64 from snapshot events
+                    if event.get("type") == "snapshot":
+                        data = event.get("data", {})
+                        if "screenshot_base64" in data:
+                            # Create copy without screenshot fields
+                            cleaned_data = {
+                                k: v
+                                for k, v in data.items()
+                                if k not in ("screenshot_base64", "screenshot_format")
+                            }
+                            event["data"] = cleaned_data
+
+                    # Write cleaned event
+                    TraceFileManager.write_event(outfile, event)
+        except Exception as e:
+            if self.logger:
+                self.logger.error(f"Error creating cleaned trace: {e}")
+            raise
+
+    def _request_screenshot_urls(self, sequences: list[int]) -> dict[int, str]:
+        """
+        Request pre-signed upload URLs for screenshots from gateway.
+
+        Args:
+            sequences: List of screenshot sequence numbers
+
+        Returns:
+            dict mapping sequence number to upload URL
+        """
+        if not self.api_key or not sequences:
+            return {}
+
+        try:
+            response = requests.post(
+                f"{self.api_url}/v1/screenshots/init",
+                headers={"Authorization": f"Bearer {self.api_key}"},
+                json={
+                    "run_id": self.run_id,
+                    "sequences": sequences,
+                },
+                timeout=10,
+            )
+
+            if response.status_code == 200:
+                data = response.json()
+                # Gateway returns sequences as strings in JSON, convert to int keys
+                upload_urls = data.get("upload_urls", {})
+                result = {int(k): v for k, v in upload_urls.items()}
+                if self.logger:
+                    self.logger.info(f"Received {len(result)} screenshot upload URLs")
+                return result
+            else:
+                error_msg = f"Failed to get screenshot URLs: HTTP {response.status_code}"
+                if self.logger:
+                    # Try to get error details
+                    try:
+                        error_data = response.json()
+                        error_detail = error_data.get("error") or error_data.get("message", "")
+                        if error_detail:
+                            self.logger.warning(f"{error_msg}: {error_detail}")
+                        else:
+                            self.logger.warning(f"{error_msg}: {response.text[:200]}")
+                    except Exception:
+                        self.logger.warning(f"{error_msg}: {response.text[:200]}")
+                return {}
+        except Exception as e:
+            error_msg = f"Error requesting screenshot URLs: {e}"
+            if self.logger:
+                self.logger.warning(error_msg)
+            return {}
+
+    def _upload_screenshots(
+        self,
+        screenshots: dict[int, dict[str, Any]],
+        on_progress: Callable[[int, int], None] | None = None,
+    ) -> None:
+        """
+        Upload screenshots extracted from trace events.
+
+        Steps:
+        1. Request pre-signed URLs from gateway (/v1/screenshots/init)
+        2. Decode base64 to image bytes
+        3. Upload screenshots in parallel (10 concurrent workers)
+        4. Track upload progress
+
+        Args:
+            screenshots: dict mapping sequence to screenshot data
+            on_progress: Optional callback(uploaded_count, total_count)
+        """
+        if not screenshots:
+            return
+
+        # 1. Request pre-signed URLs from gateway
+        sequences = sorted(screenshots.keys())
+        if self.logger:
+            self.logger.info(f"Requesting upload URLs for {len(sequences)} screenshot(s)")
+        upload_urls = self._request_screenshot_urls(sequences)
+
+        if not upload_urls:
+            if self.logger:
+                self.logger.warning(
+                    "No screenshot upload URLs received, skipping upload. "
+                    "This may indicate API key permission issue, gateway error, or network problem."
+                )
+            return
+
+        # 2. Upload screenshots in parallel
+        uploaded_count = 0
+        total_count = len(upload_urls)
+        failed_sequences: list[int] = []
+
+        def upload_one(seq: int, url: str) -> bool:
+            """Upload a single screenshot. Returns True if successful."""
+            try:
+                screenshot_data = screenshots[seq]
+                base64_str = screenshot_data["base64"]
+                format_str = screenshot_data.get("format", "jpeg")
+
+                # Decode base64 to image bytes
+                image_bytes = base64.b64decode(base64_str)
+                image_size = len(image_bytes)
+
+                # Update total size
+                self.screenshot_total_size_bytes += image_size
+
+                # Upload to pre-signed URL
+                response = requests.put(
+                    url,
+                    data=image_bytes,  # Binary image data
+                    headers={
+                        "Content-Type": f"image/{format_str}",
+                    },
+                    timeout=30,  # 30 second timeout per screenshot
+                )
+
+                if response.status_code == 200:
+                    if self.logger:
+                        self.logger.info(
+                            f"Screenshot {seq} uploaded successfully ({image_size / 1024:.1f} KB)"
+                        )
+                    return True
+                else:
+                    error_msg = f"Screenshot {seq} upload failed: HTTP {response.status_code}"
+                    if self.logger:
+                        try:
+                            error_detail = response.text[:200]
+                            if error_detail:
+                                self.logger.warning(f"{error_msg}: {error_detail}")
+                            else:
+                                self.logger.warning(error_msg)
+                        except Exception:
+                            self.logger.warning(error_msg)
+                    return False
+            except Exception as e:
+                error_msg = f"Screenshot {seq} upload error: {e}"
+                if self.logger:
+                    self.logger.warning(error_msg)
+                return False
+
+        # Upload in parallel (max 10 concurrent)
+        with ThreadPoolExecutor(max_workers=10) as executor:
+            futures = {
+                executor.submit(upload_one, seq, url): seq for seq, url in upload_urls.items()
+            }
+
+            for future in as_completed(futures):
+                seq = futures[future]
+                if future.result():
+                    uploaded_count += 1
+                    if on_progress:
+                        on_progress(uploaded_count, total_count)
+                else:
+                    failed_sequences.append(seq)
+
+        # 3. Report results
+        if uploaded_count == total_count:
+            total_size_mb = self.screenshot_total_size_bytes / 1024 / 1024
+            if self.logger:
+                self.logger.info(
+                    f"All {total_count} screenshots uploaded successfully "
+                    f"(total size: {total_size_mb:.2f} MB)"
+                )
+        else:
+            if self.logger:
+                self.logger.warning(
+                    f"Uploaded {uploaded_count}/{total_count} screenshots. "
+                    f"Failed sequences: {failed_sequences if failed_sequences else 'none'}"
+                )
+
+    def _cleanup_files(self) -> None:
+        """Delete local files after successful upload."""
+        # Delete trace file
+        if os.path.exists(self._path):
+            try:
+                os.remove(self._path)
+            except Exception:
+                pass  # Ignore cleanup errors
+
     def __enter__(self):
         """Context manager support."""
         return self