Merge branch 'add/sweep_status_bg_worker' of https://github.com/transformerlab/transformerlab-app into add/sweep_status_bg_worker

Author: dadmobile

.coderabbit.yaml

@@ -0,0 +1,34 @@
+reviews:
+  profile: 'chill' # less aggressive than default
+
+  high_level_summary: true
+  review_start_comment: false # disable the "starting review" comment
+  diff_only: true # Stop reviewing old code that is not part of the PR
+  request_changes_workflow: false # don't auto-block PRs
+  poem: false
+
+  instructions: >
+    Do NOT suggest adding type annotations.
+    Do NOT comment on missing return type hints.
+    Focus on:
+    - runtime correctness
+    - security issues
+    - performance issues
+    - race conditions
+    - architectural risks
+
+  auto_review:
+    enabled: true
+    drafts: false # don't review draft PRs
+
+  path_filters:
+    - '!**/*.md'
+    - '!docs/**'
+    - '!**/*.json'
+    - '!**/*.lock'
+    - '!dist/**'
+    - '!build/**'
+
+  tools:
+    ruff:
+      enabled: false # we already run this

.gitignore

@@ -35,6 +35,7 @@ npm-debug.log.*
 # https://docs.cursor.com/context/ignore-files#default-ignore-list
 .cursor/
 .cursorignore
+.claude/
 
 .env
 **/.env

api/transformerlab/compute_providers/local.py

@@ -1,6 +1,8 @@
 """Local compute provider: runs tasks in a uv venv synced with the base environment."""
 
+import contextlib
 import os
+import signal
 import subprocess
 import sys
 from pathlib import Path
@@ -85,6 +87,48 @@ def _get_uv_pip_install_flags() -> str:
     return ""
 
 
+def _terminate_process_tree(pid: int, sig: int = signal.SIGTERM) -> None:
+    """
+    Best-effort termination of a process and all of its descendants.
+
+    Uses psutil when available to walk the full process tree and then force-kill
+    any survivors; otherwise falls back to killing the process group (if possible)
+    and then the single pid.
+    """
+    try:
+        import psutil  # type: ignore[import-not-found]
+    except Exception:
+        psutil = None  # type: ignore[assignment]
+
+    if psutil is not None:
+        try:
+            parent = psutil.Process(pid)
+        except psutil.NoSuchProcess:
+            parent = None
+
+        if parent is not None:
+            procs = [parent] + parent.children(recursive=True)
+
+            # First try graceful termination
+            for proc in procs:
+                with contextlib.suppress(psutil.NoSuchProcess, psutil.AccessDenied, psutil.Error):
+                    proc.send_signal(sig)
+
+            # Give processes a short window to exit, then force kill survivors
+            gone, alive = psutil.wait_procs(procs, timeout=3)  # type: ignore[assignment]
+            for proc in alive:
+                with contextlib.suppress(psutil.NoSuchProcess, psutil.AccessDenied, psutil.Error):
+                    proc.kill()
+            return
+
+    # Fallback path when psutil is unavailable or parent no longer exists
+    try:
+        pgid = os.getpgid(pid)
+        os.killpg(pgid, sig)
+    except Exception:
+        os.kill(pid, sig)
+
+
 def _is_process_zombie(pid: int) -> bool:
     """
     Return True if the process is a zombie/defunct process.
@@ -172,18 +216,19 @@ def launch_cluster(self, cluster_name: str, config: ClusterConfig) -> Dict[str,
         if not job_dir or not os.path.isdir(job_dir):
             raise ValueError("Local provider requires workspace_dir (job directory) in provider_config")
         job_dir = Path(job_dir)
-        venv_path = job_dir / "venv"
-        venv_path.mkdir(parents=True, exist_ok=True)
-
-        self._ensure_venv_and_sync(venv_path)
-
         # Use a per-job workspace directory as HOME for local runs so tools that
         # rely on ~ and $HOME resolve inside the job workspace instead of the
         # user's real home directory. This makes it easier to clone and run
         # code in an isolated workspace for each job.
         workspace_home = job_dir / "workspace"
         workspace_home.mkdir(parents=True, exist_ok=True)
 
+        # Create the venv inside the per-job workspace HOME directory so that all
+        # environment state (including Python packages) lives under HOME.
+        venv_path = workspace_home / "venv"
+
+        self._ensure_venv_and_sync(venv_path)
+
         venv_bin = venv_path / "bin"
         env = os.environ.copy()
         env.update(config.env_vars or {})
@@ -226,7 +271,7 @@ def launch_cluster(self, cluster_name: str, config: ClusterConfig) -> Dict[str,
         }
 
     def stop_cluster(self, cluster_name: str) -> Dict[str, Any]:
-        """Stop the local process for this cluster (SIGTERM)."""
+        """Stop the local process tree for this cluster (SIGTERM)."""
         job_dir = self.extra_config.get("workspace_dir")
         if not job_dir:
             return {
@@ -243,8 +288,8 @@ def stop_cluster(self, cluster_name: str) -> Dict[str, Any]:
             }
         try:
             pid = int(pid_file.read_text().strip())
-            os.kill(pid, 15)
-            return {"cluster_name": cluster_name, "status": "stopped", "message": "Sent SIGTERM"}
+            _terminate_process_tree(pid, signal.SIGTERM)
+            return {"cluster_name": cluster_name, "status": "stopped", "message": "Sent SIGTERM to process tree"}
         except (ValueError, ProcessLookupError, OSError) as e:
             return {"cluster_name": cluster_name, "status": "stopped", "message": str(e)}
 

api/transformerlab/routers/compute_provider.py

@@ -2523,6 +2523,16 @@ async def stop_cluster(
         user_id_str = str(user_and_team["user"].id)
         provider_instance = await get_provider_instance(provider, user_id=user_id_str, team_id=team_id)
 
+        # Local provider needs workspace_dir (job dir) to stop the correct process tree.
+        # Derive job_id from the standard "-job-<job_id>" suffix in the cluster name.
+        if provider.type == ProviderType.LOCAL.value and hasattr(provider_instance, "extra_config"):
+            job_id_segment = None
+            if "-job-" in cluster_name:
+                job_id_segment = cluster_name.rsplit("-job-", 1)[-1] or None
+            if job_id_segment is not None:
+                job_dir = await asyncio.to_thread(get_local_provider_job_dir, job_id_segment, org_id=team_id)
+                provider_instance.extra_config["workspace_dir"] = job_dir
+
         # Stop cluster
         result = await asyncio.to_thread(provider_instance.stop_cluster, cluster_name)
 

api/transformerlab/routers/experiment/jobs.py

@@ -364,7 +364,7 @@ async def get_provider_job_logs(
 async def get_tunnel_info_for_job(
     experimentId: str,
     job_id: str,
-    tail_lines: int = Query(400, ge=100, le=2000),
+    tail_lines: int = Query(1000, ge=100, le=5000),
     user_and_team=Depends(get_user_and_team),
     session: AsyncSession = Depends(get_async_session),
 ):

knip.json

@@ -0,0 +1,43 @@
+{
+  "$schema": "https://unpkg.com/knip@5/schema.json",
+  "entry": [
+    "src/renderer/index.tsx",
+    "src/renderer/instrument.ts",
+    "src/main/**/*.ts",
+    "src/__tests__/**/*.{ts,tsx}",
+    ".erb/configs/**/*.ts",
+    ".erb/scripts/**/*.ts"
+  ],
+  "project": ["src/**/*.{ts,tsx,js,jsx}"],
+  "ignore": [
+    "api/**",
+    "cli/**",
+    "lab-sdk/**",
+    "docs/**",
+    "scripts/**",
+    "release/**",
+    "build/**",
+    "dist/**",
+    "test/**",
+    "test-results/**",
+    "playwright-report/**",
+    "assets/**",
+    "src/node_modules/**",
+    "**/*.d.ts",
+    "src/renderer/preload.d.ts"
+  ],
+  "ignoreDependencies": [
+    "@playwright/test",
+    "playwright",
+    "electron",
+    "electron-builder",
+    "electron-devtools-installer",
+    "electron-notarize"
+  ],
+  "webpack": {
+    "entry": [".erb/configs/webpack.config.*.ts"]
+  },
+  "jest": {
+    "entry": ["src/__tests__/**/*.{ts,tsx}"]
+  }
+}