fix: add random suffix to Daytona sandbox names to prevent collisions

Long task names truncated to 40 chars caused identical sandbox names
for different tasks (109 DaytonaErrors in last benchmark run). Adding
a 4-char hex random suffix ensures uniqueness while staying under the
63-char Daytona name limit.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: sjarmak

ccb_harbor/daytona.py

@@ -0,0 +1,89 @@
+from __future__ import annotations
+
+import re
+import secrets
+from pathlib import Path
+
+from harbor.environments.daytona import DaytonaEnvironment
+
+
+def _sanitize_label_value(value: str, max_length: int = 63) -> str:
+    normalized = re.sub(r"[^a-zA-Z0-9_.-]+", "-", value.strip()).strip("-.")
+    if not normalized:
+        return ""
+    return normalized[:max_length]
+
+
+class GuardedDaytonaEnvironment(DaytonaEnvironment):
+    """Repo-local Daytona environment with stable labels and sandbox names."""
+
+    def __init__(
+        self,
+        *args,
+        label_launcher: str | None = None,
+        label_run_id: str | None = None,
+        label_benchmark: str | None = None,
+        label_task_id: str | None = None,
+        label_config: str | None = None,
+        label_job_name: str | None = None,
+        label_category: str | None = None,
+        label_model: str | None = None,
+        label_mcp_type: str | None = None,
+        task_source_dir: str | None = None,
+        **kwargs,
+    ):
+        task_source_path = Path(task_source_dir) if task_source_dir else None
+        derived_task_id = label_task_id or (
+            task_source_path.name if task_source_path and task_source_path.name else None
+        )
+        derived_benchmark = label_benchmark or (
+            task_source_path.parent.name if task_source_path and task_source_path.parent.name else None
+        )
+
+        self._ccb_labels = {
+            "managed_by": "codescalebench",
+            "launcher": label_launcher or "unknown",
+            "run_id": label_run_id or "",
+            "benchmark": derived_benchmark or "",
+            "task_id": derived_task_id or "",
+            "config": label_config or "",
+            "job_name": label_job_name or "",
+            "category": label_category or "",
+            "model": label_model or "",
+            "mcp_type": label_mcp_type or "",
+        }
+        self._ccb_labels = {
+            key: sanitized
+            for key, value in self._ccb_labels.items()
+            if (sanitized := _sanitize_label_value(str(value)))
+        }
+
+        sandbox_name_parts = [
+            self._ccb_labels.get("benchmark", ""),
+            self._ccb_labels.get("task_id", ""),
+            self._ccb_labels.get("config", ""),
+        ]
+        sandbox_name_core = "-".join(part for part in sandbox_name_parts if part)
+        # Reserve 5 chars for random suffix (-xxxx) to prevent collisions
+        # when long task names truncate to the same prefix.
+        sandbox_name_core = _sanitize_label_value(sandbox_name_core, max_length=40)
+        rand_suffix = secrets.token_hex(2)  # 4 hex chars
+        session_hint = kwargs.get("session_id", "")
+        session_hint = _sanitize_label_value(str(session_hint), max_length=12)
+        if sandbox_name_core:
+            self._ccb_sandbox_name = f"ccb-{sandbox_name_core}-{rand_suffix}"
+            if session_hint:
+                self._ccb_sandbox_name = f"{self._ccb_sandbox_name}-{session_hint}"
+        else:
+            self._ccb_sandbox_name = f"ccb-{session_hint or 'sandbox'}-{rand_suffix}"
+        self._ccb_sandbox_name = _sanitize_label_value(self._ccb_sandbox_name, max_length=63)
+
+        super().__init__(*args, **kwargs)
+
+    async def _create_sandbox(self, params):
+        labels = dict(getattr(params, "labels", None) or {})
+        labels.update(self._ccb_labels)
+        params.labels = labels
+        if not getattr(params, "name", None):
+            params.name = self._ccb_sandbox_name
+        await super()._create_sandbox(params)