Avoid deleting entire cache for missing shards (#14754 follow-up)

.github/workflows/nightly-test-nvidia.yml

@@ -115,6 +115,25 @@ jobs:
           cd test
           python3 run_suite.py --hw cuda --suite nightly-8-gpu-h200 --nightly --continue-on-error
 
+      - name: Run MiniMax-M2 nightly performance test
+        timeout-minutes: 180
+        env:
+          TRACE_BASE_URL: https://raw.githubusercontent.com/sglang-bot/sglang-ci-data/main/traces/${{ github.run_id }}
+          PERFETTO_RELAY_URL: ${{ vars.PERFETTO_RELAY_URL }}
+          GPU_CONFIG: "8-gpu-h200"
+        run: |
+          rm -rf test/performance_profiles_minimax_m2/
+          cd test
+          python3 nightly/test_minimax_m2_perf.py
+
+      - name: Publish MiniMax-M2 traces to storage repo
+        env:
+          GITHUB_TOKEN: ${{ secrets.GH_PAT_FOR_NIGHTLY_CI_DATA }}
+          GITHUB_RUN_ID: ${{ github.run_id }}
+          GITHUB_RUN_NUMBER: ${{ github.run_number }}
+        run: |
+          python3 scripts/ci/publish_traces.py --traces-dir test/performance_profiles_minimax_m2
+
       - name: Run Qwen3-235B nightly performance test
         timeout-minutes: 180
         env:
@@ -172,25 +191,6 @@ jobs:
         run: |
           python3 scripts/ci/publish_traces.py --traces-dir test/performance_profiles_glm_4_6
 
-      - name: Run MiniMax-M2 nightly performance test
-        timeout-minutes: 180
-        env:
-          TRACE_BASE_URL: https://raw.githubusercontent.com/sglang-bot/sglang-ci-data/main/traces/${{ github.run_id }}
-          PERFETTO_RELAY_URL: ${{ vars.PERFETTO_RELAY_URL }}
-          GPU_CONFIG: "8-gpu-h200"
-        run: |
-          rm -rf test/performance_profiles_minimax_m2/
-          cd test
-          python3 nightly/test_minimax_m2_perf.py
-
-      - name: Publish MiniMax-M2 traces to storage repo
-        env:
-          GITHUB_TOKEN: ${{ secrets.GH_PAT_FOR_NIGHTLY_CI_DATA }}
-          GITHUB_RUN_ID: ${{ github.run_id }}
-          GITHUB_RUN_NUMBER: ${{ github.run_number }}
-        run: |
-          python3 scripts/ci/publish_traces.py --traces-dir test/performance_profiles_minimax_m2
-
   # General tests - 8 GPU H20
   nightly-test-general-8-gpu-h20:
     if: github.repository == 'sgl-project/sglang' && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-test-general-8-gpu-h20')

python/sglang/srt/model_loader/weight_utils.py

@@ -393,14 +393,17 @@ def _find_local_hf_snapshot_dir_unlocked(
                     )
                     return None
                 else:
-                    # Cannot selectively clean (e.g., missing shards) - remove entire cache
+                    # Missing shards (not corruption) - let snapshot_download handle it.
+                    # IMPORTANT: Do NOT delete the entire cache here, as other processes
+                    # (TP/EP ranks) may already be loading weights from these files.
+                    # Deleting the cache while other processes are using it causes
+                    # FileNotFoundError race conditions. Instead, just return None
+                    # to trigger a download - snapshot_download will only fetch
+                    # missing files without disturbing existing ones.
                     log_info_on_rank0(
                         logger,
                         f"Validation failed for {model_name_or_path}: {error_msg}. "
-                        "Will remove entire cache and re-download.",
-                    )
-                    _cleanup_corrupted_model_cache(
-                        model_name_or_path, found_local_snapshot_dir, error_msg
+                        "Will attempt to download missing files.",
                     )
                     return None
 

python/sglang/srt/model_loader/weight_validation.py

@@ -61,6 +61,26 @@ def _check_index_files_exist(snapshot_dir: str) -> Tuple[bool, Optional[str]]:
 
     for index_file in index_files:
         index_path = os.path.join(snapshot_dir, index_file)
+
+        # Check if index file is a broken symlink (exists in listing but blob missing)
+        if os.path.islink(index_path) and not os.path.exists(index_path):
+            # Broken symlink - clean it up so download can proceed
+            try:
+                blob_path = os.path.realpath(index_path)
+                os.remove(index_path)
+                logger.warning(
+                    "Removed broken index symlink: %s (blob missing)", index_file
+                )
+                # Also try to remove dangling blob reference if it somehow exists
+                if os.path.exists(blob_path):
+                    os.remove(blob_path)
+            except Exception as e:
+                logger.error("Failed to remove broken symlink %s: %s", index_file, e)
+            return (
+                False,
+                f"Broken index file symlink: {index_file} (cleaned up, will re-download)",
+            )
+
         try:
             with open(index_path) as f:
                 index_data = json.load(f)
@@ -85,6 +105,13 @@ def _check_index_files_exist(snapshot_dir: str) -> Tuple[bool, Optional[str]]:
                     f"Missing {len(missing_files)} file(s) from index {index_file}: {missing_files[:3]}{'...' if len(missing_files) > 3 else ''}",
                 )
 
+        except FileNotFoundError as e:
+            # Index file was listed but can't be read - could be race condition or broken state
+            logger.warning("Failed to read index file %s: %s", index_file, e)
+            return (
+                False,
+                f"Index file {index_file} unreadable (will re-download)",
+            )
         except Exception as e:
             logger.warning("Failed to read index file %s: %s", index_file, e)
             continue

test/manual/test_weight_validation.py

@@ -0,0 +1,186 @@
+"""
+Unit tests for weight validation and cache cleanup logic.
+
+Tests the fix for issue #14754 - ensuring that missing shards do not trigger
+entire cache deletion, which can cause race conditions in multi-process scenarios.
+"""
+
+import json
+import os
+import struct
+import tempfile
+import unittest
+
+from sglang.srt.model_loader.weight_validation import (
+    _check_index_files_exist,
+    _validate_sharded_model,
+)
+
+
+class TestWeightValidation(unittest.TestCase):
+    """Tests for weight validation functions."""
+
+    def test_validate_sharded_model_missing_shard(self):
+        """
+        Test that missing shards are detected correctly.
+
+        This is the core test for issue #14754 fix: when a shard is missing,
+        the validation should return is_valid=False with an error message
+        containing "Missing", but corrupted_files should be empty (indicating
+        this is a missing shard issue, not a corruption issue).
+
+        This distinction is critical because:
+        - Missing shards: should NOT delete cache (other processes may be using it)
+        - Corrupted files: should delete only the corrupted files selectively
+        """
+        with tempfile.TemporaryDirectory() as tmpdir:
+            # Create partial shards (missing shard 3)
+            for i in [1, 2]:  # Missing shard 3
+                open(
+                    os.path.join(tmpdir, f"model-0000{i}-of-00003.safetensors"), "w"
+                ).close()
+
+            # Create index file
+            index_data = {
+                "weight_map": {
+                    "layer1": "model-00001-of-00003.safetensors",
+                    "layer2": "model-00002-of-00003.safetensors",
+                    "layer3": "model-00003-of-00003.safetensors",
+                }
+            }
+            with open(os.path.join(tmpdir, "model.safetensors.index.json"), "w") as f:
+                json.dump(index_data, f)
+
+            weight_files = [
+                os.path.join(tmpdir, f"model-0000{i}-of-00003.safetensors")
+                for i in [1, 2]
+            ]
+
+            is_valid, error_msg, corrupted_files = _validate_sharded_model(
+                tmpdir, weight_files
+            )
+
+            self.assertFalse(is_valid)
+            self.assertIn("Missing", error_msg)
+            # CRITICAL: corrupted_files should be empty for missing shards
+            # This is what prevents entire cache deletion
+            self.assertEqual(corrupted_files, [])
+
+    def test_validate_sharded_model_all_present(self):
+        """Test that complete shards pass validation."""
+        with tempfile.TemporaryDirectory() as tmpdir:
+            # Create all shards with valid safetensors header
+            for i in [1, 2, 3]:
+                filepath = os.path.join(tmpdir, f"model-0000{i}-of-00003.safetensors")
+                # Create a minimal valid safetensors file
+                # Header: 8 bytes for header size + JSON header
+                header = b'{"__metadata__":{}}'
+                header_size = len(header)
+                with open(filepath, "wb") as f:
+                    f.write(struct.pack("<Q", header_size))
+                    f.write(header)
+
+            # Create index file
+            index_data = {
+                "weight_map": {
+                    "layer1": "model-00001-of-00003.safetensors",
+                    "layer2": "model-00002-of-00003.safetensors",
+                    "layer3": "model-00003-of-00003.safetensors",
+                }
+            }
+            with open(os.path.join(tmpdir, "model.safetensors.index.json"), "w") as f:
+                json.dump(index_data, f)
+
+            weight_files = [
+                os.path.join(tmpdir, f"model-0000{i}-of-00003.safetensors")
+                for i in [1, 2, 3]
+            ]
+
+            is_valid, error_msg, corrupted_files = _validate_sharded_model(
+                tmpdir, weight_files
+            )
+
+            self.assertTrue(is_valid)
+            self.assertIsNone(error_msg)
+            self.assertEqual(corrupted_files, [])
+
+    def test_validate_sharded_model_corrupted_shard(self):
+        """
+        Test that corrupted shards are detected and returned in corrupted_files.
+
+        This tests the other branch: when a file exists but is corrupted
+        (invalid safetensors format), it should be added to corrupted_files
+        so that selective cleanup can remove just that file.
+        """
+        with tempfile.TemporaryDirectory() as tmpdir:
+            # Create shard 1 as valid
+            filepath1 = os.path.join(tmpdir, "model-00001-of-00003.safetensors")
+            header = b'{"__metadata__":{}}'
+            with open(filepath1, "wb") as f:
+                f.write(struct.pack("<Q", len(header)))
+                f.write(header)
+
+            # Create shard 2 as corrupted (invalid header)
+            filepath2 = os.path.join(tmpdir, "model-00002-of-00003.safetensors")
+            with open(filepath2, "wb") as f:
+                f.write(b"invalid data that is not a valid safetensors file")
+
+            # Create shard 3 as valid
+            filepath3 = os.path.join(tmpdir, "model-00003-of-00003.safetensors")
+            with open(filepath3, "wb") as f:
+                f.write(struct.pack("<Q", len(header)))
+                f.write(header)
+
+            # Create index file
+            index_data = {
+                "weight_map": {
+                    "layer1": "model-00001-of-00003.safetensors",
+                    "layer2": "model-00002-of-00003.safetensors",
+                    "layer3": "model-00003-of-00003.safetensors",
+                }
+            }
+            with open(os.path.join(tmpdir, "model.safetensors.index.json"), "w") as f:
+                json.dump(index_data, f)
+
+            weight_files = [filepath1, filepath2, filepath3]
+
+            is_valid, error_msg, corrupted_files = _validate_sharded_model(
+                tmpdir, weight_files
+            )
+
+            self.assertFalse(is_valid)
+            self.assertIn("Corrupt", error_msg)
+            # The corrupted file should be identified
+            self.assertEqual(len(corrupted_files), 1)
+            self.assertIn("model-00002-of-00003.safetensors", corrupted_files[0])
+
+    def test_broken_index_symlink_detected(self):
+        """
+        Test that broken index symlinks are detected and cause validation to fail.
+
+        When an index file is a symlink pointing to a non-existent blob,
+        validation should fail (to trigger re-download) rather than silently
+        continuing and causing timeout during actual loading.
+        """
+        with tempfile.TemporaryDirectory() as tmpdir:
+            # Create a broken symlink for the index file
+            index_path = os.path.join(tmpdir, "model.safetensors.index.json")
+            non_existent_blob = os.path.join(tmpdir, "blobs", "nonexistent_hash")
+            os.symlink(non_existent_blob, index_path)
+
+            # Verify it's a broken symlink
+            self.assertTrue(os.path.islink(index_path))
+            self.assertFalse(os.path.exists(index_path))
+
+            # Check should fail for broken symlink
+            is_valid, error_msg = _check_index_files_exist(tmpdir)
+
+            self.assertFalse(is_valid)
+            self.assertIn("Broken", error_msg)
+            # The broken symlink should have been cleaned up
+            self.assertFalse(os.path.exists(index_path))
+            self.assertFalse(os.path.islink(index_path))
+
+
+if __name__ == "__main__":
+    unittest.main()