⚡ Bolt: [대규모 로그 파싱 최적화]

.jules/bolt.md

@@ -60,3 +60,6 @@
 ## 2026-06-25 - [Optimize Path.exists() when paired with stat()]
 **Learning:** Checking `Path.exists()` before `Path.stat()` introduces a redundant system call because `exists()` internally uses `stat()`.
 **Action:** Rely on catching the `OSError` from `Path.stat()` to simultaneously check for existence and retrieve file attributes, saving measurable I/O overhead on large filesystems.
+## 2024-11-20 - [Optimize massive log parsing with regex finditer]
+**Learning:** When parsing large text logs from command-line tools like `ffmpeg` (which use carriage returns `\r` for progress updates), using `str.splitlines()` causes massive memory allocations and high CPU usage.
+**Action:** Compile a unified regex and use `re.finditer()` directly on the raw output string for safe, memory-efficient parsing without allocating new strings for each line.

media_shrinker.py

@@ -84,8 +84,9 @@
 BITRATE_SAFETY_MARGIN = 0.92
 OPUS_MAX_BITRATE_BPS = 510_000
 OPUS_MIN_REASONABLE_BITRATE_BPS = 16_000
-SILENCE_START_RE = re.compile(r"silence_start:\s*(?P<value>[0-9]+(?:\.[0-9]+)?)")
-SILENCE_END_RE = re.compile(r"silence_end:\s*(?P<value>[0-9]+(?:\.[0-9]+)?)")
+SILENCE_EVENT_RE = re.compile(
+    r"silence_(?P<type>start|end):\s*(?P<value>[0-9]+(?:\.[0-9]+)?)"
+)
 
 
 class MediaShrinkerError(RuntimeError):
@@ -544,23 +545,19 @@ def parse_silencedetect_intervals(stderr: str) -> list[SilenceInterval]:
 
     intervals: list[SilenceInterval] = []
     current_start: float | None = None
-    for line in stderr.splitlines():
-        # Fast path: Substring search is much faster than regex.
-        # Most lines are ffmpeg progress updates (e.g., 'frame=...')
-        if "silence" not in line:
-            continue
-        start_match = SILENCE_START_RE.search(line)
-        if start_match:
-            current_start = float(start_match.group("value"))
-            continue
 
-        end_match = SILENCE_END_RE.search(line)
-        if end_match and current_start is not None:
-            end_seconds = float(end_match.group("value"))
-            if end_seconds > current_start:
+    # Using re.finditer directly on the raw string avoids creating massive
+    # lists of strings via splitlines() when parsing dense ffmpeg output.
+    for match in SILENCE_EVENT_RE.finditer(stderr):
+        event_type = match.group("type")
+        value = float(match.group("value"))
+        if event_type == "start":
+            current_start = value
+        elif event_type == "end" and current_start is not None:
+            if value > current_start:
                 intervals.append(
                     SilenceInterval(
-                        start_seconds=current_start, end_seconds=end_seconds
+                        start_seconds=current_start, end_seconds=value
                     )
                 )
             current_start = None