flywheel-bootstrap 0.1.9.202602032022__tar.gz → 0.1.9.202602032202__tar.gz

{flywheel_bootstrap-0.1.9.202602032022 → flywheel_bootstrap-0.1.9.202602032202}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: flywheel-bootstrap
-Version: 0.1.9.202602032022
+Version: 0.1.9.202602032202
 Summary: Bootstrap runner for Flywheel provisioned GPU instances
 Project-URL: Homepage, http://paradigma.inc/
 Author: Paradigma Labs
@@ -81,6 +81,8 @@ a starting point; update the paths and instructions for your machine.
    - Run `codex exec --json --cd <workspace> --skip-git-repo-check flywheel_prompt.txt` with env `FLYWHEEL_RUN_ID/TOKEN/SERVER`.
    - Start a heartbeat thread posting `/runs/{id}/heartbeat` every 30s.
    - Stream Codex stdout lines as logs to `/runs/{id}/logs`; capture Codex `run_id` if emitted.
+   - Oversized log messages are dropped and replaced with a placeholder entry; if a proxy
+     returns `413`, the bootstrap treats it as non-fatal and emits the placeholder.
 9. After Codex exits:
    - Read `flywheel_artifacts.json`; if empty and Codex `run_id` is known, attempt one `codex resume <id>` then re-read.
    - POST artifacts to `/runs/{id}/artifacts`; POST `/complete` on exit 0, else `/error` with the exit code.

{flywheel_bootstrap-0.1.9.202602032022 → flywheel_bootstrap-0.1.9.202602032202}/README.md

@@ -65,6 +65,8 @@ a starting point; update the paths and instructions for your machine.
    - Run `codex exec --json --cd <workspace> --skip-git-repo-check flywheel_prompt.txt` with env `FLYWHEEL_RUN_ID/TOKEN/SERVER`.
    - Start a heartbeat thread posting `/runs/{id}/heartbeat` every 30s.
    - Stream Codex stdout lines as logs to `/runs/{id}/logs`; capture Codex `run_id` if emitted.
+   - Oversized log messages are dropped and replaced with a placeholder entry; if a proxy
+     returns `413`, the bootstrap treats it as non-fatal and emits the placeholder.
 9. After Codex exits:
    - Read `flywheel_artifacts.json`; if empty and Codex `run_id` is known, attempt one `codex resume <id>` then re-read.
    - POST artifacts to `/runs/{id}/artifacts`; POST `/complete` on exit 0, else `/error` with the exit code.

{flywheel_bootstrap-0.1.9.202602032022 → flywheel_bootstrap-0.1.9.202602032202}/bootstrap/telemetry.py

@@ -16,6 +16,8 @@ import sys
 READ_TIMEOUT_SECONDS = 30
 MAX_ATTEMPTS = 3
 BASE_DELAY_SECONDS = 0.5
+MAX_LOG_MESSAGE_BYTES = 25 * 1024 * 1024
+LOG_DROPPED_MESSAGE = "Log entry dropped: too large"
 
 
 def utcnow() -> datetime:
@@ -43,7 +45,7 @@ def post_log(
     extra: Mapping[str, object] | None = None,
 ) -> None:
     """Send a log entry to the backend."""
-    _post(
+    _, status = _post(
         f"{server_url.rstrip('/')}/runs/{run_id}/logs",
         token,
         {
@@ -52,7 +54,29 @@ def post_log(
             "message": message,
             "extra": extra or {},
         },
+        non_retryable_statuses={413},
     )
+    if status == 413:
+        placeholder = {
+            "created_at": utcnow().isoformat(),
+            "level": "warning",
+            "message": LOG_DROPPED_MESSAGE,
+            "extra": {
+                "dropped": True,
+                "drop_reason": "payload_too_large",
+                "message_bytes": len(message.encode("utf-8")),
+                "limit_bytes": MAX_LOG_MESSAGE_BYTES,
+                "original_level": level,
+            },
+        }
+        _post(
+            f"{server_url.rstrip('/')}/runs/{run_id}/logs",
+            token,
+            placeholder,
+            non_retryable_statuses={413},
+            attempts=1,
+            base_delay_seconds=0,
+        )
 
 
 def post_artifacts(
@@ -98,7 +122,15 @@ def post_error(
     )
 
 
-def _post(url: str, token: str, body: Mapping[str, object]) -> None:
+def _post(
+    url: str,
+    token: str,
+    body: Mapping[str, object],
+    *,
+    non_retryable_statuses: set[int] | None = None,
+    attempts: int = MAX_ATTEMPTS,
+    base_delay_seconds: float = BASE_DELAY_SECONDS,
+) -> tuple[bool, int | None]:
     data = json.dumps(body).encode("utf-8")
     req = urllib.request.Request(
         url,
@@ -109,15 +141,19 @@ def _post(url: str, token: str, body: Mapping[str, object]) -> None:
         },
         method="POST",
     )
-    success = _urlopen_with_retries(
+    success, status = _urlopen_with_retries(
         req,
         timeout_seconds=READ_TIMEOUT_SECONDS,
-        attempts=MAX_ATTEMPTS,
-        base_delay_seconds=BASE_DELAY_SECONDS,
+        attempts=attempts,
+        base_delay_seconds=base_delay_seconds,
+        non_retryable_statuses=non_retryable_statuses,
     )
-    if not success:
+    if not success and (
+        not non_retryable_statuses or status not in non_retryable_statuses
+    ):
         # Best-effort: log locally but do not raise, to avoid failing the run on transient stalls.
         print(f"telemetry POST to {url} failed after retries", file=sys.stderr)
+    return success, status
 
 
 def _urlopen_with_retries(
@@ -125,7 +161,8 @@ def _urlopen_with_retries(
     timeout_seconds: int,
     attempts: int,
     base_delay_seconds: float,
-) -> bool:
+    non_retryable_statuses: set[int] | None = None,
+) -> tuple[bool, int | None]:
     """Best-effort POST with retries for transient network errors."""
     for i in range(attempts):
         try:
@@ -133,7 +170,14 @@ def _urlopen_with_retries(
                 req, timeout=timeout_seconds
             ) as resp:  # pragma: no cover - network
                 resp.read()
-            return True
+            return True, getattr(resp, "status", None)
+        except urllib.error.HTTPError as exc:
+            if non_retryable_statuses and exc.code in non_retryable_statuses:
+                return False, exc.code
+            if i == attempts - 1:
+                return False, exc.code
+            delay = base_delay_seconds * (2**i)
+            time.sleep(delay)
         except (
             TimeoutError,
             urllib.error.URLError,
@@ -144,4 +188,4 @@ def _urlopen_with_retries(
                 break
             delay = base_delay_seconds * (2**i)
             time.sleep(delay)
-    return False
+    return False, None

{flywheel_bootstrap-0.1.9.202602032022 → flywheel_bootstrap-0.1.9.202602032202}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "flywheel-bootstrap"
-version = "0.1.9.202602032022"
+version = "0.1.9.202602032202"
 description = "Bootstrap runner for Flywheel provisioned GPU instances"
 readme = "README.md"
 requires-python = ">=3.11"

flywheel_bootstrap-0.1.9.202602032202/tests/test_telemetry.py

@@ -0,0 +1,64 @@
+from __future__ import annotations
+
+from bootstrap import telemetry
+
+
+def test_post_does_not_warn_on_non_retryable_413(monkeypatch, capsys) -> None:
+    def fake_urlopen_with_retries(
+        req,
+        timeout_seconds: int,
+        attempts: int,
+        base_delay_seconds: float,
+        non_retryable_statuses: set[int] | None = None,
+    ) -> tuple[bool, int | None]:
+        return False, 413
+
+    monkeypatch.setattr(telemetry, "_urlopen_with_retries", fake_urlopen_with_retries)
+
+    success, status = telemetry._post(
+        "http://example.test/runs/logs",
+        "token",
+        {"message": "hello"},
+        non_retryable_statuses={413},
+    )
+
+    captured = capsys.readouterr()
+    assert success is False
+    assert status == 413
+    assert captured.err == ""
+
+
+def test_post_log_sends_placeholder_on_413(monkeypatch) -> None:
+    calls: list[tuple[str, str, dict, dict]] = []
+
+    def fake_post(url: str, token: str, body: dict, **kwargs):
+        calls.append((url, token, body, kwargs))
+        if len(calls) == 1:
+            return False, 413
+        return True, 202
+
+    monkeypatch.setattr(telemetry, "_post", fake_post)
+
+    telemetry.post_log(
+        "http://example.test",
+        "run-123",
+        "token",
+        "info",
+        "hello",
+        {"foo": "bar"},
+    )
+
+    assert len(calls) == 2
+    assert calls[0][3]["non_retryable_statuses"] == {413}
+
+    placeholder = calls[1][2]
+    assert placeholder["message"] == telemetry.LOG_DROPPED_MESSAGE
+    assert placeholder["level"] == "warning"
+    extra = placeholder["extra"]
+    assert extra["dropped"] is True
+    assert extra["drop_reason"] == "payload_too_large"
+    assert extra["original_level"] == "info"
+    assert extra["message_bytes"] == len("hello".encode("utf-8"))
+    assert extra["limit_bytes"] == telemetry.MAX_LOG_MESSAGE_BYTES
+    assert calls[1][3]["attempts"] == 1
+    assert calls[1][3]["base_delay_seconds"] == 0