You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
feat(replay): default to skipping recorded delays on replay
BUFFR_REPLAY_NODELAY now defaults to on, so replays emit chunks/frames
back-to-back instead of re-spending the original generation time. Set
BUFFR_REPLAY_NODELAY=0 to restore the recorded streaming cadence when
the timing itself is under test.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Copy file name to clipboardExpand all lines: README.md
+2-2Lines changed: 2 additions & 2 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -173,9 +173,9 @@ All flags have environment variable equivalents; flags take precedence.
173
173
174
174
<br />
175
175
176
-
buffr records the wall-clock delay before each SSE chunk / WebSocket frame and reproduces that cadence on replay — faithful, but it re-spends the original generation time (often seconds per call) on every run.
176
+
buffr records the wall-clock delay before each SSE chunk / WebSocket frame. By default these delays are dropped on replay so chunks/frames are emitted back-to-back — payloads are identical, only the inter-chunk timing is gone. This keeps replays fast instead of re-spending the original generation time (often seconds per call) on every run.
177
177
178
-
Set `BUFFR_REPLAY_NODELAY=1` to emit all chunks/frames back-to-back. Payloads are identical; only the inter-chunk timing is dropped. Leave it unset when the streaming cadence itself is under test.
178
+
Set `BUFFR_REPLAY_NODELAY=0` to reproduce the recorded cadence faithfully — do this when the streaming timing itself is under test.
0 commit comments