Cloning our largest internal repo over SSH intermittently fails during pack negotiation. The client sits idle for a while, then the connection drops with fatal: the remote end hung up unexpectedly.
It only reproduces on repos with a big initial pack — small repos clone fine every time. It looks like the SSH session times out before the server finishes streaming the pack.
Stack: Cloudzilla on the staging node, default config; client is plain OpenSSH.
Reproduced. The SSH channel write was blocking on a buffer that never flushed for packs over ~40 MB, so the write deadline fired mid-stream and tore down the session.
The fix flushes in bounded chunks and resets the write deadline per chunk. Patch is up for review.
Pulled the fix onto staging — the large repo clones cleanly now, three runs in a row. Thanks for the quick turnaround.
malith closed this in #137