Previously, retries happened as fast as possible unless the server
provided the Retry-After header. This is effective for certain types of
errors, but not when the LFS server is experiencing a temporary but not
instantaneous failure. Delaying between retries lets the server recover
and the LFS operation complete.
Delays start at a fixed 250ms for the first retry and double with each
successive retry up to a configurable maximum delay, 10s by default. The
maximum retry is configurable using lfs.transfer.maxretrydelay. Delays
can be disabled by setting the max delay to 0.
cf022164c6