Comment 11 for bug 1568650

Reviewed: https://review.opendev.org/690644
Committed: https://git.openstack.org/cgit/openstack/swift/commit/?id=e3739920c4db97f255ab973cf125f0091bb15428
Submitter: Zuul
Branch: stable/rocky

commit e3739920c4db97f255ab973cf125f0091bb15428
Author: Samuel Merritt <email address hidden>
Date: Wed Jun 13 14:28:28 2018 -0700

    Fix socket leak on object-server death

    Consider a client that's downloading a large replicated object of size
    N bytes. If the object server process dies (e.g. with a segfault)
    partway through the download, the proxy will have read fewer than N
    bytes, and then read(sockfd) will start returning 0 bytes. At this
    point, the proxy believes the object download is complete, and so the
    WSGI server waits for a new request to come in. Meanwhile, the client
    is waiting for the rest of their bytes. Until the client times out,
    that socket will be held open.

    The fix is to look at the Content-Length and Content-Range headers in
    the response from the object server, then retry with another object
    server in case the original GET is truncated. This way, the client
    gets all the bytes they should.

    Note that ResumingGetter already had retry logic for when an
    object-server is slow to send bytes -- this extends it to also cover
    unexpected disconnects.

    Change-Id: Iab1e07706193ddc86832fd2cff0d7c2cb6d79ad9
    Related-Change: I74d8c13eba2a4917b5a116875b51a781b33a7abf
    Closes-Bug: 1568650
    (cherry picked from commit 0e81ffd1e1481a73146fce17f61f2ab9e01eb684)