tree e9e445e1405ea4f838e107bee6b6e1ba9dd8d83f
parent baec8cae34591c92c9309333280f85f81d852f8d
author Todd Lipcon <todd@apache.org> 1510815344 -0800
committer Impala Public Jenkins <impala-public-jenkins@gerrit.cloudera.org> 1519187635 +0000

KUDU-2218. tls_socket: properly handle temporary socket errors in Writev

This fixes a bug which caused RaftConsensusITest.TestLargeBatches to
fail when run under stress, as in the following command line:

taskset -c 0-4 \
 build/latest/bin/raft_consensus-itest \
   --gtest_filter=\*LargeBat\* \
   --stress-cpu-threads=8

This would produce an error like:
Network error: failed to write to TLS socket: error:1409F07F:SSL routines:SSL3_WRITE_PENDING:bad write retry:s3_pkt.c:878

This means that we were retrying a write after getting EAGAIN, but with
a different buffer than the first time.

I tracked this down to mishandling of temporary socket errors in
TlsSocket::Writev(). In the case that we successfully write part of the
io vector but hit such an error trying to write a later element in the
vector, we were still propagating the error back up to the caller. The
caller didn't realize that part of the write was successful, and thus it
would retry the write from the beginning.

The fix is to fix the above, but also to enable partial writes in
TlsContext. The new test fails if either of the above two changes are
backed out.

Change-Id: If797f220f42bfb2e6f452b66f15e7a758e883472
Reviewed-on: http://gerrit.cloudera.org:8080/8570
Tested-by: Kudu Jenkins
Reviewed-by: Alexey Serbin <aserbin@cloudera.com>
Reviewed-on: http://gerrit.cloudera.org:8080/9361
Reviewed-by: Michael Ho <kwho@cloudera.com>
Tested-by: Impala Public Jenkins