----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/61664/#review183002 -----------------------------------------------------------
3rdparty/libprocess/src/process.cpp Lines 256-261 (patched) <https://reviews.apache.org/r/61664/#comment258985> How did you pick these values? 3rdparty/libprocess/src/process.cpp Lines 2184 (patched) <https://reviews.apache.org/r/61664/#comment258986> s/, see/; see/ 3rdparty/libprocess/src/process.cpp Lines 2170-2182 (original), 2195-2209 (patched) <https://reviews.apache.org/r/61664/#comment258989> I guess this means if writing a very large message takes greater than the timeout we abort? Do you have a rought back of the envelope calculation for what the max data size could be given the minimum timeout? - Vinod Kone On Aug. 15, 2017, 3:45 p.m., Alexander Rukletsov wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/61664/ > ----------------------------------------------------------- > > (Updated Aug. 15, 2017, 3:45 p.m.) > > > Review request for mesos, Benno Evers, Benjamin Mahler, Jie Yu, and Vinod > Kone. > > > Bugs: MESOS-7748 > https://issues.apache.org/jira/browse/MESOS-7748 > > > Repository: mesos > > > Description > ------- > > Prior to this patch, a send socket operation can wait forever for > a send to complete. Clients that drop connections or stop reading > incoming data, aka "slow reader" attack, can eventually exhaust the > resources of a libprocess-based application and cause denial of > service or an OOM event. > > This patch introduces an obligatory timeout for all send socket > operations, after which the stalled connection is dropped. The > timeout can be adjusted via the `LIBPROCESS_SOCKET_SEND_TIMEOUT` > environment variable. > > > Diffs > ----- > > 3rdparty/libprocess/src/process.cpp > dcd9c6738816764aae066fe56cd5f468c98fc9bd > > > Diff: https://reviews.apache.org/r/61664/diff/1/ > > > Testing > ------- > > Manual testing with a rogue client. > > > Thanks, > > Alexander Rukletsov > >
