----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/20817/#review42984 -----------------------------------------------------------
src/linux/cgroups.hpp <https://reviews.apache.org/r/20817/#comment76969> After discussed with Ian offline, let's push the timeout to the top level, instead of exposing it in the library function - Jie Yu On May 8, 2014, 7:35 p.m., Ian Downes wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/20817/ > ----------------------------------------------------------- > > (Updated May 8, 2014, 7:35 p.m.) > > > Review request for mesos and Jie Yu. > > > Bugs: MESOS-473 > https://issues.apache.org/jira/browse/MESOS-473 > > > Repository: mesos-git > > > Description > ------- > > The Freezer tries to converge to the "FROZEN" state by repeatedly (every > 100 ms) writing "FROZEN" to the freezer.state control file (up to a > configurable timeout). It assumes there are two possible reasons why a > process does not get frozen during an attempt: > > 1. It was in the middle of being forked and did not receive the signal; > it will receive it at the next attempt. > > 2. It is in uninterruptable sleep ("D" state). Normally, this is from > device I/O or paging and is shortlived, in which case it'll be frozen on > retry. However, processes can get stuck in "D" state, either because of > a device issue, incorrect OOM handling, or kernel bugs. Under this > scenario the correct behavior is to fail after a timeout (defaults to 60 > seconds). > > Freezer functions have been namespaced under cgroups::freezer. > > > Diffs > ----- > > src/linux/cgroups.hpp 5a5735721fb9f051eee661edb08d1cdaa163d0f3 > src/linux/cgroups.cpp 8202c282f580d027a60ded2081962e96e4860f60 > src/tests/cgroups_tests.cpp 6ba9de622953e158feadaa9950618b0b13c9e832 > > Diff: https://reviews.apache.org/r/20817/diff/ > > > Testing > ------- > > make check # Linux > > > Thanks, > > Ian Downes > >
