> On Feb. 14, 2013, 1:03 a.m., David Mackey wrote: > > Just a comment on what cpuacct.stat includes and doesn't include. > > > > user is user time + nice time > > system is system time + irq time + softirq time > > > > cpuacct.stat does not include idle, iowait, guest, guest nice and steal > > time. IDLE and IOWAIT are essentially global stats that are ill defined in > > a cgroup context and the others only really apply to virtual cpus. > > > > So, yes, looks good. > > > > Ben Mahler wrote: > Thanks for the info! > > What about cpuacct.usage, and cpuacct.usage_percpu, do you know what > those include? Why would they be different?
cpuacct.usage is essentially a for loop summing the individual values in cpuacct.usage_percpu so these two are equivalent in what they measure only at a different granularity. The difference between .usage and .stat is largely in the units they are measuring. .usage is measuring the wall clock nanoseconds whereas .stat is measuring the cpu cycles consumed. - David ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/9145/#review16544 ----------------------------------------------------------- On Feb. 13, 2013, 9:48 p.m., Ben Mahler wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/9145/ > ----------------------------------------------------------- > > (Updated Feb. 13, 2013, 9:48 p.m.) > > > Review request for mesos, Benjamin Hindman and Vinod Kone. > > > Description > ------- > > This implements resource collection for the cgroups isolation module. > > From the redhat documentation: > https://access.redhat.com/knowledge/docs/en-US/Red_Hat_Enterprise_Linux/6/html/Resource_Management_Guide/sec-cpuacct.html > > // cpuacct.usage > // reports the total CPU time (in nanoseconds) consumed by all tasks in this > cgroup (including tasks lower in the hierarchy). > I don't like this control because it can be reset back to zero! > > // cpuacct.stat > // reports the user and system CPU time consumed by all tasks in this cgroup > (including tasks lower in the hierarchy) in the following way: > // user — CPU time consumed by tasks in user mode. > // system — CPU time consumed by tasks in system (kernel) mode. > // CPU time is reported in the units defined by the USER_HZ variable. > Since USER_HZ is typically 100, the granularity here is only 10 ms. > > // cpuacct.usage_percpu > // reports the CPU time (in nanoseconds) consumed on each CPU by all tasks in > this cgroup (including tasks lower in the hierarchy). > I don't like this control because it can be reset back to zero! > > I've used cpuacct.stat since AFAICT it can't be reset to 0. > However cpuacct.stat has somewhat low granularity, see the testing comments > below. > > > This addresses bug MESOS-324. > https://issues.apache.org/jira/browse/MESOS-324 > > > Diffs > ----- > > src/linux/cgroups.hpp 1f701f3bbbe06ddf84768c68b529aba4659c19be > src/linux/cgroups.cpp 03b31e7309b9dd65f00d3b0da2abb81ddaaeea43 > src/slave/cgroups_isolation_module.cpp > 63cefc33cf34eebb82db5d8448b751be8652fa36 > src/tests/cgroups_tests.cpp b219906374764e91f1a5268469ae92dd0fe08e53 > > Diff: https://reviews.apache.org/r/9145/diff/ > > > Testing > ------- > > Added tests for cgroups::stat. > > End to end testing using the webui. > > NOTES for cpuacct.stat: > $ cat > /cgroup/mesos/framework_201302132039-2081170186-5050-60933-0001_executor_default_tag_3e1f5310-c873-42cb-9aa4-4ee4c2b9feb8/cpuacct.usage > 4672471833 > --> 4672471833ns = 4.67 seconds > > $ cat > /cgroup/mesos/framework_201302132039-2081170186-5050-60933-0001_executor_default_tag_3e1f5310-c873-42cb-9aa4-4ee4c2b9feb8/cpuacct.usage_percpu > > 831220060 463800214 319016010 184325849 840595741 441855678 294660045 > 160799890 240361561 197829862 130045719 56978804 227972655 193743493 98604097 > 70557562 > --> > 831220060+463800214+319016010+184325849+840595741+441855678+294660045+160799890+240361561+197829862+130045719+56978804+227972655+193743493+98604097+70557562 > = 4752367240ns = 4.75 seconds > > $ cat > /cgroup/mesos/framework_201302132039-2081170186-5050-60933-0001_executor_default_tag_3e1f5310-c873-42cb-9aa4-4ee4c2b9feb8/cpuacct.stat > user 111 > system 246 > --> 1.11 + 2.46 = 3.57 seconds > > So since cpuacct.stat reveals only the user + system times, we see slightly > lower times than where the total time is displayed. I'm guessing they may be > including other cpu times? > E.g. steal, guest > > I think user + system is a good measurement. > > > Thanks, > > Ben Mahler > >
