Start testing Sociomantic ocean in pipeline

[mihails-strasuns]

Recently (sociomantic-tsunami/ocean#321) we have enabled CI testing with vanilla upstream DMD compiler for WIP ocean major branch, as 2.077 provided necessary Throwable.message().

Would appreciate getting it to the list of automatically tested projects ;)

[dlang-bot]

Thanks for your pull request, @mihails-strasuns! We are looking forward to reviewing it, and you should be hearing from a maintainer soon.

Some tips to help speed things up:

• smaller, focused PRs are easier to review than big ones

• try not to mix up refactoring or style changes with bug fixes or feature enhancements

• provide helpful commit messages explaining the rationale behind each change

Bear in mind that large or tricky changes may require multiple rounds of review and revision.

Please see CONTRIBUTING.md for more information.

[mihails-strasuns]

Good, now fails because ocean is built by default with deprecations-as-errors, which, of course, is not suitable for regression testing with upstream. Will require small tweak in our makefile.

[MartinNowak]

Nice, will have a look at pre-installing the dependencies with ansible, assuming that they don't change too often.

[mihails-strasuns]

Nice, will have a look at pre-installing the dependencies with ansible, assuming that they don't change too often.

Yes, I think at this point they change less often than once a year. Alternatively, you can enable docker on the CI server and script can fetch the very same image we use for our own CI tests.

[MartinNowak]

Yes, I think at this point they change less often than once a year. Alternatively, you can enable docker on the CI server and script can fetch the very same image we use for our own CI tests.

That could potentially require nested docker support, a.k.a. root on host, and thus rule out some VM slave options. Wouldn't settle on this easily.

[mihails-strasuns]

No problem, whatever suits you best. Btw there is no need to add workarounds for deprecation bit - I will update the PR as soon as sociomantic-tsunami/ocean#374 is merged which should be pretty soon :)

[mihails-strasuns]

Shouldn't need LDFLAGS here if dependencies are pre-installed

[mihails-strasuns]

Tagged https://github.com/sociomantic-tsunami/ocean/releases/tag/v4.0.0-alpha.1 and pushed commit to switch to it

[MartinNowak]

Thx, I'd prefer help with the failing tests.
https://ci.dlang.io/blue/organizations/jenkins/dlang-org%2Fci/detail/PR-88/8/pipeline/188#step-342-log-368

ocean.core.Test.TestException@/var/lib/jenkins/dlang_projects@2/sociomantic-tsunami/ocean/test/sysstats/main.d(68): expression 'nan >= 0' evaluates to false
----------------
./src/ocean/core/Enforce.d:416 void ocean.core.Enforce.enforceImpl!(">=", float, int).enforceImpl(lazy Exception, float, int, immutable(char)[], int) [0x6281dd]
./src/ocean/core/Enforce.d:323 void ocean.core.Enforce.enforceImpl!(">=", ocean.core.Test.TestException, float, int).enforceImpl(float, int, immutable(char)[], int) [0x628138]
/var/lib/jenkins/dlang_projects@2/sociomantic-tsunami/ocean/test/sysstats/main.d:68 int 

[mihails-strasuns]

Can you run the same pipeline with released 2.077 to see if it is because of compiler or environment?

[mihails-strasuns]

Though environment is more likely - https://github.com/sociomantic-tsunami/ocean/blob/v3.x.x/test/sysstats/main.d#L68

I guess you have virtualized system where CPU numbers are not accessible?

[mihails-strasuns]

Ping @Burgos / @nemanja-boric-sociomantic - I think you are more familiar with stats module than me.

[MartinNowak]

Slaves are using LXD or KVM.

[Burgos]

I wonder if the problem is simply because the timing scheduling and CPU counting is different in LXD/KVM than on the bare metal machine. We'll try adding some busy loop there to spend some time in the CPU instead of yielding the control back.

[MartinNowak]

Stats(nan, nan, nan, 24403968, 8146944, 0.024311)
Reproducibly the uptime diff is 0, while there are as many ticks per second (100), lxcfs rounds them to full seconds.
https://insights.ubuntu.com/2015/03/02/introducing-lxcfs/

[Burgos]

Hm, ok. Then we could just sleep for a bit more than one second/two seconds before getting the results. I don't think this will hurt developers much (and they could always make -j to speed the tests). Alternatively, we could just ignore this test on the platforms that behave like this.

[MartinNowak]

@Burgos see sociomantic-tsunami/ocean#385.

[mihails-strasuns]

Updated to 4.0.0-alpha.2, looking at build log

[mihails-strasuns]

Green! D-YAML has failed though :)

[Burgos]

Such a time to be alive!

[MartinNowak]

Green! D-YAML has failed though :)

Yes, see dlang/dmd#7315 (comment) :/.

[MartinNowak]

Let's hope it doesn't run into any other issues on the KVM host, but since it works on Travis-CI/docker, I wouldn't really expect some.

[mihails-strasuns]

@MartinNowak I see. Do you want me to have a look into Jenkins instance taking over what Travis/CircleCI currently do?

[MartinNowak]

Let's see whether #90 works.

[Burgos]

Am I seeing correctly that merging this PR started failing the master? https://ci.dlang.io/blue/organizations/jenkins/dlang-org%2Fci/detail/master/47/pipeline

[MartinNowak]

Yes @Burgos, looks like it :/.
Maybe there is sth. wrong with the timer wait loop?
https://ci.dlang.io/blue/organizations/jenkins/dlang-org%2Fci/detail/master/47/pipeline/180#step-313-log-369

Currently rerunning https://ci.dlang.io/blue/organizations/jenkins/dlang-org%2Fci/detail/master/48/pipeline.

[Burgos]

The timer loop looks correct to me. I think the problem is that we're checking for equality of the ProcUptime instances: https://github.com/sociomantic-tsunami/ocean/blob/v3.x.x/src/ocean/sys/stats/linux/ProcVFS.d#L209 which includes both uptime and idle but the CPU user calculation uses only a uptime member. https://github.com/sociomantic-tsunami/ocean/blob/v3.x.x/src/ocean/sys/Stats.d#L171

[Burgos]

(So the idle progresses, but not the uptime).

[mihails-strasuns]

I think for now best course of action would be to disable this one specific test:

diff --git a/Build.mak b/Build.mak
index 11eb0f0..f875f76 100644
--- a/Build.mak
+++ b/Build.mak
@@ -20,6 +20,10 @@ clean += .*.lst
 TEST_FILTER_OUT += \
        $C/test/collectd/main.d
 
+# fails with KVM, temporarily disabled
+TEST_FILTER_OUT += \
+       $C/test/sysstats/main.d
+
 $O/test-%: override LDFLAGS += -lebtree
 
 $O/test-filesystemevent: override LDFLAGS += -lrt

It will fix pipeline for now and we will talk about best way to fix actual test later this week among us.

[Burgos]

sociomantic-tsunami/ocean#386

[mihails-strasuns]

#91