CBuild/LAVA: Benchmark support setup

[Paul Sokolovsky]

This is carry-over from https://blueprints.launchpad.net/linaro-infrastructure-misc/+spec/finalize-cbuild-lava . To support running benchmarks on LAVA boards, we need to enabled private stream for build results, which is currently blocked on lp:1110713, and setup infra to distribute private benchmarks themselves to the boards.

[Paul Sokolovsky (pfalcon)]

Working on configuring rsyncd to handle benchmark-related config distribution to lava slaves.

[Paul Sokolovsky (pfalcon)]

After studying benchmark implementation in CBuild, I got an idea that it makes sense to use rsyncd not only 1) to distribute build prerequisites to boards; 2) distribute settings to boards; but also to: 3) publish build logs/benchmark results from boards; 4) publish result binaries from boards. Such scheme essentially follows native CBuild way, and bypasses LAVA server for transfer of results completely. The latter seems "not good", but actually it seems to be the most practical way for now (while LAVA has issues with API coverage, and with current aim being just use LAVA as builds backend).

Some difference with native CBuild publishing is that it uses rsync+ssh for transfer, which is considered security issue for LAVA boards, so the idea is to use write-only rsync modules.

[Paul Sokolovsky (pfalcon)]

Created http://cbuild.validation.linaro.org/binaries/lava/ and http://cbuild.validation.linaro.org/benchmarks/lava/ to separate lava results during devel/testing phase.

https://code.launchpad.net/~pfalcon/cbuild/cbuild-init.rsyncd-lava devel branch was setup to capture rsync config.

[Paul Sokolovsky (pfalcon)]

Ok, we have successful benchmark run in LAVA: http://validation.linaro.org/lava-server/scheduler/job/49083 , http://cbuild.validation.linaro.org/benchmarks/gcc-linaro-4.7+bzr115060~christophe-lyon~4.7-backport-vect-cost-model/logs/armv7l-precise-cbuild452-panda-es03-cortexa9hfr1/ . So, LAVA is capable to run CBuild benchmarks from now on. What and how benchmarks is another question - for example, that job ran ~1 benchmark, because it could not find other benchmarks in standard places (standard to CBuild itsels, no changes for LAVA integration to that aspect). But these issues should be apparently resolved on case by case basis (each benchmark has its own licensing restriction, results access restriction, etc.), and should be driven by TCWG.