The UBSAN pre-commit jobs have been running successfully for a while now. I
propose removing them from the pre-commit job to save on build time and
costs. I've setup a separate Jenkins job that runs BE, FE, JBDC, and custom
cluster tests against an UBSAN build, on a daily basis.
On Mon, Nov 12, 20
For sharding in #4, you might be interested in what test-with-docker.py
does. It already has some ASAN support, I think, so maybe adding UBSAN will
be pretty straight-forward. In the screenshot below, you can see that I
sharded CLUSTER_TEST and EE_TEST into two halves. It seems to work ok.
[image
My only concern with this is the unfinished nature of the UBSAN work. Here
are some things left to do:
1. Make UBSAN builds work with FE and JDBC tests. These aren't UBSAN
unclean -- they plain don't work.
2. Make e2e, custom cluster, and BE tests UBSAN clean. This is in progress.
3. Make e2e test
On Mon, Nov 12, 2018 at 1:21 PM Jim Apple wrote:
> I don't think I understand the "one stone" part - are you suggesting that
> we do UBSAN testing within a centos6 container?
>
Exactly. If we're doing multiple builds, we may as well be mutating other
variables to get coverage from them. (Here, y
I don't think I understand the "one stone" part - are you suggesting that
we do UBSAN testing within a centos6 container?
On Mon, Nov 12, 2018 at 1:01 PM Philip Zeyliger wrote:
> Seems useful to me.
>
> If you're interested, we could kill multiple birds with one stone.
> Specifically, I'm also i
Seems useful to me.
If you're interested, we could kill multiple birds with one stone.
Specifically, I'm also interested in centos6/rh6 pre-merge testing. There
are a variety of ways to do so, including running with test-with-docker
stuff. I recognize it's more work, but happy to help if you want
