I've checked and we are no longer relying on the previous LatestOnlyOperator behavior for any of our DAGs.
This is not a dealbreaker (though I will need to keep it in mind). Thanks for asking, --George On Tue, Jul 25, 2017 at 2:50 PM Maxime Beauchemin < [email protected]> wrote: > [AIRFLOW-1296] is part of 1.8.2. > > Is this a dealbreaker for 1.8.2? > > Max > > On Tue, Jul 25, 2017 at 2:40 PM, George Leslie-Waksman < > [email protected]> wrote: > > > I hope that it's not too late for me to chime in but there is a breaking > > change in the behavior of LatestOnlyOperator. > > > > The change was introduced in > > PR: https://github.com/apache/incubator-airflow/pull/2365 > > Change: 333e0b3 [AIRFLOW-1296] Propagate SKIPPED to all downstream tasks > > > > Prior to this change, the LatestOnlyOperator would skip direct downstream > > but not indirect downstream; now it skips indirect downstream. > > > > This breaks the use of LatestOnlyOperator with TriggerRules that do not > > propagate skips. > > > > --George > > > > On Thu, Jul 20, 2017 at 10:08 PM Maxime Beauchemin < > > [email protected]> wrote: > > > > > I think I'm gathering a good picture of what is expected here. I'll try > > to > > > update the Confluence page as I go. > > > > > > I'm hoping to get started tomorrow and package it early next week. > > > > > > Max > > > > > > On Wed, Jul 19, 2017 at 6:16 PM, siddharth anand <[email protected]> > > > wrote: > > > > > > > FYI, can anyone pictorially describe the release process (and post it > > on > > > > the apache airflow wiki)? I think that would eliminate a lot of > > confusion > > > > in the future and avoid a rehash of this email thread on the next > > > release. > > > > > > > > -s > > > > > > > > On Wed, Jul 19, 2017 at 10:48 AM, Hitesh Shah <[email protected]> > > wrote: > > > > > > > > > To add, the main source tarball should have instructions to > generate > > > the > > > > > sdist and bdist versions. Additionally, as part of the release > > process > > > if > > > > > the plan is to publish to pypi (after the IPMC vote succeeds), then > > the > > > > > appropriate bits also need to be verified/voted upon. There are not > > > > exactly > > > > > counted as the official release bits but they do need to be > verified > > as > > > > > part of the voting process to ensure that the bits do indeed map to > > the > > > > > source release, license/notice files are correct, etc. > > > > > > > > > > thanks > > > > > -- Hitesh > > > > > > > > > > > > > > > On Tue, Jul 18, 2017 at 12:01 AM, Bolke de Bruin < > [email protected]> > > > > > wrote: > > > > > > > > > > > Thanks Hitesh. We discussed it with John Ament on the IPMC. > Python > > > has > > > > > the > > > > > > notion of 3 types of distributions, “source”, “sdist”, “bdist”, > > > > contrary > > > > > to > > > > > > Java that knows only two (source, bdist). We used to vote on > > “sdist”, > > > > > which > > > > > > was deemed incorrect. > > > > > > > > > > > > So, Max, indeed we need to vote on a tar.gz that contains build > > > > > > instructions in INSTALL to get to “sdist”. The build instructions > > > > should > > > > > > also contain instruction how to run the license checks by Apache > > Rat. > > > > > Most > > > > > > of the work probably goes in the build instructions and verifying > > > they > > > > > > work, but it should not be much. > > > > > > > > > > > > Any other clarification required? > > > > > > > > > > > > Bolke > > > > > > > > > > > > > > > > > > > > > > > > > > >
