I’m taking over some code that uses ray to distribute tasks in a kubernetes cluster. We are having some trouble with the stability.
Since we’re not on the latest version, but are using the following install:
python3 -m pip install https://s3-us-west-2.amazonaws.com/ray-wheels/master/72dbaab0c1e91846b0f5d150f7910b1b679b3ca3/ray-2.0.0.dev0-cp38-cp38-manylinux2014_x86_64.whl, I was thinking we should try out a more recent version.
However, I was wondering if the “official release” is more stable (1.3.0), or perhaps more generally, what is the difference between the two versions 1.3.0 and 2.0.0.dev0? And for a given commit, can I specify either one of the two versions?
(Sorry if the question here is too broad)
Hi @simenandresen,
That looks like the code you inherited follows the Installing from a specific commit installation method. Where 72dbaab0c1e91846b0f5d150f7910b1b679b3ca3 is the git SHA of the specific commit you’re installing from.
IMO, that’s a great solution if you’re in active development mode (also consider the nightlies here) or looking to pull in a specific change or contribute to the project. But for production-grade deployments, you’re probably better off sticking w/ the official releases (currently 1.4.0). In general, those releases will come more battle-tested and be easier to find support on.
The last question I’m not so sure about. It doesn’t appear that every ray-version/os/python-version combination is built at each SHA. That said, most of the newer commits at least appear to produce a ray-2.0.0.dev w/ cp38-cp38 w/ manylinux2014_x86_64 for you to test against. Perhaps someone else can shed some light on how that’s configured within the ci/build system.
w.r.t. to further stability concerns, if you feel like it’s still a problem in 1.4.0, you should raise a GitHub issue detailing the situation.
1 Like
Thanks for your response @paulfryzel
I think I’ll stick to the official release then. Furthermore, I’ll keep looking at the stability and report issues if there’s still some problems.
1 Like