Joseph Cool - this is precisely a key part of the motivation behind the NiFi Reg effort and all the work to make extensions sourced from the registry on-demand. As a result we'll no longer package most (perhaps even any) nars with a nifi distribution going forward. We've been steadily making progress on this front. It would be good for you to collaborate in those efforts if you're interested. The driver here is not docker but rather all forms of nifi consumption/deployment models but the docker world would certainly benefit.
Thanks On Fri, Sep 27, 2019 at 12:01 PM Joseph Thweatt <[email protected]> wrote: > tl:dr I made this to make the NiFi docker image smaller, let me know if > it's useful: https://github.com/josephthweatt/skinifi > > Hello everyone, > > My team has been been using NiFi for a couple months now and we have > recently begun adding the docker image into a stack. I noticed that NiFi's > docker image is pretty big (1.91 GB) and takes a while to download as a > result. Adding to that, the image itself is only a part of the multistage > build. After adding custom processors we were actually looking at something > closer to 2.8 GB, making itesting/building/downloading a nightmare. > > The main reason the image is so large is that the lib is full of nar files, > the majority of which are never used. Those same nars are then duplicated > to the work directory when nifi is running, making it even larger. > > My solution to this was the github project above. Basically if you provide > it a Flow from a registry or a template it will find which nars are needed > to run nifi and pack them into a smaller image. A bare-bones image (NiFi > with only the processors needed to run without breaking) is about 680 MB > and in my team's case we got the file size reduced to 1.2 GB. Integration > tests for our stack also halved as a result, from 35 mins down to 17. > > I'll probably continue to add some useful features that I feel we need, but > I figured I should put this out there if anyone else wanted to try it. > > Suggestions are also appreciated! >
