We have invested time into Angular 2 and leveraged a fair amount of its features and ecosystem, but implementation specifics on some have been a great challenge.
We watched the Google I/O presentation from 2016 where the team pitched a lot of exciting stuff coming to angular like SSR/Universal rendering, and decided to buy in and use ng2, but using universal with the CLI was near impossible to get done given how undocumented it was. There was a CLI fork that was quite promising, and with a bit of work we managed to get client side AOT optimization and SSR.
Bundle sizes after all this were unacceptably high. The documentation and talks had discussed things like tree shaking and other such optimization processes, but there was a tussle between webpack and rollup where the CLI used one and not the other, and rollup did tree shaking better, so we had to either drop tree shaking or the CLI. It was not a fun experience.
Eventually angular 4 dropped and there was a timeline on getting the CLI up to speed with universal, the guys at Google I/O 2017 did a talk on universal rendering, but they used some cryptic command line tools that were also fairly undocumented at the time. The angular 2 app sat in production using the old CLI fork that had been abandoned and we weren't able to migrate because universal documentation for angular 4 was also relatively unclear.
Finally for future projects we opted to go with Vue, which has been relatively nicer. We used localization, token based authentication via cookies to enable authenticated universal rendering for logged in users, PWA features (offline). The project was analogous to Yelp, with product and service suggestions displayed to the user depending on their browsing habits and other aspects of their profile. Roughly 300k monthly uniques. Universal was essential for SEO and performance since 90% of traffic was search.
Other than struggles with universal and SEO, there were some issues with getting customizations to the webpack build process since the CLI fork we used didn't allow for many changes like adding minification, etc. We also wanted stuff at the express end of things like HTML minification, but there was no clear-cut way to do caching across things like authenticated vs unauthenticated. Maybe we couldn't think of the best way to go about this. Other frameworks seemed to have a painless way of doing this stuff, so we spent a lot of time wondering if we had made the right choice. Most of the plain client side stuff was very satisfying to use. RxJS is great as well, and it was nice to see it become popular across the JS ecosytem. I am not sure if I would go with angular for future work because the bundle size seems to be overwhelmingly high - perhaps due to a knowledge gap at our end. What would be fantastic some sort of sample kitchen sink demo application that employs all the best practices for everything - auth, localization, seo, universal, build process customization, etc.
At the time of our evaluation, we looked at a number of quickstart and bootstrap/boilerplate/starter kit projects, but each one seemed to lack one thing that we really wanted, with no clear path to integrating it in.