#Rstats really is a great community - I'm surprised by how many ppl I know through past events. It's been so great catching up with you all - everyone I've spoken to thus far. Also delighted to be meeting some people for the first time! #useR2026 opening reception a success, onto the first conf day!
Remote
shikokuchuo
@shikokuchuo@fosstodon.org
mastodon 4.7.3Software Engineer, Open Source @ Posit, PBC
176 Followers
0 Following
18 Posts
Joined December 26, 2022
Github:
Website:
I've distilled my experience of useR! 2026 Warsaw into a blog post.
Hoping that people who couldn't attend get a taste of it, and are motivated to participate in the next one - or join us for posit::conf in Houston! For those who wanted to find me, but didn't get the chance - the userR! slack is still open, I'm also on the rOpenSci one, feel free to ping me on either.
https://opensource.posit.co/blog/2026-07-14_user-2026-warsaw/
We're continuing our work on {mori} the package for shared memory in #Rstats.
Debuted in April, we just released 0.2.2. A patch release for technical correctness, it's a prerequisite for future work we have planned.
{mori} provides a key strategy for minimising memory usage for local async and parallel operations.
We hope you'll try it out as we build out more features!
Just released mori 0.1.0 on CRAN — a new R package for sharing R objects across processes via OS-level shared memory.
Parallel R today duplicates your data in every worker's RAM. 8 workers × 1 GB = 8 GB. mori puts it in shared memory once; every worker maps the same physical pages via R's ALTREP framework.
Works with any parallel backend that uses R's standard serialization, including mirai, parallel and callr.
Today I shipped a patch version of mirai with as many changes as a full version!
- Lower dispatch overhead for `mirai_map()` and `everywhere()`
- Clearer errors from timeouts and cancellations in `mirai_map()`
- Many edge case fixes for `race_mirai()`, `daemon()` and the dispatcher
These automatically translate into wins if you use parallel purrr!
p.s. yes, we also have a shiny new hex logo!! Stickers ready for posit::conf(2026) Houston 🌎️
OpenTelemetry has come to R 🎉
We've instrumented Shiny, plumber2, mirai, httr2, ellmer, knitr, testthat, and DBI — so you can get production-grade traces from your R apps with no code changes. Just set a few env vars and telemetry flows to the backend of your choice.
Full write-up: https://opensource.posit.co/blog/2026-05-07_opentelemetry/
Async R is getting a proper memory model. mirai 2.7.0 and mori 0.2.0 are on CRAN, and between them they close the two memory gaps in production async work:
- `daemons(memory = ...)` caps the dispatcher queue in bytes, so a runaway producer can't OOM your session
- `try_mirai()` returns NULL on a full queue instead of blocking – drop, retry, or propagate backpressure
- mori out of experimental with stable wire format and path-form names
https://opensource.posit.co/blog/2026-05-12_production-async-r/
Watcher - meet the R community!
```r
install.packages("watcher")
```
Tell us how you end up using it - open a discussion show-and-tell at r-lib/watcher!
Catch you all at useR! 2026 in Warsaw next week!!!
Replying to
@Lluis_Revilla @Alejandro_P
Thanks @Lluis_Revilla you're exactly right. I'd say that simply having the capability is what's important, rather than if it lives in base or a package.
Replying to
@noamross@ecoevo.social See https://mirai.r-lib.org/articles/v07-questions.html#launching-daemons---vanilla you can use remote_config() to specify a replacement for Rscript where you can specify these
Installing mirai from source just got simpler.
nanonext 1.10.0 (mirai's communications layer) was released to CRAN this morning. We ripped out the CMake build-time dependency, with the bundled NNG + Mbed TLS sources now only needing a C compiler.
This means you can now install mirai on really ancient Linux releases! It'll install anywhere R does.
Come code and hang out with us in Houston!
Replying to
@noamross@ecoevo.social R itself has options - see ?mem.maxVSize - and I think there are env vars you can pass to mirai
Sincere thanks to all of you who attended our talk at useR!2024, where we shared the design choices behind {mirai} and {crew}.
@landau@fosstodon.org and I are speaking again at Posit::conf where we will unveil the next generation of async we designed together with Joe Cheng for Shiny ExtendedTask, Plumber and targets. A little taster currently in the rweekly highlights: https://shikokuchuo.net/posts/24-conditional-linkingto/