Laksetap

Laksetap is a R Golem Shiny App for reporting salmon and rainbow trout losses in Norway.

It is published at apps.vetinst.no/laksetap.

The app runs with dummy data locally without the need to be connected to VI’s server. The dummy data is distributed as part of the package in inst/extdata. For anyone interested in implementing the app with their own data, running the app from dev/run_dev.R will load the app with the dummy data. The dev folder also contains definitions of the data sets and scripts to test the visualization functions with the dummy data.

Since the app is built for reporting fish mortality in Norway, it assumes that the species displayed are salmon and rainbow trout. However, the available data for these species may differ in terms of aggregation levels. Additionally, Norway’s geographical areas for displaying this data are also assumed: production area as “Produksjonsområde”, county as “Fylke”, and country.

Because of this, the app has choices (adjustable in the config) for species and geo_group in mod_top_bar.R and we have implemented a logic about adjusting the UI when switching between them. This could be improved, or generalized so any species can be displayed in the app, but for the time being this remains like so.

The code that produces the dummy data also follows this assumption and does not generate data for other geographical levels than country for rainbow trout.

All of the other inputs are read from the data, such as number of year, areas for which data are available, etc.

The app can be configured and translated to different languages. Files managing this are passed as golem options in run_app.R.

The inst/app/www/config.json contains paths to images, favicons, social media previews, and similar. Additional configuration that lives within the code is the tracking script in tag_manager in fct_meta.R. This needs to be removed, or adjusted if you are using the app with your data.

All labels used in the app are in inst/app/www/labels_nb_no.json. An English translation of the .json is also provided. The long form text (i.e. more than a few words) is in separate .md files inst/app/www/.

The code in fct_data.R also contains a bit of locale specific code, like for example labeling of data for "dead", "discarded", "escaped", "other" to "Døde", "Utkast", "Rømt", "Annet". This could easily be overlooked, but the idea here is to be able to use the data functions both in the app, and in an interactive R session. One should be aware of this when working with the data and app.

We provide a mortality calculator in R and Excel as part of the repository, and downloadable from the app. The calculator itself is not part of the app UI because we don’t want to handle company or farm level data in the app and potentially run into legal or security issues. The calculator uses the same approach to calculate fish mortality that is used in Norwegian Veterinary Institute’s Fish Health Report.

Running the app with docker should work:

docker build -f Dockerfile_base --progress=plain -t laksetap_base .
docker build -f Dockerfile --progress=plain -t laksetap:latest .
docker run -p 80:80 laksetap:latest
# then go to 127.0.0.1:80

The deploy.yml (prod, triggered on release) and deploy-dev.yml (dev, triggered on push to dev) workflows call rsconnect::writeManifest() to build a Posit Connect manifest from renv.lock.

rsconnect must be present in renv.lock itself, not installed ad-hoc in the workflow (e.g. install.packages("rsconnect")). Since rsconnect 1.0.2, writeManifest()/deployApp() error out if the installed library and renv.lock are out of sync, and an unpinned install.packages() call in CI introduces exactly that drift (a package present in the library that the lockfile knows nothing about).

rsconnect is listed in DESCRIPTION under Suggests (it’s a deploy-only tool, not used at runtime), but renv’s package.dependency.fields setting (renv/settings.json) only scans Imports/Depends/LinkingTo when doing a regular renv::snapshot(), so a plain snapshot never picks it up. To add or update a package that isn’t reachable from those fields:

• Use renv::record("package_name") — it updates just that package’s lockfile entry without touching (or dropping) anything else.
• Do not use renv::snapshot(packages = "package_name") for this. Despite appearances, it recomputes the whole lockfile scoped to that package’s own dependency tree and silently drops every unrelated package.
• renv::record() only updates the named package, not its transitive dependencies, and it records the latest version available in the repository (not necessarily what’s installed locally). If that pulls in a new dependency (e.g. rsconnect needing httr2) or raises another package’s minimum required version (e.g. httr2 needing a newer rlang), record those packages too, or the CI install will fail with a namespace version-mismatch error.

Both deploy workflows read the top-level renv.lock (not renv.lock.prod, which is only used by the Docker builds).