There was a bug in the load-datasets.py script which is running as a CronJob in the linkrecommendation service. Basically, while iterating over datasets to import for each wiki in a list of wikis (there are ten wikis), we are appending to a list of datasets to import instead of re-initializing it. The result is output like:
== Importing datasets (anchors, redirects, pageids, w2vfiltered, model) for cswiki == == Importing datasets (anchors, redirects, pageids, w2vfiltered, model, anchors, redirects, pageids, w2vfiltered, model) for simplewiki == == Importing datasets (anchors, redirects, pageids, w2vfiltered, model, anchors, redirects, pageids, w2vfiltered, model, anchors, redirects, pageids, w2vfiltered, model) for arwiki ==
And so on. There's no harm other than a waste of CPU resources and time (we'd like the datasets to finish updating sooner rather than later), so if linkrecommendation-production-load-datasets-1618311600-hn6k8 could be removed so that a new cron job container is created with the latest deployed code (fix was in https://gerrit.wikimedia.org/r/c/research/mwaddlink/+/678918), that'd be nice.