app:recommendation:generate¶

Generate music recommendations with every strategy, as a recommendation job that runs in the command's own process. It is the shell counterpart of Generate on the admin Recommendations page, POST /api/admin/recommendations/generate, and leaves the same job record, which app:recommendation:job:list and the admin page list.

Quick start¶

Replace the recommendations of every song:

make exec cmd="php bin/console app:recommendation:generate"

Replace only those of songs updated in the last seven days:

make exec cmd="php bin/console app:recommendation:generate --mode incremental"

Print the result as JSON, for a script:

make exec cmd="php bin/console app:recommendation:generate --mode=incremental --json"

Options¶

Details¶

The command creates a recommendation job, then runs it in its own process: in the web server, jobs run on the CPU process pool, which a console process cannot reach. The job record records triggered_by: cli. The job moves to in_progress, names each strategy (collaborative, content, genre) as it starts, and ends completed. Follow it from another shell with app:recommendation:job:list or app:recommendation:job:show.

The run is also recorded in the job monitor as GenerateRecommendationsCommand, so app:monitor:jobs and the admin Job monitor page show it.

When the run finishes, the command prints the recommendation job's public ID, the job monitor ID, the mode, the final status and the number of recommendations each strategy saved.

Cancelling the job with app:recommendation:job:cancel or from the admin page stops the run before its next strategy starts. The strategies that already ran keep their recommendations, and the command exits with 1. A cancellation that arrives as the last strategy finishes is kept too: the job stays cancelled rather than completed. If the job completes first, the cancellation is refused. If generation throws, the job is marked failed with the error as its reason, unless it was cancelled meanwhile, and the command prints the error and exits with 1.

The command always generates, whatever the recommendations.auto_generate server setting says. That setting governs only the daily Generate recommendations scheduled job; see Scheduled recommendations.

A full run loads every song and every user's listening history, so on a large library it can take long and use a lot of memory.

With --json the command prints only the API's data object: job_id, public_id, mode, status, execution (always sync, because the job runs in this process) and counts, the recommendations each strategy saved. A job that was cancelled or stopped before it finished still prints the object and exits with 1. The job-monitor ID that the table shows is not part of the JSON.

Exit codes¶