I think that in the medium term we should get rid of flower and replicate the useful functionality through the scheduler interface.

Flower has a lot of issues with the way we work, as it stores its information in-memory instead of using a database. This breaks down badly when you run millions of jobs like we do...

I think that in the medium term we should get rid of flower and replicate the useful functionality through the scheduler interface.

Ditching flower is certainly an option. And what you mention about our own interface might be a good fit for an idea I briefly mentioned F2F a while ago: creating an "SWH admin tool" which we can use to (re)schedule jobs, monitor the current state, start/stop workers, scale up/down, etc. I had in mind more of a CLI tool, but other UI might be possible too.

That said, I don't think we fancy implementing ourselves generic monitor functionalities that, I suspect, should be the same for all celery, and even rabbitmq, users. So maybe there is room here for a more in-depth search of what other queue monitoring tools exist. We certainly aren't the only one with millions of jobs in our queues; hopefully someone else has released monitoring stuff that is more "professional" than celery.