Rich FitzJohn

Thanks Ross. Even if user error, a more informative error message is warranted.

Hi Ross. I appreciate the bug reports - but could you do us a favour and provide a minimally reproducible example for us? (i.e., 4-5 lines that we can run...

To build this, set up [toxiproxy](https://github.com/Shopify/toxiproxy) to simulate a slow Redis connection and test varying strategies. In particular, slow up/fast down would be a reasonable simulation of a home internet...

See [pipelining](http://redis.io/topics/pipelining) and [scripting](http://redis.io/commands/eval); there are a bunch of operations that could be done in a single round trip for each job that is submitted. That should see ~6x speed...

Please try this @dfalster, and if it works approve and merge the PR. Depending on how you are using the package, it will probably be quite simple to move over...

A nice feature of this is that the entire remake implementation could be to work out the topological sort, and then queue up a series of tasks and declare dependencies...

OK, the issue here is that the `key_complete` that is set is autogenerated and not that pretty (e.g., `tmpjobs:tasks:1:complete`) so it might make sense to add a new `group` to...

Let's have rrqlapply/enqueue_bulk generate user-friendly names by default, but have enqueue itself not add a name. The downside of this is that there are then two very similar keys: `key_complete`...

Most of the ideas live on in https://github.com/mrc-ide/rrq which works well for us and will likely get a bit of attention in Q4 this year

RedisAPI is deprecated with the functionality directly incorporated into the redux package. That should be used instead. rrq implements most features in rrqueue but despite writing both I could not...