Ryan Michael

It may be possible to build tensorflow as a remote repository, but building it takes about 6h, which isn't really a workable solution for my team.

Sorry for the delay - that was using `a0812a27e2367dce65dc6f40ccaa12e76282d1e6`

My guess is that you're encrypting your backups. I had the same error, tried removing backup encryption, and now it works.

I'm happy to remove them, but there are lots of files committed under vendor in master, I assumed you guys were vendoring dependencies?

Our logs as they exist today aren't very human-friendly, but I mostly find myself using them to try to diagnose a failures, either by looking at the most recent logs...

The example of `since(hourly())` doesn't specify the window _width_ with respect to time - each window's width is different. This example specifies the window's leading edge with respect to time....

A quick and dirty way of accomplishing this would be to use the key's type (i.e. `string`) as the grouping ID.

I support this direction. Any downvotes?

The thing is that grouping doesn't prevent, or in my experience meaningfully reduce, the incidence of implicit joins, however it _does_ increase the complexity of the 'user model'. Adjacent query...

Is the idea that at compute time we would read N+1 files - the N data files and a single key-inverse file? That seems like it has some downsides: *...