It turns out TRACE is a higher logging level than DEBUG which doesn't
make much sense and is confusing. We can use FINE to achieve the same
thing and it's descriptive and less confusing
The logging configuration now accepts strings "local0" through "local7"
as facility selectors.
The input is lowercased such that e.g. "LOCAL0" is valid, too.
Fixes#571. Removed the old long term and short term shards. Now if the
user doesn't manually create a shard space, a catchall called "default"
will be created the first time they write data in.
Added API endpoints to list, create, and delete shard spaces. Made a
breaking change to the old get shards API by changing the format since
shortTerm and longTerm shards aren't guaranteed to be present.
Shard spaces can be mapped to a single database. Retention policies are
enforced per server by a background routine that periodically checks
shards in spaces that have retention policies to see if they have passed
their expiration.
It turns out that smaller batch sizes are causing too much
overhead. It's not clear why, but empirical testing shows a 10x increase
in write speed when the batch size increases from the previous hard
coded 64K to 10M.
Todd Persen
John Shahid
Philip O'Toole
Kimo Rosenbaum
Malthe Borch
Paul Dix
Tony Pitale
Julien Ammous
Mitchell Bradd