Mirabelle is heavily inspired by Riemann, but still have major differences.
In Riemann, streams are function. In Mirabelle, they compile as EDN.
It means streams configurations can be easily parsed by other systems, passed between systems, verified…
In Mirabelle, all actions use the events time as wall clock. It means streams times advance depending of the event they receives.
Because of that, it’s easy to reason about (and test) streams: the same inputs always produce the same outputs. It’s also possible to have multiple streams with different times running in parallel.
You can in Mirabelle create streams which will not receive the events passed to the Mirabelle TCP server by default.
It means you can have some streams for “real time” computation, and some streams for some use cases. Each stream will run independently, will have its own time…
When you send an event to Mirabelle, you can specify the stream attribute in order to send the event to a specific stream.
Riemann mostly works on the :host and :service fields. it’s common to encode “dimensions” in :service.
For example, you could have in Riemann a service named http_requests_duration_prod_p99. In Mirabelle, you would model this event like this: {:service "http_requests_duration" :environment "prod" :quantile "0.99"}
A lot of Riemann streams/actions (index, coalesce…) were rewritten. You can now provide for each action on which keys it should work instead to have to manipulate complex :service names.
New actions which I think could be interesting were added to Mirabelle. Also, not all Riemann actions were backported, only the main ones (I could probably backport the other ones later if needed).
Mirabelle provides an HTTP API which allow you to manage streams dynamically (create, list, get, remove), and to gather information about your system:
Mirabelle does not use the Riemann query language to query the index. Instead, it uses its own language, which is used everywhere (index queries, where action…).
Streams and I/O are configured separately, and streams references I/O. These components do not have the same lifecycle, and handling I/O is less error prone in Mirabelle, especially during reloads.
In Riemann, all streams states (time windows for example) all lost during reloads. it’s not the case in Mirabelle. Only streams which were modified in the configuration are recreated.
In Mirabelle, each stream has its own dedicated index.
A lot of small new features were also added (like a new pubsub mechanism, the on-disk queue…).