Revisiting the Unified API Model
Some thoughts were discussed around the two major groups of API symantics, their use cases, and whether we can support all the current semantics for high-throughput use cases.
Thinking about batching in terms of the transport. Supporting a more stream-oriented protocol, such as WebSocket, would allow for pipelining requests and avoid connection setup/teardown cost for high-throughput use cases. Perhaps only implement for message consumption?
Proposals around how to better support messages that aren't easily modeled with JSON.
How do we define a "flavor"? What sorts of attributes would an operator want to set for a given flavor?
RPC Event Aggregator
We talked about how Marconi could be used to take in OpenStack's RPC bus firehose, acting as an aggregator, normalizer, and filter for projects that simply want to subscribe to specific types of events, with stable and consistent message schemas.
What does the notifications service need in addition to what the queues service already provides?
How do we migrate a queue to another pool with zero downtime?
Proposed implementation for sharding messages in a single queue across multiple nodes in a given pool.
Planning a benchmarking tool that can test end-to-end scenarios. Such a tool could also provide a "second opinion" to validate Tsung benchmarks.
Discussing items that came up in our graduation review, and our plan to work on them during Juno.
Rough plan for Juno.