On a clustered setup of EventStore used for pre-production testing, we have observed something concerning (we think) and need some help as to how we should proceed.
We have created a FromStreams projection with a number of different streams. While “bootstrapping” the system with a large amount of data, the code using this projection ended up failing since information from a specific event was missing.
We have inspected the stream where this event belongs, and the event is present, it just never appeared in the projection.
We have pulled the data from that eventstore back into a “lab-machine” where we can better study the data. This is a single node setup of EventStore. If we re-create the projection (copy to a new name) all the events appear on that projection, including the previosly missing event.
So something like: StreamX, StreamY combined into FromStreams Projection Foo
StreamX generates events: 1,2,3
StreamY generates events: 4,5,6
Foo receives events: 1,3,4,5,6
We are aware there are no guarantees of ordering, but missing events completely seems to be something very different from that.
We have only recently started experimenting with the clustered setup, and have never observed anything like this in any of our single node tests setups even when stressing the system.
So, a couple of questions
Are we missing some important information regarding clustered setups of EventStore and projections?
Is this maybe a bug?
what can we do to investigate this further?
We will happily provide any details we can.