Event: any class implementing the com.google.gerrit.server.events.Event abstract class
Instance: a running instance of Gerrit (a single JVM process)
Master: Gerrit master
Replica: Gerrit replica (or slave)
Topic: a broker topic, i.e.: a stream of Events of the same category (
Problem: Where is an Event coming from?
In a multi-master setup, high-availability or multi-site, events are broadcasted among all the master instances belonging to the same cluster and share the same repositories.
Currently, as a plugin developer, I cannot determine from which instances events are coming from.
Each plugin implements its own strategy to understand the source of an event and therefore we end up duplicating development effort.
Gerrit should provide a way of adding information to events that can help consumers identify the initial source of the event.
Given a multi-master, single or multi-site, Gerrit setup, it is not possible to determine, from which instance in the cluster an event has been produced.
Events don’t keep track of the instance they were generated from, they only carry information about their type and creation time.
In case of multi-master setup the events need to be broadcasted to all the instances of the cluster.
This is fundamental to notify other instances on the activities happening on the master and allow them to align.
Currently any plugin uses its own way of deducing the origin of an event.
Examples can be found in the
high-availability and the
The former, for instance, flags the in flight Events as “forwarded” in the thread local storage.
The latter, adds the origin in an “envelope” wrapping the original Event.
Considering that multi-master setups are a common and fundamental Gerrit usage, inferring of the event origin shouldn’t be delegated to external plugins.
AS A Gerrit plugin developer
GIVEN A Gerrit cluster with instances
G1 produces an Event
THEN I want to know
E1 was produced by
Being able to determine the source of an event will simplify development of plugins dealing with events. Here some examples of cases this additional information can be used for:
- Avoid loops: in a multi-site setup events are broadcasted to topics. Each instance produces and consumes from the same topic. Events end up being consumed and processed multiple times.
- Priority routing: prioritise consumption from a particular instance when the replication queues are loaded
- Troubleshooting: exposing which instance is serving a request
- Preventing events duplication: when listening to events through plugins (e.g. Slack integration) a single event would be consumed by all the instances, causing duplication of notifications (e.g. Slack would be notified by each instance with the same message).