Pubsub + Kafka

[matt2e]

https://hackmd.io/03-rIhlZTmycnNzida4qRw?view

• Add back ability to Reset Subscription #3718
• ftltest to use partition key #3576
• Add single partition mapper and check that it only is allowed for single partitions #3558
• Improve error message if //ftl:subscribe directive just has a casing issue #3525
• Update docs #3496
• Commit offsets every few seconds within a batch #3493
• Provisioner should error/warn(?) if partition count does not match #3492
• Dead letter topics #3490
• Integration tests for losing runner (signal, panic, new deployment) #3473
• Integration tests for rebalancing consumer groups #3472
• Each subscriber verb can only subscribe to one topic #3467
• Telemetry #3461
• Test publishing throughput #3459
• use headers to store original request key (and what else did we used to store in async calls?)
• Provisioning topics in dev is slow #3901
• Request keys
• Conform to could events? https://cloudevents.io/ (@jvmakine: That is pretty much the standard for kafka messages at block AFAIK)
• Changing partitions is a massive pain... even in dev...
• Go external modules do not generate ftl:topic export partitions directives
• Reset subscription does not contact all runners for a deployment, only one of them.

Out of scope / Known issues:

• Provision pubsub in AWS with IAM #3466
• Can not reset subscription if there are multiple runners for a deployment #3986
• Figure out when to reset pubsub #3723

[matt2e]

Current state of epic:

• Provisioner sets up RedPanda
• Go and JVM publish to RedPanda
• Consumption does not read from RedPanda
• Go publishes and consumes from old pubsub still, just to keep things working in maintenance mode.

Full round trip for go-runtime (publish & consumption) is coded up in a spike here: #3603