Ideas

Current implementation of read-only replicas require them to be connected to the cluster. It means that every new replica adds some overhead to the operational cluster nodes, particularly its leader. It creates limitation of the feature as one mig...

Move custom and standard projections to Connectors.

Consider allowing read only replicas to connect to single node deployments. Currently, this is not possible because replication is disabled on single node databases.

ESDB currently supports only RSA keys in certificates when specifying --certificate-private-key-file. The advantage of using EC keys would be that the key is much smaller for the same security level (EC takes 256 bit vs RSA 2048 bit for same secur...

Seems like persistent subscriptions aim to solve two problems: Server-side checkpointing for use cases when events are being forwarded to a place that doesn't have the ability to store checkpoints (message brokers), or users just don't want to dea...

Crypto-assurance provides guarantees against mutations, truncation, and reordering. Although ESDB doesn't provide nativa mutation and reordering capability, data on disk can be tampered to change the content of events. ESDB supports truncation wit...

ESDB is primarily configured using the config file, which is located on each cluster node. Some features (projections, connectors) are configured in the database itself. It can be confusing to understand where things are configured and why some co...

The idea is to rewrite the chunk file after it stops accepting writes, so it can handle reads better. Re-sort events in closed chunks to group events of the same stream together for efficient IOPS use. Could also save having to repeat the stream n...

Introduction In today's data-driven world, the storage and management of events play a crucial role in various industries. However, storing all events indefinitely can be both space-intensive and costly. To address this challenge, we propose the i...

Allow clients to control write guarantees. Right now, ESDB confirms the write when the leader has flushed data to disk, and the majority has the data in memory. Possible additional options: All followers have data in memory so they can do consiste...