Release notes

This page contains the release notes for EventStoreDB 24.10

24.10.5

19 May 2025

#4893 and #4990)

Fixed an issue with filtered subscriptions where it was possible to incorrectly send a checkpoint of (ulong.max, ulong.max) to a filtered $all subscription if:

• the filtered subscription started from the beginning of the database and
• none of the events in the database matched the filter and
• the checkpoint interval was greater than the number of events in the database.

The checkpoint will now be updated to a valid value before sending it when transitioning to live.

#5022)

Fixed the projection metrics not reporting the projection status when it had a compound state (e.g., faulted (enabled)).

#5062)

Added two persistent subscription metrics to count the number of parked message requests and replays.

eventstore_persistent_sub_park_message_requests
eventstore_persistent_sub_parked_message_replays

The park message requests are subdivided into two reason categories: client-nak and max-retries.

#5007)

Added Kurrent, Kurrent.Connectors and Kurrent.Connectors.Sinks meters. The following metrics are now available for connectors:

kurrent_connector_active_total
kurrent_sink_written_total_records
kurrent_sink_errors_total
kurrent_sink_transform_duration_milliseconds
messaging_kurrent_consumer_message_count_total
messaging_kurrent_consumer_commit_latency
messaging_kurrent_consumer_lag
messaging_kurrent_producer_queue_length
messaging_kurrent_producer_message_count_total
messaging_kurrent_producer_produce_duration_milliseconds
messaging_kurrent_processor_error_count_total

These metrics are described in the connectors documentation.

Connectors: Add data protection

We've implemented a data protection system to secure sensitive connector configurations. All connectors now use envelope encryption to protect sensitive data like passwords and tokens, ensuring secure transmission. Setup is simple, with token-based protection that can be configured directly or through separate files for added security in production.

Connectors: Include deleted connectors in results

Previously, deleted connectors were unintentionally omitted from the results, leading to incomplete data visibility. This fix ensures that both deleted and non-deleted connectors are properly included where applicable.

Connectors: Added validation for Serilog connector configuration

Added validation for Serilog configuration during setup or changes. This helps catch errors early and prevents issues with logging functionality.

Connectors: Ensure reset connector command starts at offset 0

Previously, executing the reset connector command did not correctly reset the offset to zero, leading to inconsistent data replay. This fix ensures the connector starts from the beginning as expected when reset.

Connectors: Rename connectors registry snapshot stream

A system stream for connectors was incorrectly named $$connectors-ctrl/registry-snapshots instead of $connectors-ctrl/registry-snapshots and therefore treated as a metadata stream.

This has been renamed, and the empty connectors-ctrl/registry-snapshots stream will be deleted when upgrading from older versions of 24.10 to remove the incorrect metadata stream.

24.10.4

6 March 2025

#4881)

This reduces memory pressure when running projections, especially on databases with larger events.

There is a change in behaviour between 24.10.4 and previous versions.

Previously unquoted keys, e.g., { foo : "bar" }, were erroneously considered valid JSON and would be processed by the projections. This is not the case starting from 24.10.4.

Projections won't fault after an upgrade if they previously emitted an event based on a now-invalid event.

#4882)

The MaxAppendEventSize option has been added to limit the size of individual events in an append request over gRPC and HTTP. The default for this new option is int.MaxValue (no limit). Events appended using the AtomPub HTTP API still may not exceed 4 MB. Events written by internal systems or services, such as projections, are unaffected.

This option differs from MaxAppendSize, which limits the size of entire append requests over gRPC.

A new RPC exception maximum-append-event-size-exceeded was added as part of this, and is returned when any event in an append batch exceeds the configured MaxAppendEventSize.

Note: The calculation of event size for events appended over gRPC was fixed as part of this, which could result in a change in behaviour for appending batches of events over gRPC. Previously, the event size was calculated off of the event's data, excluding the size of the event's metadata or event type. This means that append requests which previously did not trip the MaxAppendSize check could fail on 24.10.4.

#4884)

The MaxProjectionStateSize option has been added to configure the maximum size of a projection's state before the projection should fault. This defaults to int.MaxValue (no limit).

A projection will now fault if the total size of the projection's state and result exceeds the MaxProjectionStateSize. This replaces the warning log for state size in CoreProjectionCheckpointManager.

Note: Sometimes, a projection's state and result will be set to the same value and written to the projection state event twice. This can result in the MaxProjectionStateSize being exceeded even when the projection state is half the configured maximum size.

24.10.3

27 February 2025

Fix getting projection statistics over gRPC for faulted projections (PR #4863)

Before this fix, projection statistics could not be retrieved via gRPC if one of the projections immediately transitions into a faulted state.

Fix possibility of a gRPC write hanging clientside if a leader election occurs (PR #4858)

Prior to this fix, if a gRPC write was pending at the time of a leader election, the gRPC call could be terminated successfully instead of in a failed state. On the dotnet client, this results in the Append Task call hanging indefinitely; other clients may also be affected.

24.10.2

13 February 2025

Handle replayed messages when retrying events in a persistent subscription (PR #4777)

This fixes an issue with persistent subscriptions where retried messages may be missed if they are retried after a parked message is already in the buffer. This can happen if a user triggers a replay of parked messages while there are non-parked messages timing out and being retried.

When this occurs, an error is logged for each message that is missed:

Error while processing message EventStore.Core.Messages.SubscriptionMessage+PersistentSubscriptionTimerTick in queued handler 'PersistentSubscriptions'.
System.InvalidOperationException: Operation is not valid due to the object's current state.

If a retried message is missed in this way, the consumer will never receive it. The persistent subscription will need to be reset in order to recover and receive these messages again.

Validate against attempts to set metadata for the "" stream (PR #4799)

An empty string (“”) has never been a valid stream name. Attempting to set the metadata for it results in an attempt to write to the stream “$$”, which, until now, has been a valid stream name.

However, writing to “$$” involves checking on the “” stream, to see if it is soft deleted. This results in the storage writer exiting, which shuts down the server to avoid a 'sick but not dead' scenario

“$$” is now an invalid stream name, and so any attempt to write to it is rejected at an early stage.

Fix EventStoreDB being unable to start on Windows 2016 (PR #4765)

EventStoreDB would crash when running on Windows 2016 with the following error:

[ 1488, 1,16:45:11.368,FTL] EventStore Host terminated unexpectedly.
System.TypeInitializationException: The type initializer for 'EventStore.Core.Time.Instant' threw an exception.
---> System.Exception: Expected TicksPerSecond (1853322) to be an exact multiple of TimeSpan.TicksPerSecond (10000000)
at EventStore.Core.Time.Instant..cctor() in D:\a\TrainStation\TrainStation\EventStore\src\EventStore.Core\Time\Instant.cs:line 21
--- End of inner exception stack trace ---

This was caused by a bug when converting ticks between Stopwatch.Frequency and TimeSpan.TicksPerSecond.

Fixes to stats and metrics

Fixed the following issues:

• On Linux, the disk usage/capacity were showing the values for the disk mounted at / even if the database was on a different disk (PR #4759)
• Align histogram buckets with the dashboard (PR #4811)
• Fix some disk IO stats being 0 on Linux due to the assumption that the /proc/<pid>/io file was in a particular order and returned early without reading all the values (PR #4818)

Support paging in persistent subscriptions (PR #4785)

The persistent subscription UI now pages when listing all persistent subscriptions rather than loading all of them at once. By default, the UI shows the persistent subscription groups for the first 100 streams and refreshes every second. These options can be changed in the UI.

A count and offset can now be specified when getting all persistent subscription stats through the HTTP API: /subscriptions?count={count}&offset={offset}.

The response of /subscriptions (without the query string) is unchanged.

24.10.1

18 December 2024

Up to 100 license entitlements are fetched (PR #4670)

Previously, only ten license entitlements were returned, meaning that entitlements were missing on the commercial license policies with more than ten entitlements.

If you are using a commercial trial of 24.10, you must upgrade to 24.10.1 to transition onto the full license.

SLOW QUEUE messages are logged by default again (PR #4660)

Slow bus messages were still logged, but the logging of slow queue messages had been accidentally disabled by default.

Update dependency patch version due to CVEs (PR #4674)

System.Private.Uri has been upgraded from 4.3.0 to 4.3.2 (patch upgrade) due to GHSA-5f2m-466j-3848 and GHSA-xhfc-gr8f-ffwc

24.10.0

20 November 2024

Find out What's new in this release.

24.6.0

12 July 2024

Critical fix - Events in explicit transactions can be missing from $all

Events written in explicit transactions via a TCP client can be missing from $all reads and subscriptions.

These events are present when reading from a specific stream but may be absent when reading from $all. This happens only with events which were written using transactions on the TCP client.

• This bug only affects events written in explicit transactions via the deprecated TCP API
• The bug appears to affect Filtered all reads when the window is bigger than the maxcount
• The bug can affect regular all reads/subscriptions

After the fix, the events are no longer missing from $all reads and subscriptions.

This is also in v23.10.2 and v22.10.6

Breaking changes

The database now restarts after a successful truncation operation

It is expected that EventStoreDB is run with a process manager like Systemd to automatically restart the process if it shuts down. This is only a breaking change if your process manager is configured to not restart the process if it shuts down twice within a short period of time.

A node will now restart after a truncation operation. This will affect the behaviour of the node in the following cases:

Deposed leaders may shut down twice before they rejoin the cluster

When a leader loses connection to the rest of the cluster, the remaining nodes may elect a new leader among themselves. When the former leader rejoins the cluster, it may go offline to truncate uncommitted records from its log so that it can subscribe to the new leader.

The node will now restart a second time after truncation has completed successfully.

Nodes will restart after performing a file copy restore.

Part of the file copy backup and restore procedure requires copying the chaser checkpoint file over the truncate checkpoint file. This triggers the database to truncate the log on the next startup.

The node will restart after the truncation has completed.

Unbuffered config setting now has no effect

The Unbuffered config setting is deprecated and now has no effect.

Contact us if you're using this feature.

Persistent subscription checkpoint event type name change

The event type for persistent subscription checkpoints has been updated from SubscriptionCheckpoint to $SubscriptionCheckpoint. Existing checkpoints written before upgrading will still retain the old event type and remain fully functional.

This will only affect users who read from or subscribe to the persistent subscription checkpoint streams or $all, and consume those events directly. The persistent subscriptions themselves are unaffected and will continue to function as usual.

Performance improvements

Scavenged databases now open much faster on startup

Midpoints for scavenged chunks are now built later on demand. This significantly improves the performance of opening a large scavenged database.

Reduced FileHandle usage by 80%

This makes it less likely for the database to hit the file limit, and lowers the memory footprint of EventStoreDB.

In this release

Metrics

• Improved system metrics so that the following are available:
  • CPU Usage on Linux, FreeBSD, OSX, Windows
  • CPU Load Averages on Linux, FreeBSD, OSX
  • Disk IO Stats on Linux, Windows and OSX (read bytes and written bytes only)
• Added a metric to track the number of elections, allowing users to configure alerts if the count exceeds a defined threshold within a given time period.
• Added metrics for projection subsystem:
  • Projection status
  • Percent progress
  • Events processed since restart
• Added metrics for Persistent Subscriptions:
  • Total items processed
  • Connection count
  • Total in-flight messages
  • Total number of parked messages
  • Last seen and/or checkpointed event

Quality of life improvements

• EventStoreDB can now build and run on OSX for development purposes.
• Index merge will now continue without bloom filters if there is not enough memory to store them.
• Use the advertised addresses for identifying nodes in the scavenge log.
• Change status of incomplete scavenges from "Failed" to "Interrupted".
• Log an error when the node certificate does not have the necessary usages.
• Add more information to the current database options API response.
• Allow statusCode for health requests to be provided in the query string. e.g. /health/live?liveCode=200
• Projections will now log a warning when the projection state size becomes greater than 8 MB.

24.2.0

29 February 2024

New plugins

Connectors plugin preview

EventStoreDB 24.2.0 introduces a preview of the Connectors plugin, designed to streamline the integration of EventStoreDB with external services. Connectors allow for the filtering and forwarding of events directly to downstream services and remove the need for manual subscription or checkpoint management.

Each connector runs server-side, maintains its own checkpoints, and can be configured to run on nodes with a specific role (i.e., Leader, Follower, or ReadOnlyReplica).

The preview includes two sinks:

• Console Sink: For testing and development purposes.
• HTTP Sink: For POSTing events to an HTTP endpoint of an external system.

Refer to the documentation for instructions on setting up and configuring connectors and sinks.

x.509 user certificates commercial plugin

The new User Certificates plugin enables the use of x.509 user certificates to authenticate users on a cluster.

In this initial version of the plugin, x.509 certificate authentication works in addition to the basic authentication method. This allows you to continue using basic authentication while gradually transitioning to x.509 certificate authentication if desired.

This plugin is available in the commercial version of 24.2.0 and is disabled by default. Refer to the documentation for more information on how to use this feature.

Logs download commercial plugin

The Logs Endpoint plugin provides you with the ability to list and download log files for an EventStoreDB instance over HTTP.

This allows developers and users without direct access to the machines where the log files are stored to easily list and download log files for diagnostic purposes.

This plugin is part of the commercial version in 24.2.0 and is enabled by default. Refer to the documentation for more information on how to use this feature.

OpenTelemetry exporter commercial plugin

The Open Telemetry Exporter plugin allows you to export EventStoreDB metrics via the Open Telemetry Protocol to a specified endpoint. This direct export capability simplifies integration with Application Performance Monitoring (APM) providers, enhancing visibility into EventStoreDB operations without additional configuration overhead.

Refer to the documentation for more information about using and configuring this plugin.

Feature Enhancements

Catch-up Subscription Improvements

Version 24.2.0 introduces significant improvements to catch-up subscriptions over gRPC, including a seamless transition mechanism between live and catch-up modes.

This update addresses the previous challenge, where catch-up subscriptions that fell behind would be dropped by the server with the message "Consumer too slow". The client would then need to resubscribe from the last checkpoint to continue receiving events.

As of 24.2.0, catch-up subscriptions automatically revert to catch-up mode on the server side without user intervention, improving reliability and consistency.

Additionally, subscriptions can now reauthorize the user running the subscription in response to user access changes. This means if you remove a user's access to a stream (for example, through ACLs on the stream), any subscriptions that the user has to that stream will be dropped.

Better support for containerized environments

EventStoreDB 24.2.0 is now able to detect when it's running in a containerized environment and will disable certain auto-configuration options. This helps prevent the node from running out of resources and allows for more fine-tuned control of the EventStoreDB instances.

Breaking changes

External TCP API Removed

The deprecated external TCP API has been removed in 24.2.0. This means that any external clients using the TCP API will no longer work with EventStoreDB versions 24.2.0 and onwards.

A number of configuration options have been removed as part of this. EventStoreDB will not start by default if any of the following options are present in the database configuration:

• AdvertiseTcpPortToClientAs
• DisableExternalTcpTls
• EnableExternalTcp
• ExtHostAdvertiseAs
• ExtTcpHeartbeatInterval
• ExtTcpHeartbeatTimeout
• ExtTcpPort
• ExtTcpPortAdvertiseAs
• NodeHeartbeatInterval
• NodeHeartbeatTimeout
• NodeTcpPort
• NodeTcpPortAdvertiseAs

The options for the internal TCP API (ReplicationTcp*/IntTcp*) are unchanged from version 23.10.0.