Download Latest Version
7.2.5 source code.zip (95.8 MB)
Get an email when there's a new version of RavenDB
| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| 7.2.5 source code.tar.gz | 2026-07-20 | 87.4 MB | |
| 7.2.5 source code.zip | 2026-07-20 | 95.8 MB | |
| README.md | 2026-07-20 | 9.4 kB | |
| Totals: 3 Items | 183.2 MB | 0 | |
Features
[CDC Sink]Added the CDC Sink task that syncs changes from relational databases into RavenDB documents Docs | Guide[Connection Strings]Added server-wide connection strings Docs | Guide[Debug]Added the Debug Package Analyzer Docs | Guide[Integrations]Added support for Power BI DirectQuery mode Docs[Queue Sink]Added Azure Service Bus support Docs | Guide[SSO]Added Single Sign-On (SSO) access to RavenDB Studio Docs | Guide
Server
[AI]Improved the error message when the configured AI provider does not support a given file format[AI Agents]Added AI agent client code generation[AI Agents]Improved AI agent conversation paging and exposed attachment names on the conversation level[AI Agents]Added cancellation support for AI agent conversation operations[AI Agents]Fixed a shutdown race in AI workers when the database record changed while new work was being queued[AI Agents]Fixed a BadRequest error after sending a trimmed AI agent conversation[Audit Log]Fixed audit logging for pull replication sink updates[Backup]Fixed a backup hang when uploading directly to Amazon S3 or Azure Blob Storage (upload I/O no longer runs on the backup thread's synchronization context)[Backup]Added an option to disable S3 checksum validation, allowed only for custom (S3-compatible) endpoints[Compaction]Fixed an error during compaction of the TimeSeries.Stats table[Corax]Fixed faceted queries returning empty values when the facet field used an alias (the term was resolved from the display name instead of the indexed field name)[Corax]Fixed search() with wildcards on dynamic fields created with CreateField()[Corax]Fixed UTF-8 bounds checking and same-language number comparison in alphanumerical sorting[Core]Fixed inconsistent equality, hashing and string materialization for values containing control characters; databases created with this version also reject control characters in document identifiers. This change does not affect existing databases[Databases]Persisted the database's supported features in the documents storage; creating a database over an existing data directory now restores them from disk instead of assuming the newest defaults[Debug]Fixed missing information from some shards in the debug package of a sharded database[Docker]Added a multiarch Docker image tag with the full server version[GenAI]Fixed GenAI tasks failing with an authentication error on secured servers by authenticating model queries with the server certificate[Indexing]Fixed index terms of non-ASCII values being truncated after a value longer than 2048 characters was indexed into the same field[Linux]Issued a sequential read-ahead hint for journal files on database startup recovery and released already-processed journal pages during recovery (configurable viaStorage.UseSequentialReadAheadHintForJournalRecovery, enabled by default)[Linux]Added a server startup alert when a block device'sread_ahead_kbexceeds the recommended threshold (configurable viaStorage.ReadAheadKbAlertThresholdInKb, default 128 KB)[Logging]Fixed an ObjectDisposedException when audit logging an offline database migration[Monitoring]Added support for customizing the OpenTelemetry service name and namespace (configurable viaMonitoring.OpenTelemetry.ServiceNameandMonitoring.OpenTelemetry.ServiceNamespace)[Monitoring]ETL, AI task and CDC Sink metrics are exposed as gauges[Monitoring]Excluded AI tasks from the ETL metrics group in server monitoring[Notifications]Included ETL, AI and CDC Sink task-error counts in database statistics change notifications so the Studio task errors badge updates live[Performance]Fixed high CPU usage when writing large HTTP responses[Querying]Fixed a NullReferenceException in streaming queries on encrypted databases with Lucene indexes when a concurrent read transaction was disposed mid-read[Queue Sink]Fixed the Queue Sink fallback retry back-off being stuck at 5 seconds instead of escalating[Replication]Fixed active/active replication divergence for counters and time series: conflict-resolved values could be skipped by the replication send gate, and concurrent same-timestamp time series values now converge regardless of replication direction[Replication]Fixed majority write assurance to compute the required sibling count from the database group topology, excluding promotable nodes, instead of the destinations list[Replication]Fixed bidirectional filtered pull replication inflating the hub's database change vector, which could cause internal replication between hub cluster nodes to skip documents[Replication]Fixed SinkToHub pull replication re-sending all documents from scratch after a hub node failover[Replication]Fixed outgoing replication heartbeat starvation on busy nodes and cleaned up half-open incoming replication connections[Replication]Fixed HubToSink pull replication connection handoff and cleanup, including reconnecting after configuration changes and guarding against stale connectors[Replication]Fixed external replication conflicting a document recreated after a cluster-wide delete, resolving to the tombstone instead of the new document[Replication]Fixed external replication of sharded revisions to pre-6.0 servers by stripping the revision change vector to its version part[Revisions]Changed revision storage keys to fixed-size hashes of the change vector, bounding key size regardless of cluster topology history. This release includes an internal storage schema update; after upgrading, downgrading to an earlier version is not possible[RQL]Replaced a NullReferenceException with a readable RQL syntax error when a method call has no name[Sharding]Fixed re-put attachments not surviving a bucket re-send after bucket migration source failover (an attachment put over a tombstone now supersedes the tombstone's change vector)[Sharding]Fixed bucket migration getting stuck when revision or attachment tombstones were present[Sharding]Fixed deleted time series being resurrected when a bucket was re-sent after the migration source node failed over[Smuggler]Fixed smuggler import disposing the caller's stream instead of honoring leaveOpen[Storage]Fixed database compaction corrupting the attachments index on databases migrated from 6.2 or 7.1 to 7.2 (the persisted table schema was stale). The schema is now repaired automatically at database load. If such a database was already compacted on an earlier 7.2 version and shows any signs of corruption, running the compaction again after upgrading might resolve the issue; otherwise restore from a backup[Vector Search]Optimized vector search indexing and query performance[Vector Search]Fixed a use-after-free in parallel HNSW node placement during vector indexing[Voron]Fixed tree rebalancing corrupting the tree when re-adding a large separator key split ancestor pages during a delete, which could unlink subtrees and make keys unreachable; added invariant checks that fail the transaction instead of committing corruption[Voron]Prevented journal recovery from overwriting flushed data when invalid journals are skipped due toIgnoreInvalidJournalErrors[.NET]Updated to .NET 10.0.9
Client API
[Session]FixedAdvanced.HasChangesreturning true after loading an entity with a ulong property holding a large value
Studio
[Certificates]Fixed database access checks to compare certificate database permissions case-insensitively[Core]Fixed download links in the import-from-RavenDB view and the new-version-available notification to point to the correct URL[Databases]Fixed footer statistics initialization to prevent null reference errors[Databases]Fixed the databases view to load database details from the first healthy node when the database is not hosted on the current node[Sample Data]Redesigned the Sample Queries view with a searchable scripts and methods reference
