Name | Modified | Size | Downloads / Week |
---|---|---|---|
Parent folder | |||
README.md | 2025-07-08 | 3.9 kB | |
v3.4.0 source code.tar.gz | 2025-07-08 | 3.0 MB | |
v3.4.0 source code.zip | 2025-07-08 | 4.8 MB | |
Totals: 3 Items | 7.8 MB | 2 |
Memgraph v3.4.0 - July 10th, 2025
⚠️ Breaking changes
- Introduced a small breaking change under
SHOW INDEX INFO
output. Previously, all vector indexes returned the hardcoded string"vector"
as their type. Now, the output reflects the actual indexed entity more precisely: for node indexes the type islabel+property_vector
, for edge indexes:edge-type+property_vector
. #3059
✨ New features
- Quantization options have been added to the vector index. Users can now reduce memory consumption by enabling quantization in the config map, using the
scalar_kind
field to define the scalar type for vectors, at the cost of some precision. #3037 - Added function
getHopsCounter()
to return the current number of hops traversed in a given query. #3048 - Added monitoring of replica recovery process. Users will now be able to monitor how often an instance gets into recovery replication state and how often instance gets successfully recovered once in that state. #3002
- Added support for vector indexing on edges (edge_type & property combination). Users can now create and query vector indexes on edge embeddings. The syntax for creating an edge-based vector index is:
CREATE VECTOR EDGE INDEX vector_index_name ON :EDGE_TYPE(embedding) WITH CONFIG config_map;
. #3059
🛠️ Improvements
- Added support for
@neo4j/graphql
middleware v7.2.0. Users can now leverage better Neo4j / GraphQL compatibility and use Memgraph with the latest GraphQL ecosystem. #3027 #3083 - Added non-blocking index creation. Until this point, index creation was a "stop the world" operation on a given database. With this change, at the time of
CREATE INDEX
queries, memgraph will exhaust all in-flight write operations and pause/queue all new write operations. When all in-flight writes are done, Memgraph will make the index usable and allow writing. Read queries are possible at any time. From the user perspective,CREATE INDEX
operations should feel completely non-blocking. Only in extreme cases under heavy-write load might there be some pausing of write queries, but none of the writes should be lost. #2923 #3064 #3073 #3091
🐞 Bug fixes
- Queries using regex clauses on label+property indices no longer return an empty results set. Users can now use label+property indices to improve performance of regex queries (although note that, unlike exact text match queries, regex queries require an additional filtering step). #3025
- Queries now support directly accessing properties of maps nested inside lists, such as
RETURN [{name: 'Alice'}, {name: 'Bob'}][1].name;
. This allows users to more easily work with Memgraph if their schema contains nested data. #3058 - Fixed HA upgrade. Upgrade of coordinators from 3.2.1 to 3.3 was failing due to a bug in the serialisation code. With this bug fix, users can expect smooth upgrade of coordinators. #3078
- Executing
RECOVER SNAPSHOT
onMAIN
will propagate the snapshot to replicas as well. This simplifies the recovery of the whole cluster. UseRECOVER SNAPSHOT
to easily overwrite the data stored in a database. 3031