Memgraph Files

Memgraph v3.1.1 - Mar 28th, 2025

✨ New features

• Added list comprehension Cypher feature. It's now possible to execute queries like MATCH (x) RETURN [(x)-->(y) | y.prop];. #2656
• Added --storage-access-timeout-sec flag. Previously the access timeout was hardcoded to 1sec. Now users can define an appropriate timeout for their workflow at startup. #2831

🛠️ Improvements

• Added OIDC SSO quality of life change. Previously, the first role from the roles field would always get picked even if it wasn't in the mappings. Allow multiple roles to be present in the roles field, but only if all those are in role mappings and mapped to the same Memgraph role. #2799
• Allow redirecting to nonlocalhost and HTTPS in SAML SSO. CALLBACK_URL can now use HTTPS and isn't hardcoded to the localhost. #2799

🐞 Bug fixes

• Timeout for SwapMainUUIDRpc message is now properly configured with the default value set to 10s. Users shouldn't expect any difference in the replication behavior. #2826
• Install correct SAML dependencies under MAGE. SAML SSO should now work correctly MAGE. #2829
• EnableWritingOnMainRpc will now be sent when the failover occurs, even if the network identity of some of the replicas isn't stable. Users will now be able to test HA in a docker-compose environment. #2817
• Fixed deadlock caused by write queries and periodic snapshots. 2819
• Broken migration from old durable auth data. Migration is now fixed, and pre-v3.1 auth data can be correctly read. Users can upgrade to v3.1 without encountering authentication problems. 2820
• RPC timeouts for messages requesting promotion, demotion, enabling writing on main and databases' history will now work correctly. Users shouldn't observe long-running RPCs in their HA environment anymore. #2813
• Memgraph DB doesn't know how it will get deployed on K8s. It is possible to abstract multiple coordinators under the same load balancer. In that case, the Bolt server is the same for all coordinators. Previously, the code didn't allow adding multiple coordinators with the same Bolt server. Now, the users will be able to do that. #2801
• Fix the segfault issue caused by a double free. It only happens when Python procedures are called in IN_MEMORY_ANALYTICAL mode. #2804
• The serialization reader would throw if data from the buffer weren't fully ready. That caused network connections between the main and replica to be constantly dying. Now, when the replica doesn't read all data from the buffer, the exception thrown in the SLK part of the code is handled properly, avoiding the connection drop. Additionally, the PR tries to reduce the massive amount of logs produced in HA code when trace log level is used. #2770
• SnapshotRpc, WalFilesRpc, and CurrentWalRpc will now wait for a main lock but for a maximum time of 30s. If the lock cannot be acquired within the 30s on the replica, the main try to replicate again after a few seconds. Such timeout is added to prevent a possible deadlock when multiple fast failover happen during transaction execution while the recovery process is in place. Users should observe the same behavior as before. #2816