Amazon Neptune Engine Version (2023-03-08) - Amazon Neptune

Amazon Neptune Engine Version (2023-03-08)

As of 2023-03-08, engine version is being generally deployed. Please note that it takes several days for a new release to become available in every region.


If upgrading from an engine version earlier than

  • Engine release introduced a new format for custom parameter groups and custom cluster parameter groups. As a result, if you are upgrading from an engine version earlier than to engine version or above, you must re-create all your existing custom parameter groups and custom cluster parameter groups using parameter group family neptune1.2. Earlier releases used parameter group family neptune1, and those parameter groups won't work with release and above. See Amazon Neptune parameter groups for more information.

  • Engine release also introduced a new format for undo logs. As a result, any undo logs created by an earlier engine version must be purged and the UndoLogsListSize CloudWatch metric must fall to zero before any upgrade from a version earlier than can get started. If there are too many undo log records (200,000 or more) when you try to start an update, the upgrade attempt can time out while waiting for purging of the undo logs to complete.

    You can speed up the purge rate by upgrading the cluster's writer instance, which is where the purging occurs. Doing that before trying to upgrade can bring down the number of undo logs before you start. Increasing the size of the writer to a 24XL instance type can increase your purge rate to more than a million records per hour.

    If the UndoLogsListSize CloudWatch metric is extremely large, opening a support case may help you explore additional strategies for bringing it down.

  • Finally, there was a breaking change in release affecting earlier code that used the Bolt protocol with IAM authentication. Starting with release, Bolt needs a resource path for IAM signing. In Java, setting the resource path might look like this: request.setResourcePath("/openCypher"));. In other languages, the /openCypher can be appended to the endpoint URI. See Using the Bolt protocol for examples.

Subsequent Patch Releases for This Release

New Features in This Engine Release

  • Added support for TinkerPop 3.6.2, which adds many new Gremlin features such as the new mergeV(), mergeE(), element(), and fail() steps. The mergeV() and mergeE() steps are of particular note as they offer a long-awaited declarative option for performing upsert-like operations, which should greatly simplify existing code patterns and make Gremlin easier to read. The 3.6.x version also added regex predicates, a new overload to the property() step which takes a Map, and a major revision of by() modulation behavior that is far more consistent across all steps which use it.

    See the TinkerPop change log and upgrade page for information about the changes in version 3.6 and things to consider when upgrading.

    If you are using fold().coalesce(unfold(), <mutate>) for conditional inserts, we recommend that you migrate to the new mergeV/E() syntax, described here and here. Neptune uses a narrower locking pattern for Merge than for Coalesce, which can reduce concurrent modification exceptions (CMEs).

    For more information about the new features available in this TinkerPop release, see Stephen Mallette's blog, Exploring new features of Apache TinkerPop 3.6.x in Amazon Neptune.

  • Added support for R6i instance types, powered by 3rd-generation Intel Xeon Scalable processors. These are an ideal fit for memory-intensive workloads and offer up to 15% better compute/price performance and up to 20% higher memory bandwidth per vCPU than comparable R5 instance types.

  • Added graph summary API endpoints for both property graphs and RDF graphs, that let you get a fast summary report about your graph.

    For property (PG) graphs, the graph summary API provides a read-only list of node and edge labels and property keys, along with counts of nodes, edges, and properties. For RDF graphs, it provides a list of classes and predicate keys, along with counts of quads, subjects, and predicates.

    The following changes went along with the new graph summary API:

    • Added a new GetGraphSummary dataplane action.

    • Added a new rdf/statistics endpoint to replace the sparql/statistics endpoint, which is now deprecated.

    • Changed the name of the summary field in the statistics status response to signatureInfo, so as not to confuse it with graph summary information. Previous engine versions continue to use summary in the JSON response.

    • Changed the precision of the date field in the statistics status response from minute to millisecond. The previous format was 2020-05-07T23:13Z (minute precision), while the new format is 2023-01-24T00:47:43.319Z (millisecond precision). Both are ISO 8601 compliant, but this change may break existing code, depending on how the date is being parsed.

    • Added a new %statistics line magic in the Workbench that lets you retrieve DFE engine statistics.

    • Added a new %summary line magic in the Workbench that lets you retrieve graph summary information.

  • Added slow-query logging to log queries that take longer to execute than a specified threshold. You enable and control slow-query logging using the two new dynamic parameters, namely neptune_enable_slow_query_log, and neptune_slow_query_log_threshold.

  • Added support for two dynamic parameters, namely the new cluster parameters, neptune_enable_slow_query_log, and neptune_slow_query_log_threshold. When you make a change to a dynamic parameter, it takes effect immediately, without requiring any instance reboot.

  • Added a Neptune-specific openCypher removeKeyFromMap() function that removes a specified key from a map and returns the resulting new map.

Improvements in This Engine Release

  • Extended Gremlin DFE support to limit steps with local scope.

  • Added by() modulation support for the Gremlin DedupGlobalStep in the DFE engine.

  • Added DFE support for Gremlin SelectStep and SelectOneStep.

  • Performance improvements and correctness fixes for various Gremlin operators, including repeat, coalesce, store, and aggregate.

  • Improved performance of openCypher queries involving MERGE and OPTIONAL MATCH.

  • Improved performance of openCypher queries involving UNWIND of a list of maps of literal values.

  • Improved performance of openCypher queries that have an IN filter for id. For example:

    MATCH (n) WHERE id(n) IN ['1', '2', '3'] RETURN n
  • Added the ability to specify the base IRI for SPARQL queries using the BASE statement (see Default base IRI for queries and updates).

  • Shortened the load processing wait time for Gremlin and openCypher edge-only bulk loads.

  • Made bulk loads resume asynchronously when Neptune restarts to avoid a lengthy wait time caused by Amazon S3 connectivity issues before failing resume attempts.

  • Improved handling of SPARQL DESCRIBE queries that have the describeMode query hint set to "CBD" (concise bounded description) and that involve a large number of blank nodes.

Defects Fixed in This Engine Release

  • Fixed an openCypher bug where queries returned the string, "null", instead of a null value in Bolt and SPARQL-JSON.

  • Fixed an openCypher bug in list comprehension that produced a null value rather than the values provided for the list elements.

  • Fixed an openCypher bug where byte values were not correctly serialized.

  • Fixed a Gremlin bug in UnionStep that occurred when an input was an edge traversing to a vertex inside a child traversal.

  • Fixed a Gremlin bug that caused a step label associated with UnionStep not to propagate correctly to the last step of each child traversal.

  • Fixed a Gremlin bug for the dedup step with labels following a repeat step, where the labels attached to the dedup step weren't available to use in query further.

  • Fixed a Gremlin bug where translating the repeat step inside a union step failed with an internal error.

  • Fixed Gremlin correctness issues for DFE queries with limit as a child traversal of non-union steps by falling back to Tinkerpop. Queries in a form like this are affected:

    g.withSideEffect('Neptune#useDFE', true).V().as("a").select("a").by(out().limit(1))
  • Fixed a SPARQL bug where SPARQL GRAPH patterns would not consider the dataset supplied by a FROM NAMED clause.

  • Fixed a SPARQL bug where SPARQL DESCRIBE with some FROM and/or FROM NAMED clauses did not always correctly use data from the default graph and sometimes threw an exception. See SPARQL DESCRIBE behavior with respect to the default graph.

  • Fixed a SPARQL bug so that the correct exception message is returned when null characters are rejected.

  • Fixed a SPARQL explain bug that affected plans containing a PipelinedHashIndexJoin operator.

  • Fixed a bug that caused an internal error to be thrown when a query that returns a constant value was submitted.

  • Fixed an issue with deadlock detector logic that occasionally made the engine unresponsive.

Query-Language Versions Supported in This Release

Before upgrading a DB cluster to version, make sure that your project is compatible with these query-language versions:

  • Gremlin earliest version supported: 3.6.2

  • Gremlin latest version supported: 3.6.2

  • openCypher version: Neptune-9.0.20190305-1.1

  • SPARQL version: 1.1

Upgrade Paths to Engine Release

You can manually upgrade to this release from any previous Neptune engine release greater than or equal to


Starting with engine release, all custom parameter groups and custom cluster parameter groups that you were using with engine versions earlier than must now be re-created using parameter group family neptune1.2. Previous releases used parameter group family neptune1, and those parameter groups will not work with releases from onwards. See Amazon Neptune parameter groups for more information.

You will not be automatically upgraded to this major version release.

Upgrading to This Release

Amazon Neptune is now generally available.

If a DB cluster is running an engine version from which there is an upgrade path to this release, it is eligible to be upgraded now. You can upgrade any eligible cluster using the DB cluster operations on the console or by using the SDK. The following CLI command will upgrade an eligible cluster immediately:

For Linux, OS X, or Unix:

aws neptune modify-db-cluster \ --db-cluster-identifier (your-neptune-cluster) \ --engine-version \ --apply-immediately

For Windows:

aws neptune modify-db-cluster ^ --db-cluster-identifier (your-neptune-cluster) ^ --engine-version ^ --apply-immediately

Updates are applied to all instances in a DB cluster simultaneously. An update requires a database restart on those instances, so you will experience downtime ranging from 20–30 seconds to several minutes, after which you can resume using the DB cluster.

Always test before you upgrade

When a new major or minor Neptune engine version is released, always test your Neptune applications on it first before upgrading to it. Even a minor upgrade could introduce new features or behavior that would affect your code.

Start by comparing the release notes pages from your current version to those of the targeted version to see if there will be changes in query language versions or other breaking changes.

The best way to test a new version before upgrading your production DB cluster is to clone your production cluster so that the clone is running the new engine version. You can then run queries on the clone without affecting the production DB cluster.

Always create a manual snapshot before you upgrade

Before performing an upgrade, we strongly recommend that you always create a manual snapshot of your DB cluster. Having an automatic snapshot only offers short-term protection, whereas a manual snapshot remains available until you explicitly delete it.

In certain cases Neptune creates a manual snapshot for you as a part of the upgrade process, but you should not rely on this, and should create your own manual snapshot in any case.

When you are certain that you won't need to revert your DB cluster to its pre-upgrade state, you can explicitly delete the manual snapshot that you created yourself, as well as the manual snapshot that Neptune might have created. If Neptune creates a manual snapshot, it will have a name that begins with preupgrade, followed by the name of your DB cluster, the source engine version, the target engine version, and the date.


If you are trying to upgrade while a pending action is in process, you may encounter an error such as the following:

We're sorry, your request to modify DB cluster (cluster identifier) has failed. Cannot modify engine version because instance (instance identifier) is running on an old configuration. Apply any pending maintenance actions on the instance before proceeding with the upgrade.

If you encounter this error, wait for the pending action to finish, or trigger a maintenance window immediately to let the previous upgrade complete.

For more information about upgrading your engine version, see Maintaining your Amazon Neptune DB Cluster. If you have any questions or concerns, the AWS Support team is available on the community forums and through AWS Premium Support.