- Newest
- Most votes
- Most comments
Can you post the code that you use for configuring the NeptuneGremlinClusterBuilder? Thank you
GremlinCluster readCluster = NeptuneGremlinClusterBuilder.build() .port(neptuneProperties.getReader().getPort()) .enableSsl(neptuneProperties.isEnableSSL()) .maxContentLength(NeptuneConstants.NEPTUNE_MAX_CONTENT_LENGTH) .maxConnectionPoolSize(neptuneProperties.getReader().getMaxConnectionPoolSize()) .minConnectionPoolSize(neptuneProperties.getReader().getMinConnectionPoolSize()) .maxSimultaneousUsagePerConnection(neptuneProperties.getReader().getMaxSimultaneousUsagePerConnection()) .minSimultaneousUsagePerConnection(neptuneProperties.getReader().getMinSimultaneousUsagePerConnection()) .maxInProcessPerConnection(neptuneProperties.getReader().getMaxInProcessPerConnection()) .minInProcessPerConnection(neptuneProperties.getReader().getMinInProcessPerConnection()) .addContactPoints(refreshAgent.getAddresses().get(endpointsSelector)) .create(); GremlinClient client = readCluster.connect(); refreshAgent.startPollingNeptuneAPI( (OnNewAddresses) addresses -> client.refreshEndpoints(addresses.get(endpointsSelector)), 60, TimeUnit.SECONDS); DriverRemoteConnection connection = DriverRemoteConnection.using(client); GraphTraversalSource g = AnonymousTraversalSource.traversal().withRemote(connection);
When we removed all the extra propeerties as shown below the distribution of load on readReplicas were fine (used all the default properties) GremlinCluster readCluster = NeptuneGremlinClusterBuilder.build() .port(neptuneProperties.getReader().getPort()) .enableSsl(neptuneProperties.isEnableSSL()) .addContactPoints(refreshAgent.getAddresses().get(endpointsSelector)) .create();
While we added content length property to increase it started failing the distribution of load to multiple replicas GremlinCluster readCluster = NeptuneGremlinClusterBuilder.build() .port(neptuneProperties.getReader().getPort()) .enableSsl(neptuneProperties.isEnableSSL()) .maxContentLength(NeptuneConstants.NEPTUNE_MAX_CONTENT_LENGTH) .addContactPoints(refreshAgent.getAddresses().get(endpointsSelector)) .create();
We didnt find any issue or anything mentioned that this should make an issue but as per our analysis that is the cause, any suggesstion would be helpful.
The reader endpoint distributes connections, not individual requests, by changing the instance it points to every 5 seconds. During each 5 second window, every new connection will be directed to the same instance. Large connection pools that are opened eagerly will, therefore, likely attach most of the connections in the pool to a single instance. And because these are long-lived Websocket connections, they will continue forwarding traffic to the same instance for the duration of the connection.
More details on this, and some recommendations for mitigating are described here: https://aws.amazon.com/blogs/database/load-balance-graph-queries-using-the-amazon-neptune-gremlin-client/
maxContentLength shouldn't have any impact on the endpoint selection logic – it's used purely to configure the frame size for responses from the server.
One more question: what endpoiuntSelector are you using? Is it an EndpoiuntsType enum value (and if so, which one)? Or is it a custom selector? Thanks
Relevant content
- Accepted Answerasked 2 years ago
- Accepted Answer
asked 9 months ago - asked a year ago
AWS OFFICIALUpdated 3 months ago- AWS OFFICIALUpdated 6 months ago
- AWS OFFICIALUpdated a year ago
- AWS OFFICIALUpdated a year ago
Max and MIN connection pooling size we have given is 20 and 100 only, while we have tried using both gremlin-client classes
NeptuneGremlinClusterBuilder and GremlinClusterBuilder
We are passing all the reader with the cluster but still the issue is same.
We have there is one more method with is taking input of endpoint networkLoadBalancerEndpoint(networkLoadBalancerEndpoint) not sure how to configure this or from where we will get this endpoint as of now. Still trying to debug and find the solution.