Handle ServiceUnavailableException from the Java Bolt Driver

[nandosola]

Hi,

As per neo4j/neo4j-java-driver#302, the Java Driver is launching ServiceUnavailableException in case the client loses connection with the server (eg. network glitches, server restarts, …). AFAIK the philosophy is that the libs and apps using the driver should catch ServiceUnavailableException.

However, I noticed that neo4j-ogm does not handle ServiceUnavailableException yet. The only reference to a server disconnection I've found is ClientException being rethrown as a CypherException:

• BoltRequest.java
• BoltResponse.java

It seems that CypherException is too generic. For instance, spring-data-neo4j handles it this way:

org.springframework.data.neo4j.exception.UncategorizedNeo4jException: Error executing Cypher; 
  Code: N/A; Description: SSL Connection terminated while receiving data. 
  This can happen due to network instabilities, or due to restarts of the database.; 

nested exception is org.neo4j.ogm.exception.CypherException: Error executing Cypher; 
  Code: N/A; Description: SSL Connection terminated while receiving data. 
  This can happen due to network instabilities, or due to restarts of the database.

So they don't have the information that a ServiceUnavailableException conveys, which for spring-data-neo4j could mean converting it to a RecoverableDataAccessException so that the connection could be retried (see: spring-projects/spring-data-neo4j#373)

In my (maybe naïve) debugging session, I've found that a possible solution could be creating a org.neo4j.ogm.exception.ServiceCommunicationException that can be used by the upper layers (eg. spring-data-neo4j). But where should the exception be launched?

Well, it seems that both BoltRequest.java and BoltResponse.java don't check for stale sessions:

transactionManager.getCurrentTransaction() == null

Wouldn't be enough, since the connection is stalled (hence not null) and the failure happens only after the request is executed or the results are fetched.

[mangrish]

Related to #301.

I think we should try to add this fix to the 2.1.2 release but it would be nice if we can sneak it in to 2.1.1.

[obriensystems]

I get this error only when running against the bolt port of an embedded instance (out and back in) using the java driver
When I run the same code on a server-mode instance it works fine

[ikwattro]

This exception will only be thrown from a neo4j 3.1 + instance btw using Bolt.

[nsteenv]

Same issue here, trying to connect to an embedded instance with Bolt and java driver, any hint on how to do that?

[chrisbetz]

Hi. Having a similar issue: this happens to me only on Java 9, not on Java 8. Any hints on this?

[michael-simons]

Thanks for your feedback.

The Bolt transport will now translate this into org.neo4j.ogm.exception.ConnectionException. We plan on further refining in this area as well.

Regarding running embedded and opening a bolt port to the outside: We strongly recommend not doing this. Neo4j-OGM uses the database api to communicate with the embedded instance, too.