[GEODE-10012] Avoid retries for non-HA function execution

• Type: Bug
• Status: In Progress
• Priority: Major
• Resolution: Unresolved
• Component/s: native client
• Labels: pull-request-available
• Assignee: Mario Salazar de Torres
• Reporter: Mario Salazar de Torres
• Created: 02/Feb/22 05:42
• Updated: 17/Nov/23 16:54

Description

GIVEN a cluster with 3 servers and 1 locator
AND a PartitionedRegion with redundant-copies="1"
AND a user-defined Function with isHA=false
AND a geode-native client with a pool with single-hop enabled and retry attempts set to 2
WHEN execution fails due to a connectivity issue/connection timeout/internal cluster error
THEN function is retried twice

---

Additional information. The function is retried at the network code layer, and given whenever there is a timeout/connectivity error, the BSL is removed from the ClientMetadataService, whenever retries happen, you might get a InternalFunctionInvocationTargetException exception thrown indicating: 

Multiple target nodes found for single hop operation 

Also, have into account that Java client API behaviour never retries a Function Execution if isHA=false for that function, and whenever isHA=true, the function is retried but on the function execution logic and not at the networking layer.

 

Issue Links

• links to: GitHub Pull Request #920