I have simple process.
For the test I start some script for starting this process each 10 minutes.
Input data are always the same and suppose success scenario.
Receive Task has correct subscription with “name” and “correlationKey”.
Timer Boundary Event - 3 minutes. It’s much enough to download.
Test runs well but approximately 1 from 10 processes I’ve got timeout by Timer Boundary Event.
By java-logs for these cases RabbitMQ consumer works well: download file and sends success PublishMessageCommand with correct data at the end, but Receive Task doesn’t get it.
My “correlationKey” - some UUID, unique for each process instance, process gets it like a variable during the start process.
I’ve noticed a strange metadata for these unsuccessful Receive Tasks. This data you can see in Operate clicked by task and then clicked by “view” in details window.
In each case:
There aren’t fields “messageName” and “correlationKey”.
I call these cases “fake” timeout because consumer works well and sends message.
Metadata is correct when I get real timeout or success processing:
Fields “messageName” and “correlationKey” are present.
My question is why RECEIVE TASK sometimes can loose correlation information.
Camunda 8.5 self-managed
cluster with 3 nodes and 3 partitions