ADF Pipeline Error:
DF-Executor-RemoteRPCClientDisassociated
What does this error mean?
A Spark executor lost its RPC connection to the driver — almost always because the executor crashed due to an out-of-memory kill. The data flow fails with this network-level symptom rather than surfacing the underlying OOM.
Common causes
- 1An executor process was killed by the JVM out-of-memory handler and the driver received the disconnection as an RPC error instead of a clean failure message
- 2A large shuffle operation between executors exhausted disk space on a node, causing the executor to terminate abnormally
- 3The Azure IR cluster experienced a transient node failure or eviction — more common with spot-instance IR configurations
- 4A broadcast join attempted to materialize a dataset that exceeded executor heap capacity, triggering a cascading JVM crash
How to fix it
- 1Retry the pipeline — RemoteRPCClientDisassociated is a transient Spark cluster failure where an executor lost contact with the driver; retries succeed.
- 2Increase the Azure IR compute type to provide more memory per executor node, reducing the likelihood of executor crashes under memory pressure.
- 3Check whether the error coincides with a large shuffle operation — reduce partition count or add filtering to reduce the data volume being shuffled between executors.
- 4Disable broadcast joins if the error occurs near a join transformation — large broadcasts can exhaust executor memory and trigger disconnection.
- 5If the error recurs consistently on the same data, check ADF Monitor for any patterns (same time of day, same data volume) to identify the root cause.