What causes CANNOT_LOAD_STATE_STORE?

The checkpoint location in cloud storage is inaccessible, corrupted, or was deleted. The cluster running the stream was terminated while a state store write was in progress, leaving a partial checkpoint. A schema change in the stateful operation is incompatible with the existing state store format. Cloud storage credentials used to access the checkpoint location have expired or been revoked. The state store directory was manually deleted or overwritten

How do I fix CANNOT_LOAD_STATE_STORE?

Check that the checkpoint location in cloud storage is accessible from the cluster (use dbutils.fs.ls()). Verify cloud storage credentials: the cluster IAM role or managed identity must have read/write access to the checkpoint path. If the checkpoint is corrupted, consider resetting: delete the checkpoint directory and restart the stream from the beginning or an offset. Do not change stateful operation schemas without first deleting the checkpoint — state store schemas are not forward-compatible. Investigate the storage path for partial or incomplete files from a previous interrupted write

High severitystreaming

Power BI Refresh Error:
CANNOT_LOAD_STATE_STORE

What does this error mean?

A Structured Streaming query failed to load the state store, which persists stateful operation data (e.g., aggregations, joins, deduplication) between micro-batches. Without the state store, the streaming job cannot resume.

Common causes

1The checkpoint location in cloud storage is inaccessible, corrupted, or was deleted
2The cluster running the stream was terminated while a state store write was in progress, leaving a partial checkpoint
3A schema change in the stateful operation is incompatible with the existing state store format
4Cloud storage credentials used to access the checkpoint location have expired or been revoked
5The state store directory was manually deleted or overwritten

How to fix it

1Check that the checkpoint location in cloud storage is accessible from the cluster (use dbutils.fs.ls())
2Verify cloud storage credentials: the cluster IAM role or managed identity must have read/write access to the checkpoint path
3If the checkpoint is corrupted, consider resetting: delete the checkpoint directory and restart the stream from the beginning or an offset
4Do not change stateful operation schemas without first deleting the checkpoint — state store schemas are not forward-compatible
5Investigate the storage path for partial or incomplete files from a previous interrupted write

Frequently asked questions

Can I resume a streaming job after deleting the checkpoint?

Yes, but the stream will restart from the beginning (or the latest offset, depending on configuration). You may miss data or reprocess data depending on your source. Always back up the checkpoint before deleting.

How do I prevent checkpoint corruption?

Use a dedicated checkpoint directory per streaming query, never share checkpoint paths between queries, and ensure cloud storage has high durability enabled. Avoid stopping clusters with SIGKILL during active micro-batch writes.