Disk write protection overview
Fluss includes a built-in disk write protection mechanism. When the disk usage of a TabletServer node reaches a threshold, the system automatically rejects write requests to prevent a full disk from causing a node crash or data corruption.
The write protection capability requires Fluss engine version 0.9-ali-3.0 or later.
Trigger Rules
Each TabletServer node independently monitors its disk usage and executes protection policies according to the following rules:
-
Disk usage < 85%: Writes are normal and unrestricted.
-
Disk usage ≥ 85%: The node enters a write-prohibited state and rejects all write operations, including log appends and primary key table updates. Flink jobs that write to Fluss will receive an error and automatically retry until the disk usage returns to normal.
-
Disk usage drops below 75%: The system automatically removes the write-prohibited state and resumes normal write operations.
Write protection takes effect independently at the node level. If any node in a cluster reaches the threshold, write protection is triggered only for that node and does not affect writes to other nodes.
Resume writes
When a cluster triggers disk write protection, you can reduce disk usage and resume writing in the following ways:
-
Expand disks: Scales up the disks of a cluster in the console to expand storage capacity.
-
Delete historical partitions: Deletes historical partitions that are no longer needed from tables and frees up disk space.
-
Delete unused tables: Delete tables that are no longer in use to free up disk space.
When disk usage drops below 75%, the system automatically resumes write operations. No manual intervention is required.
Monitoring recommendations
We recommend that you configure disk usage alerts for your cluster in the console with a recommended alert threshold of 70% so that you can take action before write protection is triggered.