This is an earlier version of the OpenSearch documentation. For the latest version, see the current documentation. For information about OpenSearch version maintenance, see Release Schedule and Maintenance Policy.
Shard indexing backpressure adds several settings to the standard OpenSearch cluster settings. They are dynamic, so you can change the default behavior of this feature without restarting your cluster.
The high-level controls allow you to turn the shard indexing backpressure feature on or off.
| ||False||Change to |
| ||False||Run shard indexing backpressure in shadow mode or enforced mode. In shadow mode (value set as |
Node-level limits allow you to control memory usage on a node.
| ||70%||Define the percentage of the node-level memory threshold that acts as a soft indicator for strain on a node.|
Shard-level limits allow you to control memory usage on a shard.
| ||0.001d||Specify the minimum assigned quota for a new shard in any role (coordinator, primary, or replica). Shard indexing backpressure increases or decreases this allocated quota based on the inflow of traffic for the shard.|
| ||75%||Specify the lower occupancy limit of the allocated quota of memory for the shard. If the total memory usage of a shard is below this limit, shard indexing backpressure decreases the current allocated memory for that shard.|
| ||85%||Specify the optimal occupancy of the allocated quota of memory for the shard. If the total memory usage of a shard is at this level, shard indexing backpressure doesn’t change the current allocated memory for that shard.|
| ||95%||Specify the upper occupancy limit of the allocated quota of memory for the shard. If the total memory usage of a shard is above this limit, shard indexing backpressure increases the current allocated memory for that shard.|
Performance degradation factors
The performance degradation factors allow you to control the dynamic performance thresholds for a shard.
| ||2,000||The number of requests in the sampling window size on a shard. Shard indexing backpressure compares the overall performance of requests with the requests in the sample window to detect any performance degradation.|
| ||5x||The degradation factor per unit byte for a request. This parameter determines the threshold for any latency spikes. The default value is 5x, which implies that if the latency shoots up 5 times in the historic view, shard indexing backpressure marks it as a performance degradation.|
| ||300000 ms||The amount of time a request is pending in a cluster. This parameter helps identify any stuck-request scenarios.|
| ||100||The maximum number of pending requests in a cluster.|