Quick Start
Simulate Time Series Scenarios
Standard Cluster Deployment
Data Modeling
Connecting to The database
Data Writing
Data Migration
Data Query
Scene Application Examples
Federal Query
Maintenance and Monitoring
Global Maintenance
Partition Maintenance
Backup and Restore
Cluster Expansion
Monitoring
Performance Tuning
Troubleshooting
Reference Guide
Tool Guide
Data Type
Storage Engine
Execution Engine
Configuration Parameters
SQL Reference
FAQ
This document describes the fault tolerance mechanism in MatrixGate.
MatrixGate internally uses external tables to insert data into the target table in micro-batches. Therefore, data is ingested in batches, and if any single row has a formatting error, the entire batch would previously fail.
Starting from version 4.3, MatrixGate introduces a fault tolerance mechanism. A single malformed row no longer prevents the rest of the data from being ingested. MatrixGate returns information about the erroneous row and logs the error.
Note!
Fault tolerance applies only to data formatting errors. If a constraint violation occurs (e.g., duplicate key in a unique index), the entire batch will still fail.
Unlike earlier versions, where any invalid data would result in an HTTP 500 response, MatrixGate now returns HTTP 200 even when errors occur. The response body includes details about the problematic rows. For example:
At line: 2
missing data for column "c3"Fault tolerance is not unlimited. The tolerance threshold is controlled by the GUC parameter: gp_reject_limit. When the number of error rows exceeds 5 * gp_reject_limit, the batch insert fails.
For more detailed usage, refer to MatrixGate.