@Rocky Is the recommendation to use

ext4

Linux filesystem still current for the BE/FE storage volumes ? Should we consider for example

xfs

instead ?

Yes, the recommendation to use

xfs

over

ext4

is now the preferred approach for StarRocks BE and FE storage volumes, especially in production environments where performance is a priority. While the documentation still lists both

ext4

and

xfs

as supported journaling filesystems, the official stance has shifted toward

xfs

because StarRocks typically performs better with it, specifically in terms of QPS (Queries Per Second) and high-concurrency write scenarios. Why

xfs

is considered superior for StarRocks: * Parallel I/O & Scalability:

xfs

was designed for high-performance parallel I/O. It handles concurrent reads and writes much better than

ext4

by using "Allocation Groups," which reduce lock contention during data allocation. This is critical for BE nodes that handle multiple concurrent data ingestion and query tasks. * Large File Handling: StarRocks BEs store data in large, immutable files (segments).

xfs

is optimized for these workloads, offering better throughput and less fragmentation over time for large-file operations. * Dynamic Inode Allocation: Unlike

ext4

, which pre-allocates a fixed number of inodes at creation time (potentially leading to "out of space" errors even when disk space is available),

xfs

allocates inodes dynamically as needed. Recommendations for FE (Frontend) For FE nodes, which store metadata using BDB JE (Berkeley DB Java Edition),

xfs

is also recommended. Metadata operations often involve frequent, small, synchronous writes to the journal and image files.

xfs

provides stable performance for these metadata-heavy operations. Additional Storage Tuning Regardless of the filesystem you choose, StarRocks recommends tuning the I/O scheduler at the OS level to match your hardware: * For SSD/NVMe: Use the

kyber

or

none

(noop) scheduler. * For SATA (HDD): Use the

mq-deadline

scheduler. You can check your current scheduler with:

bash
cat /sys/block/${disk}/queue/scheduler

Summary Table Feature |

ext4

|

xfs

| Recommendation Performance | Good (General Purpose) | Better (High QPS) |

xfs

Large Files | Good | Excellent |

xfs

Scalability | Standard | High Parallelism |

xfs

Resizing | Can grow/shrink | Can only grow |

ext4

(if shrinking is needed) Conclusion: If you are setting up a new cluster or adding new volumes,

xfs

is the current best-practice recommendation for StarRocks. References • faq/operation_maintenance_faq.md • deployment/environment_configurations.md • administration/management/FE_configuration.md • faq/Others.md