@Rocky whats FSLIB and STARTLET?

In the context of StarRocks, Starlet and FSLIB are core internal components specifically used in the shared-data (compute-storage separation) architecture. 1. Starlet Starlet is the agent service responsible for managing storage and resources in StarRocks shared-data clusters. It acts as an abstraction layer between the StarRocks execution engine (Compute Nodes/CNs or Backends/BEs) and the underlying cloud-native storage system (often referred to as StarOS). * Role: It manages tablet metadata, data sharding, and communication with the StarManager (the control plane for shared-data). * Service Port: By default, it runs an agent service on port

9070

(

starlet_port

). * Monitoring: StarRocks provides a dedicated "Starlet Dashboard" in Grafana to monitor its health and performance, focusing on how data is being fetched and managed from remote object storage. 2. FSLIB (File System Library) FSLIB is a sub-component or library within Starlet that provides a unified File System Interface. It abstracts the complexities of different storage protocols so StarRocks can interact with various backends using a consistent API. * Role: It handles the actual I/O operations for remote and local storage. It supports: * Remote Storage: S3, HDFS, Google Cloud Storage (GCS), Azure Blob Storage. * Local Caching: Managing the CacheFS (local data cache) to speed up queries by keeping hot data on local disks. * Metrics: When you look at Starlet monitoring, you will see

FSLIB

metrics, which track: * Read/Write IOPS and Throughput: Performance of data transfers to/from S3 or HDFS. * Cache Hit Ratio: How often the system finds data in the local

fslib

cache vs. fetching it from the cloud. * Latency: The time taken for file system operations like

open

,

read

, or

list_dir

. Summary Table Component | Function | Architecture Starlet | Management agent for shared-data/cloud-native storage. | Shared-data only FSLIB | I/O library and abstraction layer for S3/HDFS/Cache. | Shared-data & Data Lake If you are troubleshooting performance in a shared-data cluster, these are the primary areas to check for bottlenecks related to object storage latency or cache efficiency of the local data cache. References • introduction/Architecture.md • administration/management/monitoring/metrics-shared-data.md • deployment/shared_data/feature-support-shared-data.md • quick_start/shared-data.md

@Rocky whats

instance

filter in the starlet fslib grafana dashboard?

In the StarRocks Starlet Dashboard (used for monitoring shared-data or cloud-native storage), the

instance

filter typically has two contexts: 1. The Global Dashboard Filter (Node Selection) Like most StarRocks Grafana dashboards, the

instance

dropdown at the top of the page is a standard Prometheus filter. It allows you to select which specific Backend (BE) or Compute Node (CN) you want to view data for. * Format: It usually appears as

IP:Port

(e.g.,

192.168.1.10:9070

). * Purpose: Since a cluster can have many nodes, this filter helps you isolate performance issues to a specific server. 2. Internal Metric Context (Filesystem Instances) When looking at specific charts within the FSLIB FS METRICS section, you may see the term "instance" referring to internal Filesystem (FS) Instances. * Definition: These are internal software objects created by the FSLIB (File System Library) to manage connections to different storage backends. * Example Metric:

fslib_alive_fs_instances_count

tracks the number of these active objects. * What they represent: FSLIB creates separate instances for different tasks, such as: * An instance for a specific S3 bucket or HDFS cluster. * An instance for the Local Data Cache (CacheFS). * Instances for background tasks like prefetching or data eviction. Summary of Difference Where you see it | Meaning Top Filter Bar | The physical or virtual Node (Server) running the Starlet service. FSLIB Metrics Charts | Internal Filesystem Objects created by the code to handle I/O for specific storage targets. If you notice a high

fslib_alive_fs_instances_count

, it usually indicates that Starlet is managing many concurrent storage connections or cache handles. If you are troubleshooting a specific node's latency, use the

instance

filter at the top to select that node's IP. References • administration/management/monitoring/metrics-shared-data.md • administration/management/BE_configuration.md • administration/management/monitoring/Monitor_and_Alert.md References • administration/management/monitoring/metrics-shared-data.md • administration/management/BE_configuration.md • administration/management/FE_configuration.md • administration/http_interface/query_detail.md