[HDDS-7223] Caching container location information in OM

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.4.0
• Component/s: Ozone Manager
• Labels: Performance
• Assignee: Duong
• Reporter: Duong
• Created: 15/Sep/22 00:02
• Updated: 08/May/23 16:49
• Resolved: 08/May/23 16:49

Description

A container can move and change the Datanode details in it's lifecycle post close of a container.

When a key needs to be read by the client, OM queries SCM to update the location information for a container. This adds additional latency per request and load on SCM which is directly proportional to the read load on OM.

This Jira tracks the design and implementation of improving read performance by addressing this code path.

 

Issue Links

• causes: HDDS-8496 Hive with s3a connector for OBS bucket hits error

Sub-Tasks

• Introduce container location cache in ScmClient: Resolved
• Implement GetKeyInfo API: Resolved
• Integrate the GetKeyInfo API to key read flows: Resolved
• Experiment the location cache size: Resolved
• Add container location cache metrics.: Resolved
• Normalize DatanodeDetails to reduce cache memory footprint.: Resolved
• S3G: Avoid calling getS3Volume by using GetKeyInfo API.: Resolved
• Integrate the GetKeyInfo API to OFS: Resolved
• Use container cache in Key listing API.: Resolved