[OAK-6806] RDBDocumentStore: version GC does not scale when there are many docs with long paths

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.7.14, 1.8.0
• Component/s: rdbmk
• Epic Link: RDB Support
• Assignee: Julian Reschke
• Reporter: Julian Reschke
• Created: 11/Oct/17 08:13
• Updated: 12/Jan/18 12:00
• Resolved: 07/Dec/17 14:19

Description

Due to the way the RDB RevisionGC looks for split documents, it doesn't perform well at all in the presence of many documents with long paths.

The reason is that we currently do not have a column for SDTYPE, and thus use pattern matching on the document IDs instead. However, once a document has long path in the document store, it appears to the GC has candidate split document, and thus is always read upon GC.

Issue Links

• is related to: OAK-6839 RDBDocumentStore: queries done by GC processes should not thrash the cache
• is related to: OAK-6809 Allow RevisionGCTest to operate on "long" paths
• relates to: OAK-3878 Avoid caching of NodeDocument while iterating in BlobReferenceIterator
• relates to: OAK-6965 RDBDocumentStore: allow schema evolution part 5: add rows for performant VGC