Details
-
Bug
-
Resolution: Unresolved
-
Critical
-
None
-
Security Level: Public
-
None
-
CBG Sprint 15, CBG Sprint 16, CBG Sprint 19
-
5
Description
1.x push replications (using _bulk_docs) reserve blocks of sequences in order to reduce the number of incr operations being performed.
If a document included in the bulk_docs batch fails before a sequence is allocated (most commonly due to Sync Function rejection, but potentially a few additional corner cases associated with invalid data), the sequence is not released, and won't be used until the next document write.
When a SG node is under low load, this can result in the sequence being abandoned after CacheSkippedSeqMaxWait, but subsequently allocated by the node on the next write. This may result in the document not being replicated until the document is next updated.
Attachments
Issue Links
- depends on
-
CBG-258 Spec and implementation for more efficient sequence allocation
- Resolved