Details
-
Bug
-
Resolution: Won't Fix
-
Critical
-
Cheshire-Cat
-
None
-
Untriaged
-
1
-
Unknown
Description
Currently, indexer uses multiple cluster info client objects i.e. each component in indexer has a separate cluster info client object. When there is a pool/service change notification, all these clients would refresh their cache resulting in significant load at ns_server.
Instead, indexer/projector process can maintain only on global cluster info client rather than initialising one per component. To do this, the updation of the cache data has to be made "Copy-On-Write".
With out copy-on-write, it may not be possible to maintain a global cluster info cache as it would lead to significant RLock and WLock contention between different components.
Attachments
Issue Links
- blocks
-
MB-47506 Reduce polling by services
- Resolved
- Clones
-
MB-43886 Usage of global cluster info client object
- Closed
- relates to
-
MB-49446 Update in terse bucket streaming endpoint with only change in rev field
- Closed
-
MB-49448 Add serverGroup.Name in poolsStreaming endpoint update
- Closed
-
MB-49450 Add a hash value of "nodes" data in pools/default endpoint
- Closed
-
MB-49451 Addition of "nodeServicesURI" in pools/default endpoint with hash value on nodesExt data
- Closed