Couchbase Server

Performance Status

06/11/2014

KV

  • MB-11378: Cluster gets stuck in warmup state after re-creation of empty buckets
  • MB-11363: Rebalance after failover (delta recovery mode) fails at the early beginning
  • MB-11347: mem_used stat differs from memory usage reported by tcmalloc and OS
  • MB-11323: Rebalance is 2-3 times slower due to actions not related to backfilling or persistence
  • MB-11287:  Idle server with 30 empty buckets utilizes more than 2000% CPU
  • MB-11279:  UPR-based rebalance (3 -> 4) is extremely slow due to occasionally slow backfilling

    • KV

  • MB-11143:  Avg. BgFetcher wait time is still 5-6 times higher in 3.0 (on slow HDD)  
  • MB-11005:   Erlang memory in KV test case with 10 buckets is too high (up to 30GB, leading to OOM situation)
  • MB-10771: Delta recovery is slower than full recovery. Meanwhile several performance issues encountered.
  • MB-10679: It takes almost 10 minutes to stop Couchbase service in setup with 10 empty buckets
  • MB-9676:  Dramatically increasing latency of SET operations during rebalance tests

    • XDCR

    • MB-10437: XDCR replication rate drops almost to zero in presence of light write workload on src side

    Views

    • MB-11387: Latency of stale=false queries easily reaches 1 minute and more (moderate workloads)
    • MB-11386: Latency of stale=false queries is 50% higher in 3.0 (light workloads)
    • MB-11384: Erlang memory usage in cases with views is 4x higher than in CBEE 2.5.1
    • MB-11204: Leaking memory in performance tests with view queries (both memcached and Erlang)
    • MB-10273:   View compaction doesn't catch up in basic non-DGM tests with view queries
    • MB-9822:  One of nodes is too slow during indexing

    Windows

  • MB-10191: CouchDB crashed due to 'Cannot allocate 467078560 bytes of memory (of type "heap").'
  • MB-9825: Rebalance exited with reason bad_replicas

    • DEFERRED

  • MB-10370: ep-engine deadlock in write-heavy DGM cases
  • MB-9461: In heavy-DGM (<5%) scenarios with views and high cache miss rate rebalance/client operations can fail due to timed out requests to memcached/ep-engine
  • MB-7907:  Issues when scaling XDCR on single node

    • n1ql

    • MB-11141: SELECT COUNT(*) ... WHERE should avoid memcached operations
    • MB-11140: SELECT DISTINCT scans all docs
    • MB-11048: Range queries result in thousands of GET operations/sec

    BACKLOG


    • Multitenancy
    • More "observe" tests
    • MongoDB performance testing
    • [SmartOS] + DTrace

    CBIT ZONE


    • CBIT-1182: client machines for regression tests
    • CBIT-1158: RAID 10 drives for performance testing
    Made with Slides.com