Permalink
Commits on Dec 21, 2010
  1. @afeinberg
  2. @afeinberg

    Bump up curr.release

    afeinberg committed Dec 21, 2010
  3. @afeinberg

    Fix an issue where after failure, metadata requests from the server w…

    …ould be received out of order
    afeinberg committed Dec 20, 2010
Commits on Dec 17, 2010
  1. @afeinberg

    Increase curr.release

    afeinberg committed Dec 17, 2010
Commits on Dec 15, 2010
  1. @afeinberg

    Fix unit tests

    afeinberg committed Dec 15, 2010
  2. @afeinberg
  3. @afeinberg

    Disable RepairJob by default

    afeinberg committed Dec 15, 2010
  4. @afeinberg
  5. @afeinberg

    Add MigratePartitions. Use VoldemortConfig#getRebalancingTimeOut() in…

    …stead of AdminClient#getSocketTime()
    afeinberg committed Dec 15, 2010
Commits on Dec 13, 2010
  1. @rsumbaly
  2. @rsumbaly

    Added BDB cleaner threads

    rsumbaly committed Dec 13, 2010
Commits on Dec 11, 2010
  1. @rsumbaly
  2. @rsumbaly
Commits on Dec 10, 2010
  1. @rsumbaly
  2. @rsumbaly

    Added repair thread

    rsumbaly committed Dec 10, 2010
Commits on Dec 6, 2010
  1. @afeinberg
Commits on Dec 3, 2010
  1. @rsumbaly

    a) Modified proximity based handoff strategy to use client zones inst…

    …ead of origin node's zone to reorder the list
    
    b) Made early termination of pusher job better
    rsumbaly committed Dec 3, 2010
  2. @afeinberg
Commits on Dec 2, 2010
  1. @afeinberg
Commits on Nov 19, 2010
  1. @rsumbaly
  2. @afeinberg
Commits on Nov 12, 2010
  1. @rsumbaly

    Bumped up version + Updated jar

    rsumbaly committed Nov 12, 2010
  2. @rsumbaly
  3. @afeinberg

    Bump up release number

    afeinberg committed Nov 12, 2010
  4. @kirktrue @afeinberg

    Fixed a 'leaking sockets' issue wherein the client was creating more …

    …sockets than specified by the configuration.
    
    The root cause was that when connection or protocol negotiation requests in ClientRequestExecutorFactory timed out, we weren't closing the socket before throwing the error. Because the sockets are handled asynchronously, the socket would eventually connect and then be abandoned. Nonetheless, the socket was taking up a thread on the server (using the thread-per-connection socket store implementation). On node bannage (and subsequent pool recycling) this lead to a large number of these "zombie" sockets which left "live" sockets with no available threads on the server.
    kirktrue committed with afeinberg Nov 12, 2010
Commits on Nov 11, 2010
  1. @rsumbaly
Commits on Nov 9, 2010
  1. @afeinberg
  2. @afeinberg

    Bump up release to 0.90.li9

    afeinberg committed Nov 9, 2010
  3. @afeinberg
  4. @afeinberg
  5. @afeinberg
Commits on Nov 8, 2010
  1. @rsumbaly
  2. @rsumbaly
Commits on Nov 5, 2010
  1. @rsumbaly
  2. @rsumbaly