Permalink
Commits on Jan 4, 2016
  1. Explicitly move to using llvm::legacy::FunctionPassManager to clean u…

    …p compilation hiccup using Clang
    lairofthegoldinblair committed Jan 4, 2016
Commits on Oct 15, 2015
  1. When creating a large model string literal make sure the string itsel…

    …f is allocated with a null terminator.
    lairofthegoldinblair committed Oct 15, 2015
Commits on Sep 22, 2015
  1. Fixed a bug with VARCHAR literals that seems to be LLVM version (or b…

    …uild type?) specific. Basically we were declaring a global variable for a varchar literal using a struct type but then intializing with a constant value with the corresponding anonymous struct. This trigger a type mismatch assertion in LLVM
    lairofthegoldinblair committed Sep 22, 2015
Commits on Aug 25, 2015
Commits on Jun 24, 2015
  1. Merge pull request #1 from akamai/master

    Merge pull request #1 from akamai-tech/master
    lairofthegoldinblair committed Jun 24, 2015
  2. Merge pull request #1 from akamai-tech/master

    Pull from akamai-tech
    lairofthegoldinblair committed Jun 24, 2015
Commits on Jun 17, 2015
  1. Hadoop/HDFS/MapReduce support is now optional. Use CMAKE_DISABLE_FIND…

    …_PACKAGE_HADOOP to build without Hadoop if it is present on your system and you don't want to use it.
    lairofthegoldinblair committed Jun 17, 2015
Commits on Jun 16, 2015
Commits on Mar 23, 2015
Commits on Apr 30, 2014
  1. Changes to build with more recent Boost libraries (tested against 1.5…

    …5.0) and some smaller miscellaneous changes.
    lairofthegoldinblair committed Apr 30, 2014
Commits on Dec 17, 2013
  1. Maintain a few internal statistics in the scheduler that help one und…

    …erstand the IO service behavior
    lairofthegoldinblair committed Mar 11, 2013
  2. Handle query strings with newlines and treat these as multiple query …

    …strings so that we can handle Akamai aggregated download receipts. I'm not sure if such functionality makes good sense in general but it seems to be necessary. Also fixed egregious memory leak.
    lairofthegoldinblair committed Mar 8, 2013
  3. Have the HTTP read operator properly use the new API for requesting a…

    … mix of reads and writes to fix some pretty broken behavior in how this was working previously.
    lairofthegoldinblair committed Mar 8, 2013
  4. Two new pieces of functionality. First is the ability to have an oper…

    …ator request a mix of read and write requests. The second is the ability of the scheduler to flush partially full local write buffers. It does this prior to any blocking IO waits so we get low latency processing even when there isn't a big fat stream of data pushing things through. The current implementation of this idea is rather simplistic and more experimentation will be necessary to see if it navigates the latency/throughput tradeoff curve successfully
    lairofthegoldinblair committed Mar 8, 2013
  5. Raw TCP operators and integration with Boost ASIO. The operators them…

    …selves aren't well motivated at this point (ncat like functionality) but are just serving as a testbed for how to integrate the dataflow scheduler with an event demuxer like ASIO
    lairofthegoldinblair committed Feb 11, 2013
  6. A gzip decompressor implemented as an operator: byte stream to byte s…

    …tream. Initial prototype indicates that using the file operator to cat into a gunzip operator is faster than the equivalent using Unix shell (which is should be because there is no IPC). There is still a ways to go but I was happy that the operator abstraction is performing well enough to support this increased modularity.
    lairofthegoldinblair committed Jan 30, 2013
  7. Operator using the initial async file service implementation and a re…

    …factored parser to boot. The parser and file operator connect using a stream channel. This should allow us to reuse the parser with any byte stream source (e.g. a socket operator)
    lairofthegoldinblair committed Jan 30, 2013
  8. A stab at an async file service interface and implementation. Totally…

    … in flux as next steps will be to rationalize this with networked async I/O (probably boost::asio specifically).
    lairofthegoldinblair committed Jan 30, 2013