1.12.002 (2014-12-19)

     -- attempt to fix bugs: 89037, 100594, 101012

     -- cpan tester report 586f5968-8489-11e4-adb9-802de0bfc7aa is - ugly - fixed too

     -- refactoring: Client/Worker base class added

     -- dynamic tests by using environment variable GEARMAN_SERVERS

1.12 (2014-12-14)

     -- Repoint HACKING file.

     -- Make a jobserver connection use the command timeout during exception negotiation.

     -- Make $taskset->add_task use the command timeout to not hang during job submission.

     -- Add a client option 'command_timeout' to indicate how long we should wait before considering a
        gearman command to have failed. This not the same as a job timeout, and only affects commands
        that should generally not block apart from the roundtrip on the network.

     -- When a connection to a gearman server fails, start counting how many failures we've had and
        do an exponential backoff (1s, 4s, 9s, 16s...) to a maximum of 90 seconds (default) that we
        treat the server as 'gone' quickly.

     -- Change dispatch_background to share code paths with other dispatching, this will make
        background jobs now be hashed to a particular server (rather than handed to a random server)

1.11 (2010-01-17)

     -- Switch read_res_packet to use sysread with a large buffer and offsets. This improves
        memory usage (sometimes drastically) and performance.

     -- Fix from Martin Atkins, don't print() in the Gearman::Util code and close sockets
        when you have an error of some kind (indicating we're out of sync)

1.10 (2009-10-04)

     -- Make workers wake up periodically for a particular server to make sure they aren't
        stale connections. This happened naturally (although at relatively low interval) in
        previous releases.

     -- Help prevent leaks when Gearman::Task->add_hook is used.

     -- Change worker to only 'wake up' against a gearmand that has woken it up, this prevents
        a constant flood of pre-sleep command from arriving at an otherwise silent gearmand.

     -- Fix issue where prefixes were double-prepended on function names on worker
        upon reconnect.

     -- Fix issue where the other response parser code in Util would silently truncate
        argument strings longer than 20*the socket buffer size.

     -- Fix issue where the ResponseParser class would not correctly handle messages with
        zero-length bodies.

     -- Make the Gearman::Task class autoload Storable and fail gracefully when it's not
        loadable.

     -- Initial fold-in of exceptions support in the gearman client, makes an option
        to the gearman server to enable it, and is disabled by default. Workers will
        will attempt to throw exceptions anytime Storable is available.

     -- fix jobs of > 32kilobytes in size so they work properly (workers would disconnect
        when a job greater than 32kb would arrive.)

     -- expose the time that the last job was processed in the stop_if hook of worker.
        Since a jobserver wakes up all workers in the case of a new job to be processed
        the concept of is_idle doesn't actually reflect if a worker has procssed jobs,
        rather it indicates whether the jobserver has been silent for 10 seconds.

     -- change server polling order in workers to start at a random point in the list
        during every worker pass. So we drain jobs from all servers rather than
        working on each of them in order.

1.09 (2007-06-29)

     -- document the license and copyright

1.08 (2007-05-21)

     -- add "package Gearman::Objects" line to Objects.pm, to ease RPM
        building

     -- reserve the packet number for "all_yours", not yet implemented

     -- In pack_res_command, silence any warnings about 'undefined' or 'uninitialized'
        variables if we aren't sending an argument along with the command.

1.07 (2007-05-07)

     -- continued improved support for child processes as workers,
        making workers go away cleanly if parent pipe dies/EPIPEs.

1.06 (2007-05-04)

     -- better support for Gearman::Workers as child processes talking
        to a Gearman::Server over unix pipe to parent

     -- In worker class, broke out the on connection work, and use that better
        in the case of being a sub process of the gearmand.

1.05 (2007-04-26)

     -- update the prefix code to not break Gearman::Client::Async, which
        was also updated to be compatible.

     -- Add ability for workers to be launched as sub processes of the
        gearmand, using a duped socketpair for communication.

     -- just some extra debugging/deaths during weird cases (helped find
        bugs when server code was massively refactored)

1.04 (2007-04-16)

     -- Let Gearman::Worker use hostnames for job_servers (like
        Gearman::Client does)  (Ask)

     -- Add prefix setting to Gearman::Client and Gearman::Worker for simple
        namespace separation of different instances of the same worker and
        client code sharing job servers. (Ask Bjoern Hansen)

     -- Refactor tests to have less duplicate code (Ask Bjoern Hansen)

     -- Make Gearman::Client / Gearman::Worker tests use $^X instead
        of hardcoded 'perl' to start gearmand (Ask Bjoern Hansen)

1.03
     -- Workers can now specify a timeout that when a task exceeds the time, the
        jobserver will reassign the task and ignore results from the defunct job.

1.02
     -- Task objects weren't reusable before anyway, and now they're really not:
        when they complete or fail, they wipe all their internal subref members
        to break any potential loops.  also now support a new internal hook
        that happens after your on_complete or on_failure callback.  this
        is used by Gearman::Client::Async now, for cancelling timers.

1.01
     -- when workers are writing status messages up to parent, die on SIGPIPE,
        or really any write errors, but just trap SIGPIPE now, if parent goes away.
        fixes a bug found via Gearman::Client::Async's test suite.

1.00
     -- finally package it up and call it 1.00 now that we've been using
        it in (LiveJournal) production for quite a while, finding/fixing
        the bugs that happen when you put something into production.