1.13.001 2014-12-14
* Feature Request: #115368 for Gearman-Server: Allow bind to specific IP address
* Fix bug #115458 Distributions contain no tests.
Actually only use ok and version tests
* Fix bug #89033 typo fixes
* Fix bug #107045 [PATCH] fix pod whatis
* Fix bug #115350 Uses old port by default (7003) should use 4730
* pod usage
1.12 2014-12-14
* Add HACKING file
* Fix bug #70728 memory leak when clients disconnect (sleeper list isn't pruned).
Fixes CPAN RT 70728 (Marsh Yamazaki)
1.11 2010-01-17
* Don't get stuck in the wakeup loop. Dummy.
1.10 2009-10-04
* Read client input more aggressively. Speed improvement.
* Add text commands 'jobs' and 'clients' which give information allowing you to trace
jobs from clients to workers while they are running.
* Flush buffers to client on EOF (assume half-closed). This makes things like netcat
work better as a way to speak text protocol to the gearmand.
* Add command-line option to adjust a delay before more workers are woken up.
This acts as an anti-starvation mechanism in case of lower wake up counts.
-Default is .1 seconds, formerly this option was not needed because all workers
were woken up at the time of job submission.
* Add command-line option to adjust number of workers to wake up per job injected.
-Default is 3, formerly was -1 (wake up all as fast as possible)
* Add command-line option to change the number of sockets accepted at once per
listener socket.
-Default is now 10, formerly used to be 1.
* Add exceptions passing support to gearman server classes, using new options
support.
* Add options support to server clients, so they can subscribe to newer protocol
features.
* Change listening socket to be a real Danga::Socket subclass, this allows
pausing for a period of time when we run into accept errors. This will
fix the problem of gearmand spinning 100% cpu in those cases.
* Make gearmand a little more vocal about socket accept errors.
* add fast read concept to server reading from client codepath. This drastically
improves performance of jobs over 1k in size.
* fix in-process client and start_worker method calls to use non-blocking
sockets like they should.
1.09 2007-05-09
* make start_worker (for making worker child processes), return
the pid in scalar context, or ($pid, $conn) in list context
* make the text command "status" show functions which have
registered workers, even if nothing is pending for that func.
1.08 2007-05-01
* let start_worker take a coderef for the exec, not just an $exe name
1.07 2007-05-01
* fix up bug in Gearman::Server: we set OtherFds, instead of adding with
Danga::Socket->AddOtherFds, blowing away other watched fds.
* Add Gearman::Server->to_inprocess_server that returns a socket connecting to
the server object, we can then use this socket with a Gearman::Client::Async
object to run clients and servers in the same thread.
* Add ability for workers to be launched as sub processes of the
gearmand, using a duped socketpair for communication.
1.06 2007-04-25
* split up the monolithic gearmand script (which had classes, but
all internally), into separate files per class, and also make
a new Gearman::Server class, so Gearman servers can be embedded
into other processes. (running Gearman router inside Perlbal,
MogileFS, DJabberd, etc... then having child processes be workers)
1.05
* add --pidfile=/path/file.pid option and document the command
line options (Ask Bjoern Hansen)
1.04 2007-02-20
* Add maximum jobqueue size feature, with a test.
1.03 2006-09-25
* "version" text command
* weakref client connections in listeners list so they go away.
still wondering why the jobs holding them lasted so long.
or was it a loop? time will tell. go gladiator.
1.02 2006-09-20
* Devel::Gladiator support for tracking elusive memory leak
* support for "shutdown" and "shutdown graceful" commands,
the latter of which immediate stops listens, and exits
when convenient
1.01 2006-07-01
* Artur Bergman backed out some changes seconds before I did
the 1.00 release. this puts them back in.
1.00 2006-07-01
* the get-it-on-CPAN-finally release. but forgot this changes
entry.
0.30 2006-06-27
* fix memory leak with sleepers list. change it from list to hash
so there can't be dups, as there were before, when a worker
could do multiple functions.