LAJANDY / Helios-2.80 / Changes

Revision history for Perl distribution Helios.

2.80  Sat Mar 15 21:19:36 2014

	[New Features]
	- New Jobtype System:  Before Helios 2.80, the jobtype of a job 
	  had to be the same name as the service class that ran the job. 
	  If you had a service class you wanted to run multiple jobtypes,
	  name a jobtype something different than the service class, or 
	  divide a workload in a complex way, you had to create multiple 
	  service classes by creating empty subclasses of your service 
	  class. This could lead to a lot of boilerplate code to generate
	  and manage. With the new Helios::JobType class and 
	  helios_jobtype_add command, new jobtypes can be created as 
	  necessary for your jobs, and the new --jobtypes option of 
	  helios.pl allows your service to work with whatever jobtype(s) 
	  you need, regardless of the service's name. This also eliminates
	  the need for the HELIOS_CLASS_MAP table, a virtually 
	  undocumented feature only used by the submitJob.pl CGI. This 
	  separation of services and jobtypes will also allow more advanced
	  job queuing features in the future. 
	- Worker Launch Patterns:  Previous versions of Helios launched 
	  workers to service jobs in a one-at-time manner.  With Helios 
	  2.80, the algorithm used to determine the number of workers to
	  launch is configurable, with 3 algorithms to choose from: 
	  "linear", "dynamic", and "optimistic". "Linear" is the old, 
	  one-at-time algorithm. The "dynamic" pattern attempts to keep 
	  as many workers running as there are available jobs in the queue,
	  adjusting the number of workers launched based on available job 
	  count. The "optimistic" pattern launches as many workers as there
	  are available jobs in the queue, regardless of the number of 
	  workers already running (limited by the MAX_WORKERS limit). The
	  default is "linear" for backward compatibility reasons. Most 
	  Helios services should work well with the "dynamic" pattern, and
	  it has been extensively tested in production environments. The 
	  "optimistic" pattern, while theoretically spooling up workers the
	  fastest, is best used with services with large numbers of jobs of
	  short duration (1-2 sec or less), as shared resources could be 
	  taxed and job contention may result. See the 
	  Helios::Configuration POD for more information.
	- New Commands:  New shell commands greatly reduce the need to 
	  consult the Helios::Panoptes web interface or touch the Helios
	  database directly with SQL commands: 
	  helios_config_get/set/unset 
	  helios_job_status 
	  helios_job_info 
	  helios_service_status 
	  Additionally, the new helios_config_import command allows you to 
	  bulk-import configuration parameters from an INI-style config 
	  file, easing deployment of services with lots of config params. 
	- Helios::TS:  Helios no longer relies directly on TheSchwartz job
	  queuing system, but on its own Helios::TS superset system.  This
	  allows new Helios-specific features to be added (including some 
	  of the changes listed above).
	- Helios::Service->JobLockInterval():  This new method allows your
	  service to specify the amount of time jobs are locked in the job 
	  queue. This was previously set at 1 hour unless you reached into
	  the TheSchwartz layer and redefined the grab_for() setting, which
	  caused problems if you wanted to use MaxRetries() and 
	  RetryInterval() to retry failed problems more frequently than 
	  once an hour.  If you wish to retry failed jobs more frequently 
	  than once an hour, define a JobLockInterval() method in your 
	  service class with the same value as your RetryInterval(). 
	  ALWAYS REMEMBER to make sure these intervals always allow enough
	  time to complete or fail a job; otherwise, two different workers
	  can end up running the same job simultaneously. 
	- Helios::Config:  New get/set/unsetParam() methods retrieve 
	  individual config parameter values for a service, instead of 
	  having to retrieve the entire configuration hash. These are 
	  primarily used in the helios_config_* shell commands, but can be 
	  used by your own programs and scripts to access Helios config 
	  parameters outside of Helios services. 
	- Job Prioritization:  Normally with the Helios::TS job queuing 
	  system, workers pull jobs from the job queue at random to reduce
	  job contention between workers.  Using the PRIORITIZE_JOBS config
	  parameter and Helios::Job->setPriority() method, you can set 
	  priorities for jobs.  Due to the way Helios::TS works, this will
	  not be a strict "one-after-the-other" priority, but still can be
	  used in a general sense to give a job or group of jobs a higher
	  chance of being run before another group of jobs.

	[Other Improvements] 
	- Helios::Config:  Improvements have reduced the number of database
	  queries by 50%.  Additionally, a new get/setService() accessor 
	  method has been added for consistency; the old 
	  get/setServiceName() methods are deprecated but retained for 
	  backward compatibility. 
	- Helios::Job:  The Helios::Job API also receives some cleanup with
	  the addition of 3 new accessor pairs: 
	  get/setArgString() - replaces get/setArgXML() 
	  get/setJobType() - replaces get/setFuncname() 
	  get/setJobtypeid() - replaces get/setFuncid() 
	  These new methods keep Helios::Job consistent with the rest of 
	  the Helios system and also allow for future changes. As with the
	  Helios::Config changes, the old methods are deprecated but 
	  retained for backward compatibility. 
	- helios_job_submit.pl:  Added the --verbose option to echo the 
	  jobid of the job if job submission was successful. Echoing the 
	  jobid by default could cause backward compatibility problems, so
	  it was decided to add an option instead. This behavior will most 
	  likely become the default in Helios 3.0. 
	- Bugfix for [RT79690] - A patch to prevent this bug from causing 
	  lost jobs was incorporated into Helios 2.61, but a full fix 
	  required a change at the job queuing layer (and thus required 
	  Helios::TS).  As mentioned previously in the Helios 2.61 Changes 
	  and the bug's RT thread, if you have experienced this bug you 
	  should also upgrade your DBD:: driver module as well. 
	  
2.61  Fri Sep 13 20:02:14 2013
	- Helios 2.61 is a bugfix release:
	- helios.pl: Fix for [RT81709].
	- helios.pl: Fix for [RT81914].
	- Helios::Service & Helios::Job: Patch for [RT79690].  This bug 
	  actually appears to be caused by LOB binding problems with the
	  underlying DBD module.  This problem can cause a loss of jobs 
	  in the job queue under certain circumstances, so if you are 
	  experiencing this problem you are encouraged to upgrade to this 
	  version and upgrade your DBD module to the latest version.  
	  Consult the rt.cpan.org page for this bug for more information.
	- Makefile.PL: Changes for proper version handling of LICENSE. 
	  Additions so the generated META.* files will have website, 
	  bugtracking, and GitHub information for sites like MetaCPAN and
	  search.cpan.org. 
2.60  Mon Nov 06 00:00:00 2012
	-Significant changes to Helios database connection handling, 
	 including a specialized Data::ObjectDriver class and TheSchwartz 
	 extensions.  Helios now aggressively caches database connections
	 at the DBI layer and uses DBI features to properly handle 
	 database connect/disconnect between daemon and worker processes. 
	 These improvements result in significant performance increases, 
	 especially when using Oracle as a collective database.  One 
	 user's application experienced job throughput increases of ~70%. 
	 MySQL collectives should also experience speed increases and 
	 a reduced number of open connections.
	-NEW Helios::Config modular configuration API.  Similar to 
	 Helios::Job and Helios::Logger, Helios::Config defines the basic 
	 Helios configuration API while also serving as a base class to 
	 build custom configuration modules for Helios applications.  To 
	 define a custom configuration API for a Helios application, 
	 specify a ConfigClass() method in your Helios::Service subclass. 
	 Auxiliary application pieces such as scripts and cronjobs can use 
	 Helios::Config to get the application's configuration without 
	 needing to instantiate the application class and calling the 
	 prep() method as in the past.  Helios::Config makes the Helios 
	 configuration API more robust, modular, and flexible.  See the 
	 Helios::Config POD for more details.
	 NOTE: Helios::Service->getConfigFrom*() methods are now 
	 deprecated in favor of Helios::Config->parseConfig().  The prep()
	 method uses Helios::Config and will continue to be supported.
	-NEW Helios exception class for configuration errors:  
	 Helios::Error::ConfigError.
	-NEW WORKER_MAX_TTL_WAIT_INTERVAL parameter to fine-tune use of 
	 WORKER_MAX_TTL with specific applications.
	-NEW WORKER_BLITZ_FACTOR configuration parameter to launch 
	 more workers sooner.
	-NEW configuration parameter documentation.  All Helios 
	 out-of-box configuration parameters are now defined in a section 
	 of the Helios::Config POD.  For more information about the new 
	 configuration parameters or any others, check the Helios::Config 
	 POD!
	-Helios::Service->dbConnect() has been updated to work with the 
	 new database connection caching scheme and to better support 
	 DBI options.
	-Bundle::Helios::Generic has been changed to Bundle::Helios and 
	 can be used to install Helios prerequisite modules with generic 
	 Perl installs.  Bundle::Helios::RedHat was dropped in 2.41.  See 
	 the INSTALL.* files in the main distribution directory to 
	 determine the right type of install method for your environment.
	-Improved module loading code.
	-Removal of most instances of Error module's try {} catch {} 
	 blocks in favor of Perl generic eval {} or do {}.
	-More tests.
	-Updated INSTALL.* files for generic Perl and CPAN installs and 
	 Red Hat/CentOS 6.x.  BSD, Mac OS X, and other *nix should use the 
	 generic install instructions.  (Debian and Ubuntu based 
	 instructions continue to be the quickest and easiest.  Thank you 
	 Debian for your excellent Perl and CPAN packaging!)
2.41  Sun Jun 03 22:00:00 2012
	-Official Oracle support!  DDL for a Helios Oracle schema is in 
	 'sql/helios_schema_oracle.sql'.  Please note: 
	 1) you will need to adjust the Oracle schema to fit your 
	    environment (especially in regards to security and tablespaces)
	 2) while the Helios core supports Oracle, the Helios::Panoptes 
	    web admin interface still only supports MySQL (Panoptes 
	    support for Oracle is being worked on) 
	-Helios::Service and Helios::Job: added deferredJob() and 
	 deferred() methods to defer processing of a job even though it 
	 was available in the job queue.  These methods (for now) simply 
	 expose the declined() feature added in TheSchwartz 1.10; no Helios
	 job history is recorded for a deferred job.  Existing Helios 
	 users will need to upgrade their TheSchwartz library to 1.10 to 
	 use this feature.
	-Helios::Logger & Helios::Logger::Internal:  moved driver 
	 creation to base Helios::Logger class so (if necessary) any 
	 loggers can get a Data::ObjectDriver to the collective database
	-Helios::Service & Helios::Logger:  refactored getDriver() into 
	 initDriver() and getDriver() to eventually allow caching of 
	 Data::ObjectDriver objects.  Anything needing an ObjectDriver to 
	 the collective database should just continue to call getDriver() 
	 and everything will work fine.
	-helios.pl: calls the service class prep() method on startup 
	 rather than getConfigFromIni() and getConfigFromDb() separately.
	 This ensures proper setup of the loggers and Data::ObjectDriver 
	 db connection.
	-helios.pl: changed startup message so perl doesn't carp a warning 
	 if the service class doesn't have a $VERSION defined.
	-helios.pl: added registration_interval as a configuration 
	 parameter (previously it was hard-coded to 300 seconds in 
	 helios.pl).  The "registration_interval" parameter is the amount 
	 of time (in secs) between service daemon "collective check-ins" 
	 (when a service daemon records its current status in the 
	 HELIOS_WORKER_REGISTRY_TB table).  This table is used to create 
	 the Helios::Panoptes Collective Admin page, and is also useful 
	 for system monitoring and statistics.
	-helios.pl: set new defaults for zero_sleep_interval (10) and 
	 registration_interval (60).  These new defaults should make 
	 Helios more responsive in low load situations and generally work 
	 better for more users out-of-the-box.  (The old defaults geared 
	 Helios more for large volumes of long-running jobs on heavily
	 loaded databases.  The new defaults should work for a better 
	 range of small- to mid-sized loads.)  If these new defaults 
	 create too much traffic for your database, you can set new 
	 larger values in your helios.ini file to decrease database 
	 traffic.
	-helios.pl: changes to prevent a worker from transmuting to a new 
	 service daemon (caused by database instability during transition 
	 between helios.pl, TheSchwartz->work(), and Helios::Service->work() 
	-updated tests
	-submitJob.pl:  fixed example in POD
	-Removed Bundle::Helios::RedHat.  Ultimately, it didn't work any 
	 better (or worse) than using 'install Helios' command with the 
	 CPAN shell.  Bundle::Helios::Generic seems to work well on other 
	 platforms, however.  If you wish to use Helios with RHEL/CentOS 
	 5.x or earlier, it is recommended you install a more recent 
	 version of Perl than 5.8.8 for ease of installation reasons.  
2.40  Sun Jan 22 22:30:00 2012
	-NEW Extensible logging subsystem (Helios::Logger) 
	-NEW Helios::Error::BaseError exception class; acts as parent for
	 entire Helios::Error:: hierarchy
	-NEW Helios::Service->metarun() method to modularize metajob 
	 handling and provide for extensible metajob support 
	-NEW service run() and metarun() methods now run in an eval 
	 block as a process safety measure to catch uncaught exceptions.
	-NEW MP3IndexerService example application including in eg/
	 directory (StubService also rewritten for clarity)
	-REVISED Helios::Service->work() method no longer pays attention 
	 to return value of run() method (this feature was under-
	 documented, misunderstood, and caused problems with eval {} and 
	 Try::Tiny exception handling
	-REWRITTEN submitJob.pl CGI script; supports sending XML stream 
	 in addition to form encoding, no longer dependent on 
	 HELIOS_CLASS_MAP
	-REWRITTEN Helios::TestService; more modern Perl and easier to 
	 understand
	-REWRITTEN install instructions, including specific instructions 
	 for Red Hat- and Ubuntu-based systems, and including CPAN install 
     bundles (still experimental)
    -REWRITTEN Helios::Tutorial 
    -Helios::Service:  fixed MaxRetries() and RetryInterval() retry 
     job methods
	-Helios::Job:  changed failed() and failedNoRetry() to limit the 
	 char length of the error recorded to 254 chars.  That's one less 
	 than the default size of the db field in the ERROR table, so it 
	 should prevent long errors like stack traces from unexpectedly 
	 blowing up worker processes and potentially losing jobs.
	-Revised dependencies:
	 DBI 1.52 (shipped with RHEL/CentOS 5)
	 XML::Simple 2.14 (shipped with RHEL/CentOS 5)
	 Removed XML::SAX dependency (unnecessary)
	 Removed XML::Namespace dependency (unnecessary)
	 Removed XML::Parser dependency (unnecessary)
	-Removed deprecated classes, methods, and other files:
	 Helios::Service->parseArgXML() (deprecated since 2.00)
	 Helios::Worker (deprecated since 2.00)
	 sql/helios_schema_20upgrade.sql	 
	-Updated copyright notices
	-Switch to "Linux-style" versioning: 
	 odd minor version == development version
	 even minor version == stable version
	 second minor number == patch level
	 (we're still also using Perl-style _ for devel versions too)
2.23  Fri Sep 10 12:22:00 2010
	-Helios::Job: fixed bug where getArgXML() returned undef if job
	 object was vivified from job queue
2.22  Thu May 21 11:30:00 2009
	-Helios::Service: fixed a bug in getFuncidFromDb() that caused 
	 services to fail to load properly if there was no entry for them 
	 in FUNCMAP table.
	-helios.pl: Several revisions to make it a more robust daemon. Now
	 honors TERM signal; 'kill <pid>' from a shell will attempt to 
	 safely shut down workers and daemon process w/o using Panoptes
	-Documentation updates
2.21  Thu May 14 17:00:00 2009
	-added set/getJob() accessors to facilitate access to current 
	 Helios::Job object from methods other than run()
2.20  Tue Apr 28 18:00:00 2009
    -various tweaks for cross-platform compatibility
    -documentation additions/corrections
    -added Helios::LogEntry::Levels to remove dependencies on 
     Sys::Syslog
    -helios.pl: Cleaned up messages to console after daemonize
	-helios.pl + Helios::Service: added support for WORKER_MAX_TTL 
	 config param
    -Helios::Service: added JobClass() allow extensible Helios::Job 
     classes
    -Helios::Service: added MaxRetries() and RetryInterval() to replace
     TheSchwartz's max_retries() and retry_interval() (this is not an 
     abitrary change; it's to allow for extensions at the Helios layer 
     in the future)
    -Helios::Service: added default run() method for class completeness
     (it throws a Helios::Error::FatalNoRetry error because you forgot 
     to implement a run() method)
    -Helios::Service: removed logging to syslogd to eliminate 
     Sys::Syslog dependency (see HeliosX::ExtLoggerService)
    -Helios::Service: added set/getFuncid(), getFuncidFromDb() and 
     changed jobWaiting() to use them.  This eliminates an unnecessary 
     database query previously done every time jobsWaiting() was called
    -Helios::Service: added support for log_priority_threshold config 
     param
    -Helios::Job: added getFailures() to access the number of times a 
     job has previously failed
2.00  Fri Aug 01 02:00:00 2008
	-First public production release (same as 1.90_26 w/more documentation)
1.90_26 Thu Jun 26 11:30:00 2008
	-Helios 2.0 RC1
	-NEW Helios 2.0 API
	 Helios::Service
	 Helios::Job
	 Helios::MetajobBurstService
	 Helios::ConfigParam
	 Helios::JobHistory
	 Helios::LogEntry
	 Helios::TestService
	-helios.pl: added PID check to prevent 2 instances of a service 
	 running on the same server
	-helios.pl: added --clear-halt option to clear Worker Admin HALTs 
	 at the cmd line
	-helios.sql: new helios_worker_registry_tb.start_time field for 
	 uptime stats
	-submitJob.pl, helios_job_submit.pl, helios_job_clean.pl use new 
	 2.0 API
	-eg/Stub/StubService.pm for example of new 2.0 API service class
	-new tests to test both 1.x and 2.x API	
1.19_07 Thu May 29 14:00:00 2008
	-added Helios.pm (for version checking & documentation)
	-added Helios::Job (for eventual refactoring)
	-added Helios::Test (for system testing)
1.19_06 Wed May 28 12:00:00 2008
	-added LICENSE file that was left out
	-add helios_log_clean.pl utility to clean log, job history
	-added eg/Stub/StubWorker.pm as a worker class stub example
1.19  Mon May 19 12:00:00 2008
	-First public release



Hosting generously
sponsored by Bytemark