ALFRED RELEASE NOTES - Version 4.1
The 4.1 Release The 4.0 Architecture Version 4.1 Release Notes	Earlier Releases Version 3.2 and older

The Basic Alfred 4.0 Architecture

Alfred 4.1 - Release Notes

New Features

 

• Maitre-d Rank Determines Metrics Authority when there are several maitre-ds running on the network, anyone of them can respond to requests from alfservers for the current metrics definitions. The new maitre-d ranking scheme allows administrators to strictly determine which maitre-d is the definition authority. By default, all maitre-ds have a default rank of zero, and when ranks are equal the maitre-d with the newest schedule file is considered to be the most up-to-date. Here's an example which sets a maitre-d's rank, its definitions will prevail over those with lower numeric rank:

  alfred -maitred -rank 5

   

• Automatic Deletion of Old Error-Wait Jobs - there is a new preference in the alfred.ini file which controls the time, in hours, before jobs stalled in "Error-Wait" state are automatically deleted. The preference is called "userPrefs(timerErrWaitExpire)". The default value is 0, which disables automatic deletion. Error-wait state occurs when a job is unable to proceed because all remaining tasks are blocked waiting for dependent tasks which have errors.

   

• Message Glossary - this is a new item which has been added to the Help menu on the main Alfred window. The glossary window lists and explains most of the brief status messages which are displayed in the main job window.

   

• Cmd [...] -retryrc {nnn nnn nnn} The new "-retryrc" option allows a command which exits with a non-zero exit status to be automatically requeued for dispatching if the exit status matches one of the numbers in the 'retryrc' list. This is useful for commands that internally detect network or resource errors and wish to be retried at a later time or with a different remote server.

   

• Server Uptime pseudo-Metric is now generated automatically by alfserver. This is simply the number of minutes since the server was started. It provides a simple way to see which servers are very old, or to find those that might have been restarted unexpectedly.

   

• Tasks Requeued on Server Shutdown There is a new task requeue feature which can be enabled with a setting in the alfred.ini file. The "allowServerShutdownEjects" setting controls whether or not tasks currently running on a remote server should be requeued onto other servers when the alfserver process is shutdown in a controlled way (using SIGTERM, nrmcontrol, etc).

   

• RemoteCmd Fallback to rsh. At some sites the integrated Alfred remote execution server, alfserver, is not available. In this situation the Alfred scripting command "RemoteCmd" will be unable to launch commands on remote servers. There is a new setting in the alfred.ini file which allows RemoteCmd to fallback to launching rsh instead. This limits the capabilities of RemoteCmd in some ways, and requires that each server have a copy of "rshd" installed as well. The rsh daemon is typically available by default on Unix systems, whereas it is usually an add-on for other systems.

   

• Simpler Shared-Server Syntax when several Cmds are required to reuse the same remote server (which is checked out once). The %h substitution will now automatically resolve to the host part of the referenced id, as shown in the second command:

  	Task {test} -service pixarNRM -id bar -cmds {
  		Cmd {echo "1 %bar(host)"} -refersto bar
  		Cmd {echo "2 %h"} -refersto bar
  	}
  	

   

• Fallback Maitre-d Grace Period has been added as an additional way to guard against inadvertent use of the fallback maitre-d(s) when the primary is temporarily unavailable (due to network load, etc).

   

• Save Output to File has been added to the per-task pop-up menu in the TaskTree (DAG) window. This option prompts for a filename and then copies the current output logs for the given task to the specified file.

   

• Server "Kill File" similar to a NIMBY control, but using the file system instead of a UI. See the alfserver discussion.

Bug Fixes

• Spooled Job Scripts Containing Errors would sometimes create a "ghost" entry in the user interface. Jobs are automatically deleted when parsing errors are detected, although sometimes a non-executing entry would still appear in top-level user interface. Partial jobs are now fully deleted and better error messages appear in the command log.

   

• Voluminous Command Output was sometimes truncated in the command logs at approximately 40k bytes. Large output logs should appear in full now.

   

• Error Opening Task-Tree Window - sometimes an error message would appear rather than the job "DAG" window when the triangular open button was pressed. This was caused by a bug in the auto-placement mechanism which is used when there are no previously saved window locations, and it has been fixed.

   

• Command Tokenizer was sometimes incorrectly removing double-quotes from arguments. Alfred scripts contain Cmd directives which specify an executable and arguments to be launched. In some instances, double-quotes embedded within argument strings were being stripped before being passed to the executable. The 4.1 release now tokenizes arguments in the same manner as the 3.x releases.

• Bug Fix:   Disabling Walk-Ahead Limits was not possible in the 4.0 release. The limit on the number of ready but currently unlaunchable tasks generated by expand and Iterate nodes can be set using the Walk-Ahead entry in the Session->Preferences dialog. As with the 3.3 release, setting the value to "-1" will once again allow unlimited unrolling when these nodes are first processed.

• Empty Windows Back-up Schedules would sometimes be created when a new schedule was written. This has been fixed.

• Limit Counts Exceeded have been fixed in the case where the command name was also specified as explicit tag. Limits were also sometimes exceeded in the 4.1-beta releases due to race conditions which have been resolved.