2019.01 Release

Licensing robustness and network fault-tolerance have been significantly improved. The software can operate in "disconnect mode", which means that it connects, gets a checkout, then disconnects from the license server. Checkout is refreshed every hour. The software will not fail for until 5 full days have passed since the last successful checkout. Alerts are issued for this as well.

As before, license servers may be specified as a colon-separated list via the RMonitor_LICENSE environment variable

substituteArrayIdInResources 
substituteArrayIndexInResources 
substituteJobIdInResources 

The -signals, -include and -exclude options to nc stop and vovstop -f are now functional.

The trace parameters defaultStopSignalCascade and defaultSuspendSignalCascade have been added to specify default behavior when requesting to stop or suspend a job, respectively.

The job property STOP_SIGNALS can now be used to set stop-signal cascades on a per-job basis.

Internally, all commands that change node status now pass along a reason for the change, which is logged in the journal if node status logging is enabled. The most recent reason is in the @WHYSTATUS@ field.

In vtk API commands that change status, a new "-why REASON" option has been added. Any reason can be passed, and it will get added to the @WHYSTATUS@ field, which sometimes gets stored as the @WHYSTATUS@ property on nodes (if the text reason is not in the internal list of known reasons). Client information is also included (user, executable name, security level, etc).

Node status logging can be enabled server-wide using a global setting (see vtk_node_status_logging command), or individually on jobs or files. It can also be enabled on sets, which apply only to the nodes in the set at the time the command is called (it applies it to the nodes in the set, not to the set itself).

Please provide feedback on this feature during the beta period.

OWNER:     ATTACH DETACH EDIT VIEW FORGET DELEGATE EXISTS     
ADMIN:     ATTACH DETACH EDIT VIEW FORGET DELEGATE EXISTS     
EVERYBODY: ATTACH DETACH VIEW EXISTS     

Additionally the AND, OR and NOT keywords are now case-insensitive.

This protection can be enabled and configured via the following server configuration parameters defined in the policy.tcl file:

hog.protection.enable (default=0, which is off)
hog.protection.jobcountthreshold (default=100000, min=1000, max=999999)
hog.protection.clientdelay (default=1, min=1, max=600, specified in seconds)

In addition, the fields MAXNUMACORES and MAXFREENUMACORES have been added to the SAllocatorVES object for the purposes of vtk_select_loop, vovselect and related commands.

This is done by setting the LM(licmon) setting in the vovresourced configuration file to the name of the Monitor instance, and making sure that the LM(flexlmd) setting that normally contains the Monitor host and port information is not set. If LM(flexlmd) is set, the location specified therein will be honored as before.

Additionally, a new configuration parameter, LM(enabled), has been added to provide the ability to disable communications between vovresourced and Monitor altogether. The data type of this parameter is boolean and the default value is true.

• diskspacecheck.minfreemb (default=1000, min=1, max=50000)
• diskspacecheck.minfreepercent (default=1, min=1, max=50)

Increased the vovslave almost-full free disk space threshold from 5MB to 100MB. This can be overridden via the -mindisk option to vtk_slave_set_defaults and/or vtk_slave_define. Added a new notification health check (doTestHealthServerDiskSpace) so that email notifications will be sent in addition to the alerts and log messages that are generated.

For this reason, when using this new default behavior, equivalences that reference an environment variable should not resolve the variable in this file, in environments that will have both Unix and Windows clients. Instead, they will need to be resolved by the client upon receipt. This is done by enclosing the equivalence value inside curly braces and referring the environment variable as $VARNAME as opposed to the Tcl format of $env(VARNAME).

The previous behavior can be restored by setting VOVEQUIV_CACHE_FILE to "legacy" in the $SWD/setup.tcl file for a given instance or globally for all instances via a $VOVDIR/local/vovsetup script (vovsetup.csh or vovsetup.sh).

Older clients connecting to a new vovserver will work without any changes as long as the $SWD/equiv.caches directory exists and has world-write permissions. Variables in the new form will resolve correctly.

New clients connecting to an older vovserver will require legacy mode to be enabled.

Also, for all properties that exceed the max value length, instead of setting the value to an error message only, the value is now set to the same error message as before, followed by a truncated version of the value intended, ending with a truncation indicator.

% vovselect resources.X from slaves 
% vovselect resources.X from 012345678 where 'resources.X>100' 

The key to this statement is escaping each single quote with another single quote in the where clause (joe's-host becomes joe''s-host).