- A new limit called "default" can be used to dynamically
initialize the limits for tags and app names which aren't specifically
listed in the schedule limits list.
- The Cmd and RemoteCmd scripting commands now support the "-samehost 1"
option which in conjunction with "-atleast n" specifies that
multiple acquired server slots must all be on the same physical host.
This can be useful for multi-threaded or parallel-processing applications.
There is a new RenderGlobals setting in MTOR to emit this flag as well.
- The "samehost" support also required a change to the default maitre-d
assigner. Sites that have created a custom variant assigner DSO
should merge in the changes from the new version of the provided source code.
- Several sources of latency associated with job traversal and slot request
queuing have been removed. This should be reflected in quicker restarts
of error tasks, and also in maintaining a the maitre-d wait queue.
- alfserver.ini has been updated, including a new
parameter which controls the time delay used by alfserver between
escalating its shutdown signals to processes that aren't responding.
- Jobs from users whose userids contain dots are now spooled correctly.
- When restarting a Windows PC, alfred services (alfserver, maitre-d)
now respond more quickly to the system shutdown message and check-in
their licenses before the process is killed. Previously these licenses
remained unavailable for new invocations until the license manager
reclaimed them, usually after a few hours.
- On Windows systems, "foreground" (-debug) invocations of alfred
programs now respond more quickly to ctrl-c interrupt events.
- On Windows systems, the alfserver logs provide the correct process
id when listing process exit status.
- Alfserver now applies signals more consistently to entire process groups,
for all signal types, so that if the program launched by alfserver spawns
additional subprocesses, they will receive termination and pause signals
as well (unix/linux only). Note that if alfred/alfserver launch a program
which is actually a /bin/csh script, and that script creates subprocesses
via the '&' operator, then those subprocesses are not in the same process
group as the original script and therefore will not receive these signals.
This is a feature of csh, use a different scripting language such as bash
if you want signal propagation.