This is an outdated version of the HTCondor Manual. You can find current documentation at http://htcondor.org/manual.

10.2 Upgrading from the 8.0 series to the 8.2 series of HTCondor

Upgrading from the 8.0 series of HTCondor to the 8.2 series will bring new features introduced in the 8.1 series of HTCondor. These new features include: configuration is more powerful with new syntax and features, and the default configuration policy does not preempt jobs, monitoring is enhanced and now integrates with Ganglia, automated detection and management of GPUs, numerous scalability enhancements improve performance, an improved Python API including support for Python 3, new native packaging and ports are available for the latest Linux distributions including Red Hat 7 Beta and Debian 7, cloud computing improvements including support for EC2 spot instances, OpenStack, and condor_ssh_to_job directly into EC2 jobs, grid universe jobs can now target Google Compute Engine and BOINC servers, partitionable slots now are compatible with condor_startd RANK expressions, and consumption policies permit partitionable slots to be split into dynamic slots at negotiation time, improved data management including dynamic adjustment of the level of file transfer concurrency based on disk load (see section 3.3.10), and experimental support to allow the execution of a job to be overlaid with the transfer of output files from the previous different job, and the new condor_sos tool helps administrators manage overloaded daemons.

Upgrading from the 8.0 series of HTCondor to the 8.2 series will also introduce changes that administrators of sites running from an older HTCondor version should be aware of when planning an upgrade. Here is a list of items that administrators should be aware of.

• New configuration syntax and features change:
  • The interaction of comments and the line continuation character has changed. See section 3.3.1 for the current interaction.
  • The use of a colon character (:) instead of the equals sign (=) in assigning a value to a configuration variable causes tools that parse configuration to output a warning. Therefore, any custom parsing of tool output may need to be updated to handle this warning. Previous versions of the default configuration set variable RUNBENCHMARKS using a colon character; HTCondor code explicitly suppresses the warning in this case.

• The default user priority factor for new users has changed from 1 to 1000. Therefore, unless the accountant log is discarded, existing users will still have a priority factor of 1, while new users will have a priority factor of 1000. Use condor_userprio to change the priority factor of existing users if the accountant log is maintained across the upgrade. (Ticket #4282).

• For Windows platforms, HTCondor has switched to use the newer 2012 Microsoft compiler, which uses the Visual C++ 2012 Runtime components. Therefore, the HTCondor MSI installer will acquire this Runtime, if it is not already installed.

• The meaning of cpus=auto when there are more slots than CPUs has changed within the configuration. In the SLOT_TYPE_<N> configuration variable, cpus=auto previously resulted in 1 CPU per slot. Now, all slots with cpu=auto get an equal share of the CPUs, rounded down. (Ticket #3249).

• The DAGMan node status file formatting has changed. The format of the DAG node status file is now New ClassAds, and the amount of information in the file has increased.

• Setting configuration variable DAGMAN_ALWAYS_USE_NODE_LOG to False or using the corresponding -dont_use_default_node_log option to condor_submit_dag is no longer recommended. Note that at strictness setting 1 (the default), setting DAGMAN_ALWAYS_USE_NODE_LOG to False will cause a fatal error. If the DAG must be run with DAGMAN_ALWAYS_USE_NODE_LOG set to False, a good way to deal with upgrading is to use DAGMan Halt files to cause all of the running DAGs to drain from the queue, and then do the upgrade after the DAGs have stopped. After the upgrade is done, edit the per-DAG configuration files to have DAGMAN_ALWAYS_USE_NODE_LOG set to True, or set DAGMAN_USE_STRICT to 0 and re-submit the DAGs, which will then run the Rescue DAGs.

• If using ENABLE_IPV6 = True, the configuration must also set ENABLE_IPV4 = False. If both are enabled simultaneously, daemons will listen on both IPv4 and IPv6, but will only advertise one of the two addresses.

• Globus 5.2.2 or a more recent version is now required for grid universe jobs of grid-type nordugrid and cream. Globus version 5.2.5 is included in this 8.2.0 release of HTCondor. HTCondor will prefer to use libraries already installed in /usr/lib[64], when present. (Ticket #4243).

• If referencing attribute SubmittorUserPrio in configuration, such as in the PREEMPTION_REQUIREMENTS expression, you will need to change it to SubmitterUserPrio Note the spelling difference in the ClassAd attribute name. (Ticket #4369).

• HTCondor can not distinguish normal from abnormal job exit for Nordugrid ARC grids. Therefore, all grid-type nordugrid jobs will be recorded as terminating normally, with an exit code from 0 to 255. (Ticket #4342).

• For configuration, parameter substitution now honors per-daemon overrides. This long standing bug's fix may result in subtle changes to the way that your configuration files are processed.