Next: 8.7 Stable Release Series Up: 8. Version History and Previous: 8.5 Stable Release Series Contents Index

Subsections

8.6 Development Release Series 6.5

This is the development release series of Condor, The details of each version are described below.

Version 6.5.5

New Features:

Condor-G jobs can now use matchmaking.
condor_ advertise has a new option, -tcp.
Added initial support for the condor_ gridshell. Documentation is not yet available, but will be coming soon.
Added the TRUST_UID_DOMAIN config file setting. For more information about this feature, see section 3.3.7 on page .
Added the REQUIRE_LOCAL_CONFIG_FILE config file setting. If this setting is False, the absence of the file specified in the LOCAL_CONFIG_FILE config file setting is not treated as an error.
Added a new command, condor_ wait, which will watch a log file until a job or set of jobs complete.

Bugs Fixed:

The condor_ starter was reporting ImageSize to be much bigger than reality for multi-threaded jobs in Linux. If the jobs were ever evicted, this could cause them to never match to another machine or to be unnecessarily restricted.
The condor_ starter no longer seg-faults when attempting to run a job as nobody on HPUX. This bug was introduced in version 6.3.2 and effected all job universes.
Fixed a bug related to defining CONDOR_IDS in the configuration file. This new feature was added in version 6.5.3, but it did not work properly with the Standard and PVM universes. Now, all parts of Condor should work correctly without permission problems if CONDOR_IDS is defined in the config file.
Under certain rare circumstances (e.g., running condor_ rm on a condor_ dagman job which was running a POST script for a node which had no PRE script defined), DAGMan could dereference a nonexistent pointer and segfault. This has now been fixed.
Fixed a minor bug in the -jobad option for the condor_ cod command-line tool when activiating Computing On Demand (COD) jobs. Now, the parser for the user-specified ClassAd will ignore comments and whitespace.
In Standard Universe, a transient failure to read the user's password file entry (e.g. with an overloaded NIS server) could result in the job running to completion and exiting the queue without updating the user log.
Fixed a bug which prevented the user from specifying Hawkeye jobs with colons and / or spaces in the executable name. The HAWKEYE_JOBS macro now allows for the individual fields to be quoted, solving this problem.

Known Bugs:

There is a suspected-but-not-yet-confirmed bug in DAGMan's -MaxJobs feature that can lead to a segfault which re-occurs during recovery. Users are advised to avoid using this feature until the bug is found and fixed. condor_ dagman jobs submitted without the -MaxJobs feature are not affected. Likewise, the -MaxPre and -MaxPost features are not affected and can be safely used.
For large numbers of jobs, condor_ analyze may run out of memory and fail.

Version 6.5.4

New Features:

Added the -jobad option to the condor_ cod command-line tool when activiating Computing On Demand (COD) jobs. This allows the user to specify dynamic attributes at the point when they request the job to start, instead of having to define everything ahead of time in the Condor configuration files at the execution site.

Bugs Fixed:

The condor_ startd will now immediately exit with a fatal error if there is a syntax error in any of the policy expressions (START, SUSPEND, PREEMPT, etc) as defined in the configuration file. Before, the condor_ startd might run for a long time before reporting the error, or would silently ignore the expression. For example, if the START expression contained a syntax error, the condor_ startd would never match against any resources without warning the administrator that the START expression could not be parsed.
The condor_ starter now checks to make sure that the value of the JOB_RENICE_INCREMENT expression is within the valid range of 0 to 19, and adjusts out-of-range values to the closest valid value, printing a warning to the StarterLog. Previously, no warnings were printed. Also, under Windows, values above 19 used to be incorrectly adjusted to 0 (i.e., normal priority) instead of 19 (i.e., idle priority class).
When Condor is testing to see if a given submit machine is in the same UID_DOMAIN as the execution machine, it now uses a case-insensitive test. Previous versions of Condor that used a case-sensitive comparison caused problems for sites that have mixed-case hostnames in their DNS records.
When the condor_ shadow is trying to claim a condor_ startd, if the IP address of the condor_ shadow cannot be resolved, the condor_ startd used to refuse the request. Now, the condor_ startd accepts the request and uses the IP address of the condor_ shadow in all log messages, tool output, and the ClientMachine ClassAd attribute that are normally set to the hostname.
Process handles would intermittantly be lost on Windows, causing the condor_ startd and other daemons to EXCEPT(). This has been fixed.

Known Bugs:

None.

Version 6.5.3

Table 8.3: Condor 6.5.3 supported platforms

Architecture	Operating System
Hewlett Packard PA-RISC (both PA7000 and PA8000 series)	HPUX 10.20
Sun SPARC Sun4m,Sun4c, Sun UltraSPARC	Solaris 2.6, 2.7, 8, 9
Silicon Graphics MIPS (R5000, R8000, R10000)	IRIX 6.5
Intel x86	Red Hat Linux 6.2, 7.2
	Red Hat Linux 8 (clipped)
	Red Hat Linux 9 (clipped)
	Windows NT 4.0 (clipped)
	Windows NT 2000 (clipped)
	Windows NT XP (clipped)
ALPHA	Digital Unix 4.0
	Red Hat Linux 7.2 (clipped)
	Tru64 5.1 (clipped)
PowerPC	Macintosh OS X (clipped)
Itanium	Red Hat Linux 7.2 (clipped)

New Features:

Added support for $RANDOM_CHOICE(xxx,yyy,...) in the configuration file parsing and in the submit description file, so we can now do things like:
```
   CKPT_SERVER_HOST = $RANDOM_CHOICE(check1.example.com,check2.example.com)
```
Added a new installation and configuration tool to Condor. Instead of using the complicated condor_ install script to install and configure Condor, users can now try the simplified condor_ configure script. The new method does not ask any interactive questions. Instead, you set the few installation options you need to specify for your site as command-line arguments. For most sites, this method is much better than the question-driven condor_ install script, and we recommend that new users of Condor try condor_ configure. All the options it supports are documented in the new condor_ configure man page, section 9 on page . However, since this is the first public release of the new script, there may be some problems with how it works under certain conditions, so we are still including condor_ install in the 6.5.3 release if you run into problems with condor_ configure and would prefer to use the old method.
You can now define CONDOR_IDS in the Condor configuration files, not just as an environment variable. This setting is used to specify what effective user id the Condor daemons should run as when they are spawned with root privileges. This is the effective user id that Condor daemons write to their log files as, manipulate the job queue, and so on. Therefore, the log and spool directories should be owned by this user. If the Condor daemons are spawned as root and CONDOR_IDS is not set, Condor searches for a ``condor'' user in the local user information database (the /etc/passwd file, NIS, etc). For more information about the CONDOR_IDS setting, see section 3.6.10 on page and/or section 3.3.3 on page .
Added the D_ HOSTNAME debugging level to print out verbose messages regarding how Condor resolves hostnames, domain names, IP addresses, and so on. If set, the Condor daemons also print out messages showing how they initialize their own local hostname and IP address. This is useful for sites that are having trouble getting Condor to work because of problems with their DNS or NIS installation.
Improvements for Computing On Demand (COD) support:
- Simplified the interface to the condor_ cod command-line tool for managing COD claims. All commands that require a ClaimID argument no longer require -name or -addr, since the ClaimID contains the contact information for the condor_ startd that created it.
- If you request a claim from a specific virtual machine with the -name option to condor_ cod_request (for example condor_cod_request -name vm3@hostname) the tool will now automatically append a requirement to your request to ensure that your claim comes from the virtual machine you requested.
- condor_status -cod -long now provides the ClaimID string for any COD claims in your pool. This is useful in case you misplace or do not save the ClaimID returned to you from condor_ cod_request.
Improvements for PVM support:
- Added support for pvm_export.
- Added support for any tasks the Master spawns to also spawn other tasks. Before this fix, only the master process could spawn tasks.
- Added the ability for multiple tasks to be run on a resource simultaneously.
- Added support for Condor-PVM to run multiple tasks under a single Condor-PVM starter. Currently, this feature must be turned on by a specific option in the submit description file.
Improvements for Windows:
- Added support for encrypting the execute directory on Windows 2000/XP. This feature is enabled by setting ENCRYPT_EXECUTE_DIRECTORY = True.
- Added the VMx_USER config file parameter for specifying an account to run jobs as (instead of the condor nobody accounts). A different account can be specified for each VM.
Added new option -dagman to condor_ submit_dag which allows specification of an alternate condor_ dagman executable.
Added new entry in a DAG input file that adds a macro (using the "+" syntax) for use in the submit description file. The new entry is called VARS.
Improved -config option to condor_ config_val to print descriptive lines to stderr and filenames to stdout, so that its output can be cleanly piped into tools like xargs.
On Unix, added the VMx_USER config file parameter for specifying an account to run jobs as instead of ``nobody''. (i.e. when UID_DOMAIN s do not match and SOFT_UID_DOMAIN = false). A different account can be specified for each VM.
Added EXECUTE_LOGIN_IS_DEDICATED . When set to True, this tells the condor_ starter to track job processes by login, instead of by process tree, which prevents so-called lurker processes. This is turned off by default.

Bugs Fixed:

Fixed the serious bug with periodic checkpointing that was introduced in version 6.5.2.
Fixed a security hole in the condor_ schedd when it was running not as root on UNIX. Starting with Condor version 6.5.1, if you did not run the condor_ schedd as root, it would incorrectly allow other users to modify the job queue. In particular, a user could remove another user's jobs, to submit jobs to another user's personal condor_ schedd, and so on. This bug has now been fixed, and the condor_ schedd will no longer allow users to remove or modify other user's jobs.
Fixed some bugs in the condor_ kbdd that prevented it from communicating with the condor_ startd to send updates about console activity (key presses and mouse movements). The condor_ kbdd is only needed on Digital Unix and IRIX platforms, so this bug did not effect most Condor users.
Fixed a minor bug in the condor_ startd that caused it to exit with a fatal error when it fails to execute the condor_ starter. This problem is now handled gracefully and is not considered a fatal error.
Fixed a minor problem introduced in version 6.5.2 in condor_ submit. New attributes for specifying Condor's File Transfer mechanism were added in version 6.5.2. condor_ submit was supposed to be backwards compatible with old job description files, but in a few cases, it would incorrectly give a fatal error when it saw certain combinations of the old syntax. Now, all old job description files are properly recognized by the new condor_ submit. For more information on using the Condor File Transfer mechanism, see section 2.5.4 on page , or the condor_ submit man page on page .
Fixed a minor problem (introduced in version 6.5.2 while fixing another bug) that made it difficult to use a condor_ master to spawn the Hawkeye daemons. Now, if you install Hawkeye on a machine, add HAWKEYE to your DAEMON_LIST, and define HAWKEYE to be the path to the hawkeye_master program, the condor_ master will spawn everything correctly.
PVM-related bugs:
- Fixed the Environment attribute in the submit description to be honored by both the master process and the slave tasks.
- Fixed a few PVM deadlock scenarios during PVM startup.
Fixed a bug which caused the ``KFlops'' published in the startd's ClassAd to have a large variance on fast CPUs.
Fixed a bug which could cause inter-daemon communication problems. A side effect of this fix, however, causes all daemons to use 3 more file descriptors.
Fixed a misleading error message generated by condor_ q when it can not find the address of the condor_ schedd to query. Previously, it suggested a fairly obscure problem as the likely source of the error. The same error occurs when the condor_ schedd is not running, which is a far more common situation.
Fixed a couple of related of bugs which could cause the condor_ collector to seg-fault under unusual circumstances.
Fixed a minor bug where Condor could get confused if a machine had only a fully qualified domain name in DNS, but no simple hostname without the domain name. In this case, Condor daemons and tools used to exit with a fatal error. Now, they function properly, since there's no reason for them to consider this a problem.
Fixed a minor bug on Windows where the condor_ starter would incorrectly determine the VM number if StarterLog path contained periods.

Known Bugs:

The VARS entry within a DAG input file is not propagated to the rescue DAG.

Version 6.5.2

Known Bugs:

There is a serious bug with periodic checkpointing that was introduced in version 6.5.2. This bug only effects jobs submitted to the Standard universe, since those are the only jobs that can perform periodic checkpoints in Condor. As soon as the PERIODIC_CHECKPOINT expression evaluates to TRUE on a machine, the condor_ startd will get in a loop where it continues to request a periodic checkpoint every 5 seconds until the job is evicted from the machine. The work-around for this bug in version 6.5.2 is to set PERIODIC_CHECKPOINT = False in your condor_config file.

New Features:

The collector and negotiator can now run on configurable ports, instead of relying on hard-coded values. To use this feature, many places in Condor where you could previously only provide a hostname now understand ``hostname:port'' notation. For example, in your config file, you can now use:
```
  COLLECTOR_HOST = $(CONDOR_HOST):9650
  NEGOTIATOR_HOST = $(CONDOR_HOST):9651
  FLOCK_TO = your-other.collector.domain.org:7002
```
If you define COLLECTOR_HOST in this way, all Condor tools will automatically use the specified port if you are using them in the local pool (so you do not need to use any special options to the tools to get them to find your condor_ collector listening on the new port).
In addition, the -pool option to all Condor tools now understands the ``hostname:port'' notation for remote pools. To query a remote pool with a collector listening on a non-standard port, you can use this:
```
condor_status -pool your-other.collector.domain.org:7002
```
When the Condor daemons start up, they now log the names of the Configuration files they are using, right after the startup banner.
The condor_ config_val program now has a -verbose option which will tell you in which configuration file and line number a condor configuration parameter is defined, and a -config option which will simply list all of the configuration files in use by Condor.
There are new attributes to control Condor's file transfer mechanism. Not only will they hopefully be more clear and easy to use than the old ones, they also provide new functionality. There is now an option to only transfer files ``IF_NEEDED''. In this case, if the job is matched with a machine in the same FileSystemDomain, the files are not transfered, but if the job runs at a machine in a different FileSystemDomain, the files are transfered automatically. For more information on using the Condor File Transfer mechanism, see section 2.5.4 on page , or the condor_ submit man page on page .
Added support for Condor's new ``Computing on Demand'' functionality. Documentation is not yet available, but will be coming soon.
DAGMan no longer requires that all jobs in a DAG have the same log file.
The value of the JOB_RENICE_INCREMENT configuration parameter can now be an arbitrary ClassAd expression rather than just a static integer. The expression will be evaluated by the condor_ starter for each job just before it runs, and can refer to any attribute in the job ClassAd.
The condor_ collector now publishes statistics about the running jobs. In particular, it now publishes the number of jobs running in each Universe, both as a snapshot, and the maximum of the snapshots for each month.
The UNIX condor_ collector has the ability to fork off child processes to handle queries. The COLLECTOR_QUERY_WORKERS parameter is used to specify the maximum number of these worker processes and defaults to zero.
All daemons now publish ``sequence number'' and ``start time'' information in their ClassAds.
The Collector now maintains and publishes update statistics using the above ClassAd ``sequence number'' and ``start time'' information. History information is stored for the past COLLECTOR_CLASS_HISTORY_SIZE updates and is also published in the Collector's ClassAd as a hex string.
The condor_ schedd can now use TCP connections to send updates to pools that it is configured to flock to. You can now define TCP_UPDATE_COLLECTORS list and any collectors listed there, including ones the condor_ schedd is flocking with, will be updated with TCP. Also, the condor_ master uses the same list to decide if it should use TCP to update any collectors listed in the SECONDARY_COLLECTORS_LIST . For more infomation on TCP collector updates in Condor and if your site would want to enable them, read section 3.7.4 on ``Using TCP to Send Collector Updates'' on page .
You no longer need to define FLOCK_VIEW_SERVERS in your config file if you have configured a condor_ schedd to flock to other pools. This is now handled automatically, so you only have to define FLOCK_TO .
If no STARTD_EXPRS is specified, the condor_ startd now defaults to ``JobUniverse''.
Globus universe jobs under Condor-G now send email on job completion based on the notification setting in the submit file.
When submitting a Globus universe job to Condor-G the input, output, and error files can now optionally be an http, https, ftp, or gsiftp URL. (The actual file transfer is handled by globus-url-copy on the remote site.)

Bugs Fixed:

DAGMan now correctly reports an error and rejects DAGs which contain two nodes with the same name, regardless of their case. (DAGMan has rejected duplicate node names since Condor 6.4.6, but until now it would fail to do so if there was any difference in their case.)
When condor_ submit_dag checks job submit files for proper ``log'' statements, it now correctly recognizes lines with leading whitespace.
Fixed a minor bug whereby DAGMan was not removing its lock file after successful completion.
Fixed a bug introduced in version 6.5.0 whereby the UID_DOMAIN attributes of jobs and resources were being compared in a case-sensitive manner, resulting in erroneous failures.
The condor_ master used to always pass a -f on the command line to all daemons defined in the DAEMON_LIST config file setting. However, if you include entries which are not Condor daemons, the condor_ master will no longer add a -f.
The files specified in transfer_input_files and/or transfer_output_files can now contain spaces in the filenames. If there are multiple names, they must be seperated by a comma. Any spaces are considered part of a filename, not separators between filenames. This allows filenames containing spaces (which are common on Windows platforms) to be easily described.

Version 6.5.1

New Features:

DAGMan now supports both Condor computational jobs and Stork data placement (DaP) jobs. (See http://www.cs.wisc.edu/condor/stork/ for more info on Stork.)
Starter exceptions, such as failure to open standard input, are now recorded in the user log.
Added new -forcex argument to condor_ rm to force the immediate local removal of (typically Globus universe) jobs in the 'X' state, regardless of their remote state.

Bugs Fixed:

When transfer_files is being used, the path to the stdout/stderr files was not being respected. After hese files have been transferred, they are now copied to the location specified in the submit file.
A DAGMan bug introduced in Condor 6.5.0 has been fixed, where DAGMan could crash (with a failed assertion) when recovering from a rescue DAG.
Fixed a bug in the example condor_config.generic and hawkeye_config files, where COLLECTOR_HOST was being included in the default STARTD_EXPRS in non-string form, resulting in an invalid value for that attribute in machine classads.

Known Bugs:

DAGMan doesn't detect when users mistakenly specify two DAG nodes with the same node name; instead it waits for the same node to complete twice, which never happens, and so DAGMan goes off into never-never land.

Version 6.5.0

New Features:

A fresh value of RemoteWallClock is now used when evaluating user policy expressions, such as periodic_remove.
The IOProxy handler now handles escaped characters (whitespace) in filenames.
condor.boot is now configured to work automatically with Red Hat chkconfig.
A new log_xml option has been added to condor_submit. It is documented in the condor_submit portion of the manual.
A new DAGMan option to produce dot files was added. Dot is a program that creates visualizations of DAGs. This feature is documented in Section 2.11.
The email report from condor_preen is now less cryptic, and more self-explanatory.
Specifying full device paths (e.g., ``/dev/mouse'') instead of bare device names (e.g., ``mouse'') in CONSOLE_DEVICES in the config file is no longer an error.
The condor_submit tool now prints a more helpful, specific error if the specified job executable is not found, or can't be accessed.
The startd ``cron'' (Hawkeye) now permits zero length ``prefix'' strings.
A number of new Hawkeye modules have been added, and most have various bug fixes and improvements.
Added support for a new config parameter, Q_QUERY_TIMEOUT, which defines the timeout that condor_ q uses when communicating with the condor_ schedd.
Added the ability to use TCP to send ClassAd updates to the Condorcollector, though the feature is disabled by default. Read section 3.7.4 on ``Using TCP to Send Collector Updates'' on page for more details and a discussion of when a site would need to enable this functionality.
Scheduler Universe has been ported to Windows. This enables condor_ dagman to run on Windows as well.

Bugs Fixed:

Hawkeye will no longer busy-loop if a ``continuous mode'' module with a period of 0 fails to execute for some reason. (Now, for continuous mode modules, a period of 0 is automatically reset to be 1, and a warning appears in the log.)
Fixed a very rare potential bug when initializing a user log file. Improved the error messages generated when there are problems initializing the user log to include string descriptions of the errors, not just the error number (errno).
The default value in the config files for LOCK is now defined in terms of LOG , instead of using LOCAL_DIR and appending ``log''. This is a very minor correction, but in the rare cases where the log directory is being redefined for some reason, we usually want that to apply to the lock files, as well. Of course, if the log directory is on NFS, LOCK should still be customized to point to a directoy on a local file system.
Fixed a bug in the condor_ schedd where Scheduler Universe jobs with copy_to_spool = false would fail.

Known Bugs:

DAGMan doesn't detect when users mistakenly specify two DAG nodes with the same node name; instead it waits for the same node to complete twice, which never happens, and so DAGMan goes off into never-never land.

Next: 8.7 Stable Release Series Up: 8. Version History and Previous: 8.5 Stable Release Series Contents Index

condor-admin@cs.wisc.edu