HTCondor uses a few different directories, some of which are role-specific. Do not use these directories for any other purpose, and do not share these directories between machines. The directories are listed in here by the name of the configuration option used to tell HTCondor where they are; you will not normally need to change these.
Directories used by More than One Role
Each HTCondor daemon writes its own log file, and each log file is placed in the LOG directory. You can configure the name of each daemon’s log by setting <SUBSYS>_LOG, although you should never need to do so. You can also control the sizes of the log files or how often they rotate; see Daemon Logging Configuration File Entries for details. If you want to write your logs to a shared filesystem, we recommend including
$(HOSTNAME)in the value of
LOGrather than changing the names of each individual log to not collide. If you set
LOGto a shared filesystem, you should set
LOCKto a local filesystem; see below.
HTCondor uses a small number of lock files to synchronize access to certain files that are shared between multiple daemons. Because of problems encountered with file locking and network file systems (particularly NFS), these lock files should be placed on a local filesystem on each machine. By default, they are placed in the
Directories use by the Submit Role
The SPOOL directory holds two types of files: system data and (user) job data. The former includes the job queue and history files. The latter includes:
the files transferred, if any, when a job which set
the input and output files of remotely-submitted jobs.
the checkpoint files stored by self-checkpointing jobs.
Disk usage therefore varies widely based on the job mix, but since the schedd will abort if it can’t append to the job queue log, you want to make sure this directory is on a partition which won’t run out of space.
To help ensure this, you may set JOB_QUEUE_LOG to separate the job queue log (system data) from the (user) job data. This can also be used to increase performance (or reliability) by moving the job queue log to specialized hardware (an SSD or a a high-redundancy RAID, for example).
Directories use by the Execute Role
The EXECUTE directory is the parent directory of the current working directory for any HTCondor job that runs on a given execute-role machine. HTCondor copies the executable and input files for a job to its subdirectory; the job’s standard output and standard error streams are also logged here. Jobs will also almost always generate their output here as well, so the
EXECUTEdirectory should provide a plenty of space.
EXECUTEshould not be placed under /tmp or /var/tmp if possible, as HTCondor loses the ability to make /tmp and /var/tmp private to the job. While not a requirement, ideally
EXECUTEshould be on a distinct filesystem, so that it is impossible for a rogue job to fill up non-HTCondor related partitions.
Usually, the per-job scratch execute directory is created by the startd as a directory under
EXECUTE. However, on Linux machines where HTCondor has root privilege, it can be configured to make an ephemeral, per-job scratch filesystem backed either by LVM, if it is configured, or a large existing file on the filesystem.
There are several advantages to this approach. The first is that disk space is more accurately measured and enforced. HTCondor can get the disk usage by a single system call, instead of traversing what might be a very deep directory hierarchy. There may also be performance benefits, as this filesystem never needs to survive a reboot, and is thus mounted with mount options that provide the least amount of disk consistence in the face of a reboot. Also, when the job exits, all the files in the filesystem can be removed by simply unmounting and destroying the filesystem, which is much faster than having condor remove each scratch file in turn.
To enable this, first set STARTD_ENFORCE_DISK_LIMITS to
true. Then, if LVM is installed and configured, set THINPOOL_NAME to the name of a logical volume.
"condor_lv"might be a good choice. Finally, set THINPOOL_VOLUME_GROUP to the name of the volume group the LVM administrator has created for this purpose.
"condor_vg"might be a good name. If there is no LVM on the system, a single large existing file can be used as the backing store, in which case the knob THINPOOL_BACKING_FILE should be set to the name of the existing large file on disk that HTCondor will use to make filesystems from.
The per job filesystem feature is a work in progress and not currently supported.