Managing a Job¶
This section provides a brief summary of what can be done once jobs are submitted. The basic mechanisms for monitoring a job are introduced, but the commands are discussed briefly. You are encouraged to look at the man pages of the commands referred to (located in Command Reference Manual (man pages)) for more information.
Checking on the progress of jobs¶
You can check on your jobs with the condor_q command. This command has many options, by default, it displays only your jobs queued in the local scheduler. An example of the output from condor_q is
$ condor_q
-- Schedd: submit.chtc.wisc.edu : <127.0.0.1:9618?... @ 12/31/69 23:00:00
OWNER BATCH_NAME SUBMITTED DONE RUN IDLE HOLD TOTAL JOB_IDS
nemo batch23 4/22 20:44 _ _ _ 1 _ 3671850.0
nemo batch24 4/22 20:56 _ _ _ 1 _ 3673477.0
nemo batch25 4/22 20:57 _ _ _ 1 _ 3673728.0
nemo batch26 4/23 10:44 _ _ _ 1 _ 3750339.0
nemo batch27 7/2 15:11 _ _ _ _ _ 7594591.0
nemo batch28 7/10 03:22 4428 3 _ _ 4434 7801943.0 ... 7858552.0
nemo batch29 7/14 14:18 5074 1182 30 19 80064 7859129.0 ... 7885217.0
nemo batch30 7/14 14:18 5172 1088 28 30 58310 7859106.0 ... 7885192.0
2388 jobs; 0 completed, 1 removed, 58 idle, 2276 running, 53 held, 0 suspended
The goal of the HTCondor system is to effectively manage many jobs. As you may have thousands of jobs in a queue, by default condor_q summarizes many similiar jobs on one line. Depending on the types of your jobs, this output may look a little different.
Often, when you are starting out, and have few jobs, you may want to see one line of output per job. The -nobatch option to condor_q does this, and output might look something like:
$ condor_q -nobatch
-- Schedd submit.chtc.wisc.edu : <127.0.0.1:9618?...
ID OWNER SUBMITTED RUN_TIME ST PRI SIZE CMD
1297254.0 nemo 5/31 18:05 14+17:40:01 R 0 7.3 condor_dagman
1297255.0 nemo 5/31 18:05 14+17:39:55 R 0 7.3 condor_dagman
1297256.0 nemo 5/31 18:05 14+17:39:55 R 0 7.3 condor_dagman
1297259.0 nemo 5/31 18:05 14+17:39:55 R 0 7.3 condor_dagman
1297261.0 nemo 5/31 18:05 14+17:39:55 R 0 7.3 condor_dagman
1302278.0 nemo 6/4 12:22 1+00:05:37 I 0 390.6 mdrun_1.sh
1304740.0 nemo 6/5 00:14 1+00:03:43 I 0 390.6 mdrun_1.sh
1304967.0 nemo 6/5 05:08 0+00:00:00 I 0 0.0 mdrun_1.sh
14 jobs; 4 idle, 8 running, 2 held
This still only shows your jobs. You can display information about all the users with jobs in this scheduler by adding the -allusers option to condor_q.
The output contains many columns of information about the queued jobs. The ST column (for status) shows the status of current jobs in the queue:
- R
The job is currently running.
- I
The job is idle. It is not running right now, because it is waiting for a machine to become available.
- H
The job is the hold state. In the hold state, the job will not be scheduled to run until it is released. See the condor_hold and the condor_release manual pages.
The RUN_TIME time reported for a job is the time that has been committed to the job.
Another useful method of tracking the progress of jobs is through the
job event log. The specification of a log in the submit description
file causes the progress of the job to be logged in a file. Follow the
events by viewing the job event log file. Various events such as
execution commencement, checkpoint, eviction and termination are logged
in the file. Also logged is the time at which the event occurred.
When a job begins to run, HTCondor starts up a condor_shadow process on the submit machine. The shadow process is the mechanism by which the remotely executing jobs can access the environment from which it was submitted, such as input and output files.
It is normal for a machine which has submitted hundreds of jobs to have
hundreds of condor_shadow processes running on the machine. Since the
text segments of all these processes is the same, the load on the submit
machine is usually not significant. If there is degraded performance,
limit the number of jobs that can run simultaneously by reducing the
MAX_JOBS_RUNNING configuration
variable.
You can also find all the machines that are running your job through the
condor_status command.
For example, to find
all the machines that are running jobs submitted by
breach@cs.wisc.edu, type:
$ condor_status -constraint 'RemoteUser == "breach@cs.wisc.edu"'
Name Arch OpSys State Activity LoadAv Mem ActvtyTime
alfred.cs. INTEL LINUX Claimed Busy 0.980 64 0+07:10:02
biron.cs.w INTEL LINUX Claimed Busy 1.000 128 0+01:10:00
cambridge. INTEL LINUX Claimed Busy 0.988 64 0+00:15:00
falcons.cs INTEL LINUX Claimed Busy 0.996 32 0+02:05:03
happy.cs.w INTEL LINUX Claimed Busy 0.988 128 0+03:05:00
istat03.st INTEL LINUX Claimed Busy 0.883 64 0+06:45:01
istat04.st INTEL LINUX Claimed Busy 0.988 64 0+00:10:00
istat09.st INTEL LINUX Claimed Busy 0.301 64 0+03:45:00
...
To find all the machines that are running any job at all, type:
$ condor_status -run
Name Arch OpSys LoadAv RemoteUser ClientMachine
adriana.cs INTEL LINUX 0.980 hepcon@cs.wisc.edu chevre.cs.wisc.
alfred.cs. INTEL LINUX 0.980 breach@cs.wisc.edu neufchatel.cs.w
amul.cs.wi X86_64 LINUX 1.000 nice-user.condor@cs. chevre.cs.wisc.
anfrom.cs. X86_64 LINUX 1.023 ashoks@jules.ncsa.ui jules.ncsa.uiuc
anthrax.cs INTEL LINUX 0.285 hepcon@cs.wisc.edu chevre.cs.wisc.
astro.cs.w INTEL LINUX 1.000 nice-user.condor@cs. chevre.cs.wisc.
aura.cs.wi X86_64 WINDOWS 0.996 nice-user.condor@cs. chevre.cs.wisc.
balder.cs. INTEL WINDOWS 1.000 nice-user.condor@cs. chevre.cs.wisc.
bamba.cs.w INTEL LINUX 1.574 dmarino@cs.wisc.edu riola.cs.wisc.e
bardolph.c INTEL LINUX 1.000 nice-user.condor@cs. chevre.cs.wisc.
...
Peeking in on a running job’s output files¶
When a job is running, you may be curious about any output it has created. The condor_tail command can copy output files from a running job on a remote machine back to the submit machine. condor_tail uses the same networking stack as HTCondor proper, so it will work if the execute machine is behind a firewall. Simply run, where xx.yy is the job id of a running job:
$ condor_tail xx.yy
or
$ condor_tail -f xx.yy
to continuously follow the standard output. To copy a different file, run
$ condor_tail xx.yy name_of_output_file
Starting an interactive shell next to a running job on a remote machine¶
condor_ssh_to_job is a very powerful command, but is not available on all platforms, or all installations. Some administrators disable it, so check with your local site if it does not appear to work. condor_ssh_to_job takes the job id of a running job as an argument, and establishes a shell running on the node next to the job. The environment of this shell is a similar to the job as possible. Users of condor_ssh_to_job can look at files, attach to their job with the debugger and otherwise inspect the job.
Removing a job from the queue¶
A job can be removed from the queue at any time by using the condor_rm command. If the job that is being removed is currently running, the job is killed without a checkpoint, and its queue entry is removed. The following example shows the queue of jobs before and after a job is removed.
$ condor_q -nobatch
-- Schedd: froth.cs.wisc.edu : <128.105.73.44:33847> : froth.cs.wisc.edu
ID OWNER SUBMITTED CPU_USAGE ST PRI SIZE CMD
125.0 raman 4/11 14:37 0+00:00:00 R 0 1.4 sleepy
132.0 raman 4/11 16:57 0+00:00:00 R 0 1.4 hello
2 jobs; 1 idle, 1 running, 0 held
$ condor_rm 132.0
Job 132.0 removed.
$ condor_q -nobatch
-- Schedd: froth.cs.wisc.edu : <128.105.73.44:33847> : froth.cs.wisc.edu
ID OWNER SUBMITTED CPU_USAGE ST PRI SIZE CMD
125.0 raman 4/11 14:37 0+00:00:00 R 0 1.4 sleepy
1 jobs; 1 idle, 0 running, 0 held
Placing a job on hold¶
A job in the queue may be placed on hold by running the command condor_hold. A job in the hold state remains in the hold state until later released for execution by the command condor_release.
Use of the condor_hold command causes a hard kill signal to be sent to a currently running job (one in the running state).
Jobs that are running when placed on hold will start over from the beginning when released.
The condor_hold and the condor_release manual pages contain usage details.
Changing the priority of jobs¶
In addition to the priorities assigned to each user, HTCondor also provides each user with the capability of assigning priorities to each submitted job. These job priorities are local to each queue and can be any integer value, with higher values meaning better priority.
The default priority of a job is 0, but can be changed using the condor_prio command. For example, to change the priority of a job to -15,
$ condor_q -nobatch raman
-- Submitter: froth.cs.wisc.edu : <128.105.73.44:33847> : froth.cs.wisc.edu
ID OWNER SUBMITTED CPU_USAGE ST PRI SIZE CMD
126.0 raman 4/11 15:06 0+00:00:00 I 0 0.3 hello
1 jobs; 1 idle, 0 running, 0 held
$ condor_prio -p -15 126.0
$ condor_q -nobatch raman
-- Submitter: froth.cs.wisc.edu : <128.105.73.44:33847> : froth.cs.wisc.edu
ID OWNER SUBMITTED CPU_USAGE ST PRI SIZE CMD
126.0 raman 4/11 15:06 0+00:00:00 I -15 0.3 hello
1 jobs; 1 idle, 0 running, 0 held
It is important to note that these job priorities are completely different from the user priorities assigned by HTCondor. Job priorities do not impact user priorities. They are only a mechanism for the user to identify the relative importance of jobs among all the jobs submitted by the user to that specific queue.
Why is the job not running?¶
Users occasionally find that their jobs do not run. There are many possible reasons why a specific job is not running. The following prose attempts to identify some of the potential issues behind why a job is not running.
At the most basic level, the user knows the status of a job by using condor_q to see that the job is not running. By far, the most common reason (to the novice HTCondor job submitter) why the job is not running is that HTCondor has not yet been through its periodic negotiation cycle, in which queued jobs are assigned to machines within the pool and begin their execution. This periodic event occurs by default once every 5 minutes, implying that the user ought to wait a few minutes before searching for reasons why the job is not running.
Further inquiries are dependent on whether the job has never run at all, or has run for at least a little bit.
For jobs that have never run, many problems can be diagnosed by using the -analyze option of the condor_q command. Here is an example; running condor_q ‘s analyzer provided the following information:
$ condor_q -analyze 27497829
-- Submitter: s1.chtc.wisc.edu : <128.104.100.43:9618?sock=5557_e660_3> : s1.chtc.wisc.edu
User priority for ei@chtc.wisc.edu is not available, attempting to analyze without it.
---
27497829.000: Run analysis summary. Of 5257 machines,
5257 are rejected by your job's requirements
0 reject your job because of their own requirements
0 match and are already running your jobs
0 match but are serving other users
0 are available to run your job
No successful match recorded.
Last failed match: Tue Jun 18 14:36:25 2013
Reason for last match failure: no match found
WARNING: Be advised:
No resources matched request's constraints
The Requirements expression for your job is:
( OpSys == "OSX" ) && ( TARGET.Arch == "X86_64" ) &&
( TARGET.Disk >= RequestDisk ) && ( TARGET.Memory >= RequestMemory ) &&
( ( TARGET.HasFileTransfer ) || ( TARGET.FileSystemDomain == MY.FileSystemDomain ) )
Suggestions:
Condition Machines Matched Suggestion
--------- ---------------- ----------
1 ( target.OpSys == "OSX" ) 0 MODIFY TO "LINUX"
2 ( TARGET.Arch == "X86_64" ) 5190
3 ( TARGET.Disk >= 1 ) 5257
4 ( TARGET.Memory >= ifthenelse(MemoryUsage isnt undefined,MemoryUsage,1) )
5257
5 ( ( TARGET.HasFileTransfer ) || ( TARGET.FileSystemDomain == "submit-1.chtc.wisc.edu" ) )
5257
This example also shows that the job does not run because the platform requested, Mac OS X, is not available on any of the machines in the pool. Recall that unless informed otherwise in the Requirements expression in the submit description file, the platform requested for an execute machine will be the same as the platform where condor_submit is run to submit the job. And, while Mac OS X is a Unix-type operating system, it is not the same as Linux, and thus will not match with machines running Linux.
While the analyzer can diagnose most common problems, there are some situations that it cannot reliably detect due to the instantaneous and local nature of the information it uses to detect the problem. Thus, it may be that the analyzer reports that resources are available to service the request, but the job still has not run. In most of these situations, the delay is transient, and the job will run following the next negotiation cycle.
A second class of problems represents jobs that do or did run, for at least a short while, but are no longer running. The first issue is identifying whether the job is in this category. The condor_q command is not enough; it only tells the current state of the job. The needed information will be in the log file or the error file, as defined in the submit description file for the job. If these files are not defined, then there is little hope of determining if the job ran at all. For a job that ran, even for the briefest amount of time, the log file will contain an event of type 1, which will contain the string Job executing on host.
A job may run for a short time, before failing due to a file permission problem. The log file used by the condor_shadow daemon will contain more information if this is the problem. This log file is associated with the machine on which the job was submitted. The location and name of this log file may be discovered on the submitting machine, using the command
$ condor_config_val SHADOW_LOG
Job in the Hold State¶
Should HTCondor detect something about a job that would prevent it from ever running successfully, say, because the executable doesn’t exist, or input files are missing, HTCondor will put the job in Hold state. A job in the Hold state will remain in the queue, and show up in the output of the condor_q command, but is not eligible to run. The job will stay in this state until it is released or removed. Users may also hold their jobs manually with the condor_hold command.
A table listing the reasons why a job may be held is at the Job ClassAd Attributes section. A string identifying the reason that a particular job is in the Hold state may be displayed by invoking condor_q -hold. For the example job ID 16.0, use:
$ condor_q -hold 16.0
This command prints information about the job, including the job ClassAd
attribute HoldReason.
In the Job Event Log File¶
In a job event log file are a listing of events in chronological order that occurred during the life of one or more jobs. The formatting of the events is always the same, so that they may be machine readable. Four fields are always present, and they will most often be followed by other fields that give further information that is specific to the type of event.
The first field in an event is the numeric value assigned as the event
type in a 3-digit format. The second field identifies the job which
generated the event. Within parentheses are the job ClassAd attributes
of ClusterId value, ProcId value, and the node number for
parallel universe jobs or a set of zeros (for jobs run under all other
universes), separated by periods. The third field is the date and time
of the event logging. The fourth field is a string that briefly
describes the event. Fields that follow the fourth field give further
information for the specific event type.
A complete list of these values is at Job Event Log Codes section.
Job Termination¶
From time to time, and for a variety of reasons, HTCondor may terminate a job before it completes. For instance, a job could be removed (via condor_rm), preempted (by a user a with higher priority), or killed (for using more memory than it requested). In these cases, it might be helpful to know why HTCondor terminated the job. HTCondor calls its records of these reasons “Tickets of Execution”.
A ticket of execution is usually issued by the condor_startd, and includes:
when the condor_startd was told, or otherwise decided, to terminate the job (the
whenattribute);who made the decision to terminate, usually a Sinful string (the
whoattribute);and what method was employed to command the termination, as both as string and an integer (the
HowandHowCodeattributes).
The relevant log events include a human-readable rendition of the ToE, and the job ad is updated with the ToE after the usual delay.
As of version 8.9.4, HTCondor only issues ToE in three cases:
when the job terminates of its own accord (issued by the starter,
HowCode0);and when the startd terminates the job because it received a
DEACTIVATE_CLAIMcommmand (HowCode1)or a
DEACTIVATE_CLAIM_FORCIBLYcommand (HowCode2).
In both cases, HTCondor records the ToE in the job ad. In the event log(s), event 005 (job completion) includes the ToE for the first case, and event 009 (job aborted) includes the ToE for the second and third cases.
Future HTCondor releases will issue ToEs in additional cases and include them in additional log events.
Job Completion¶
When an HTCondor job completes, either through normal means or by abnormal termination by signal, HTCondor will remove it from the job queue. That is, the job will no longer appear in the output of condor_q, and the job will be inserted into the job history file. Examine the job history file with the condor_history command. If there is a log file specified in the submit description file for the job, then the job exit status will be recorded there as well, along with other information described below.
By default, HTCondor does not send an email message when the job completes. Modify this behavior with the notification command in the submit description file. The message will include the exit status of the job, which is the argument that the job passed to the exit system call when it completed, or it will be notification that the job was killed by a signal. Notification will also include the following statistics (as appropriate) about the job:
- Submitted at:
when the job was submitted with condor_submit
- Completed at:
when the job completed
- Real Time:
the elapsed time between when the job was submitted and when it completed, given in a form of
<days> <hours>:<minutes>:<seconds>- Virtual Image Size:
memory size of the job, computed when the job checkpoints
Statistics about just the last time the job ran:
- Run Time:
total time the job was running, given in the form
<days> <hours>:<minutes>:<seconds>- Remote User Time:
total CPU time the job spent executing in user mode on remote machines; this does not count time spent on run attempts that were evicted without a checkpoint. Given in the form
<days> <hours>:<minutes>:<seconds>- Remote System Time:
total CPU time the job spent executing in system mode (the time spent at system calls); this does not count time spent on run attempts that were evicted without a checkpoint. Given in the form
<days> <hours>:<minutes>:<seconds>
The Run Time accumulated by all run attempts are summarized with the
time given in the form <days> <hours>:<minutes>:<seconds>.
And, statistics about the bytes sent and received by the last run of the job and summed over all attempts at running the job are given.
The job terminated event includes the following:
the type of termination (normal or by signal)
the return value (or signal number)
local and remote usage for the last (most recent) run (in CPU-seconds)
local and remote usage summed over all runs (in CPU-seconds)
bytes sent and received by the job’s last (most recent) run,
bytes sent and received summed over all runs,
a report on which partitionable resources were used, if any. Resources include CPUs, disk, and memory; all are lifetime peak values.
Your administrator may have configured HTCondor to report on other resources, particularly GPUs (lifetime average) and GPU memory usage (lifetime peak). HTCondor currently assigns all the usage of a GPU to the job running in the slot to which the GPU is assigned; if the admin allows more than one job to run on the same GPU, or non-HTCondor jobs to use the GPU, GPU usage will be misreported accordingly.
When configured to report GPU usage, HTCondor sets the following two attributes in the job:
GPUsUsageGPU usage over the lifetime of the job, reported as a fraction of the the maximum possible utilization of one GPU.
GPUsMemoryUsagePeak memory usage over the lifetime of the job, in megabytes.
Summary of all HTCondor users and their jobs¶
When jobs are submitted, HTCondor will attempt to find resources to run the jobs. A list of all those with jobs submitted may be obtained through condor_status with the -submitters option. An example of this would yield output similar to:
$ condor_status -submitters
Name Machine Running IdleJobs HeldJobs
ballard@cs.wisc.edu bluebird.c 0 11 0
nice-user.condor@cs. cardinal.c 6 504 0
wright@cs.wisc.edu finch.cs.w 1 1 0
jbasney@cs.wisc.edu perdita.cs 0 0 5
RunningJobs IdleJobs HeldJobs
ballard@cs.wisc.edu 0 11 0
jbasney@cs.wisc.edu 0 0 5
nice-user.condor@cs. 6 504 0
wright@cs.wisc.edu 1 1 0
Total 7 516 5