1 Using Condor An Introduction ICE 2010

2 The Condor Project (Established ‘85) Distributed High Throughput Computing research performed by a team of ~35 faculty, full time staff and students.

3 Definitions › Job  The Condor representation of your work › Machine  The Condor representation of computers and that can perform the work › Match Making  Matching a job with a machine “Resource”

4 Job Jobs state their requirements and preferences: I need a Linux/x86 platform I need the machine at least 500 Mb I prefer a machine with more memory

5 Machine Machines state their requirements and preferences: Run jobs only when there is no keyboard activity I prefer to run Frieda’s jobs I am a machine in the econ department Never run jobs belonging to Dr. Smith

6 The Magic of Matchmaking › Jobs and machines state their requirements and preferences › Condor matches jobs with machines based on requirements and preferences

7 Getting Started: Submitting Jobs to Condor › Overview:  Choose a “Universe” for your job  Make your job “batch-ready”  Create a submit description file  Run condor_submit to put your job in the queue

Choose the “Universe” › Controls how Condor handles jobs › Choices include:  Vanilla  Standard  Grid  Java  Parallel  VM

9 Using the Vanilla Universe The Vanilla Universe: – Allows running almost any “serial” job – Provides automatic file transfer, etc. – Like vanilla ice cream Can be used in just about any situation

Make your job batch- ready Must be able to run in the background No interactive input No GUI/window clicks No music ;^)

11 Make your job batch-ready (continued)…  Job can still use STDIN, STDOUT, and STDERR (the keyboard and the screen), but files are used for these instead of the actual devices  Similar to UNIX shell: $./myprogram output.txt

Create a Submit Description File › A plain ASCII text file › Condor does not care about file extensions › Tells Condor about your job:  Which executable, universe, input, output and error files to use, command-line arguments, environment variables, any special requirements or preferences (more on this later) › Can describe many jobs at once (a “cluster”), each with different input, arguments, output, etc.

13 Simple Submit Description File # Simple condor_submit input file # (Lines beginning with # are comments) # NOTE: the words on the left side are not # case sensitive, but filenames are! Universe = vanilla Executable = my_job Output = output.txt Queue

Run condor_submit › You give condor_submit the name of the submit file you have created:  condor_submit my_job.submit › condor_submit:  Parses the submit file, checks for errors  Creates a “ClassAd” that describes your job(s)  Puts job(s) in the Job Queue

15 The Job Queue › condor_submit sends your job’s ClassAd(s) to the schedd › The schedd (more details later):  Manages the local job queue  Stores the job in the job queue Atomic operation, two-phase commit “Like money in the bank” › View the queue with condor_q

16 Example condor_submit and condor_q % condor_submit my_job.submit Submitting job(s). 1 job(s) submitted to cluster 1. % condor_q -- Submitter: perdita.cs.wisc.edu : : ID OWNER SUBMITTED RUN_TIME ST PRI SIZE CMD 1.0 frieda 6/16 06: :00:00 I my_job 1 jobs; 1 idle, 0 running, 0 held %

17 Input, output & error files › Controlled by submit file settings › You can define the job’s standard input, standard output and standard error:  Read job’s standard input from “input_file”: Input= input_file Shell equivalent: program <input_file  Write job’s standard ouput to “output_file”: Output= output_file Shell equivalent: program >output_file  Write job’s standard error to “error_file”: Error= error_file Shell equivalent: program 2>error_file

18 about your job Condor sends about job events to the submitting user Specify “notification” in your submit file to control which events: Notification= complete Notification= never Notification= error Notification= always Default

19 Feedback on your job › Create a log of job events › Add to submit description file: log = sim.log › Becomes the Life Story of a Job  Shows all events in the life of a job  Always have a log file

20 Sample Condor User Log 000 ( ) 05/25 19:10:03 Job submitted from host: ( ) 05/25 19:12:17 Job executing on host: ( ) 05/25 19:13:06 Job terminated. (1) Normal termination (return value 0)...

21 Example Submit Description File With Logging # Example condor_submit input file # (Lines beginning with # are comments) # NOTE: the words on the left side are not # case sensitive, but filenames are! Universe = vanilla Executable = /home/frieda/condor/my_job.condor Log = my_job.log ·Job log (from Condor) Input = my_job.in ·Program’s standard input Output = my_job.out ·Program’s standard output Error = my_job.err ·Program’s standard error Arguments = -a1 -a2 ·Command line arguments InitialDir = /home/frieda/condor/run Queue

22 Let’s run a job › First, need a terminal emulator  (or similar) › Login to chopin.cs.wisc.edu as  cguserXX, and the given password

23 Logged In? › condor_q › condor_status

24 Create submit file › nano submit.your_initials universe = vanilla executable = /bin/echo Arguments = hello world Output = out Log = log queue

25 And submit it… › condor_submit submit.your_initials › (wait… remember the HTC bit?) › Condor_q xx › cat output

26 “Clusters” and “Processes” › If your submit file describes multiple jobs, we call this a “cluster” › Each cluster has a unique “cluster number” › Each job in a cluster is called a “process”  Process numbers always start at zero › A Condor “Job ID” is the cluster number, a period, and the process number (i.e. 2.1)  A cluster can have a single process Job ID = 20.0 ·Cluster 20, process 0  Or, a cluster can have more than one process Job ID: 21.0, 21.1, 21.2·Cluster 21, process 0, 1, 2

27 Submit File for a Cluster # Example submit file for a cluster of 2 jobs # with separate input, output, error and log files Universe = vanilla Executable = my_job Arguments = -x 0 log = my_job_0.log Input = my_job_0.in Output = my_job_0.out Error = my_job_0.err Queue ·Job 2.0 (cluster 2, process 0) Arguments = -x 1 log = my_job_1.log Input = my_job_1.in Output = my_job_1.out Error = my_job_1.err Queue ·Job 2.1 (cluster 2, process 1)

28 % condor_submit my_job.submit-file Submitting job(s). 2 job(s) submitted to cluster 2. % condor_q -- Submitter: perdita.cs.wisc.edu : : ID OWNER SUBMITTED RUN_TIME ST PRI SIZE CMD 1.0 frieda 4/15 06: :02:11 R my_job –a1 –a2 2.0 frieda 4/15 06: :00:00 I my_job –x frieda 4/15 06: :00:00 I my_job –x 1 3 jobs; 2 idle, 1 running, 0 held % Submitting The Job

29 Organize your files and directories for big runs › Create subdirectories for each “run”  run_0, run_1, … run_599 › Create input files in each of these  run_0/simulation.in  run_1/simulation.in  …  run_599/simulation.in › The output, error & log files for each job will be created by Condor from your job’s output

30 Submit Description File for 600 Jobs # Cluster of 600 jobs with different directories Universe = vanilla Executable = sim Log = simulation.log... Arguments = -x 0 InitialDir = run_0 ·Log, input, output & error files -> run_0 Queue ·Job 3.0 (Cluster 3, Process 0) Arguments = -x 1 InitialDir = run_1 ·Log, input, output & error files -> run_1 Queue ·Job 3.1 (Cluster 3, Process 1) ·Do this 598 more times…………

31 Submit File for a Big Cluster of Jobs › We just submitted 1 cluster with 600 processes › All the input/output files will be in different directories › The submit file is pretty unwieldy (over 1200 lines) › Isn’t there a better way?

32 Submit File for a Big Cluster of Jobs (the better way) #1 › We can queue all 600 in 1 “Queue” command  Queue 600 › Condor provides $(Process) and $(Cluster)  $(Process) will be expanded to the process number for each job in the cluster 0, 1, … 599  $(Cluster) will be expanded to the cluster number Will be 4 for all jobs in this cluster

33 Submit File for a Big Cluster of Jobs (the better way) #2 › The initial directory for each job can be specified using $(Process)  InitialDir = run_$(Process)  Condor will expand these to “ run_0 ”, “ run_1 ”, … “ run_599 ” directories › Similarly, arguments can be variable  Arguments = -x $(Process)  Condor will expand these to “-x 0”, “-x 1”, … “-x 599”

34 Better Submit File for 600 Jobs # Example condor_submit input file that defines # a cluster of 600 jobs with different directories Universe = vanilla Executable = my_job Log = my_job.log Input = my_job.in Output = my_job.out Error = my_job.err Arguments = –x $(Process) ·–x 0, -x 1, … -x 599 InitialDir = run_$(Process) ·run_0 … run_599 Queue 600 ·Jobs 4.0 … 4.599

35 Now, we submit it… $ condor_submit my_job.submit Submitting job(s) Logging submit event(s) job(s) submitted to cluster 4.

36 And, Check the queue $ condor_q -- Submitter: x.cs.wisc.edu : : x.cs.wisc.edu ID OWNER SUBMITTED RUN_TIME ST PRI SIZE CMD 4.0 frieda 4/20 12: :00:05 R my_job -arg1 –x frieda 4/20 12: :00:03 I my_job -arg1 –x frieda 4/20 12: :00:01 I my_job -arg1 –x frieda 4/20 12: :00:00 I my_job -arg1 –x frieda 4/20 12: :00:00 I my_job -arg1 –x frieda 4/20 12: :00:00 I my_job -arg1 –x jobs; 599 idle, 1 running, 0 held

37 Removing jobs › If you want to remove a job from the Condor queue, you use condor_rm › You can only remove jobs that you own › Privileged user can remove any jobs  “root” on UNIX  “administrator” on Windows

38 Removing jobs (continued) › Remove an entire cluster:  condor_rm 4 ·Removes the whole cluster › Remove a specific job from a cluster:  condor_rm 4.0 ·Removes a single job › Or, remove all of your jobs with “-a”  condor_rm -a ·Removes all jobs / clusters

39 Submit cluster of 10 jobs › nano submit universe = vanilla executable = /bin/echo Arguments = hello world $(PROCESS) Output = out.$(PROCESS) Log = log Queue 10

40 And submit it… › condor_submit submit › (wait…) › Condor_q xx › cat log › cat output.yy

41 My new jobs run for 20 days… › What happens when a job is forced off it’s CPU?  Preempted by higher priority user or job  Vacated because of user activity › How can I add fault tolerance to my jobs?

42 Condor’s Standard Universe to the rescue! › Support for transparent process checkpoint and restart › Remote system calls (remote I/O)  Your job can read / write files as if they were local

43 Remote System Calls in the Standard Universe › I/O system calls are trapped and sent back to the submit machine Examples: open a file, write to a file › No source code changes typically required › Programming language independent

44 Process Checkpointing in the Standard Universe › Condor’s process checkpointing provides a mechanism to automatically save the state of a job › The process can then be restarted from right where it was checkpointed  After preemption, crash, etc.

45 Checkpointing: Process Starts checkpoint: the entire state of a program, saved in a file  CPU registers, memory image, I/O time

46 Checkpointing: Process Checkpointed time 123

47 Checkpointing: Process Killed time 3 3 Killed!

48 Checkpointing: Process Resumed time 3 3 goodputbadput goodput

49 When will Condor checkpoint your job? › Periodically, if desired  For fault tolerance › When your job is preempted by a higher priority job › When your job is vacated because the execution machine becomes busy › When you explicitly run condor_checkpoint, condor_vacate, condor_off or condor_restart command

50 Making the Standard Universe Work › The job must be relinked with Condor’s standard universe support library › To relink, place condor_compile in front of the command used to link the job: % condor_compile gcc -o myjob myjob.c - OR - % condor_compile f77 -o myjob filea.f fileb.f - OR - % condor_compile make –f MyMakefile

51 Limitations of the Standard Universe › Condor’s checkpointing is not at the kernel level.  Standard Universe the job may not: Fork() Use kernel threads Use some forms of IPC, such as pipes and shared memory › Must have access to source code to relink › Many typical scientific jobs are OK

52 Submitting Std uni job › #include › int main(int argc, char **argv) { › int i; for(i = 0 ; i < ; i++) { } › }

53 And submit… › condor_compile gcc –o foo foo.c  -- Change "vanilla" to "standard"  -- Change "/bin/echo" to "foo" (or above)

54 My jobs have have dependencies… Can Condor help solve my dependency problems?

55 Condor Universes: Scheduler and Local › Scheduler Universe  Plug in a meta-scheduler  Developed for DAGMan (more later)  Similar to Globus’s fork job manager › Local  Very similar to vanilla, but jobs run on the local host  Has more control over jobs than scheduler universe

56 DAGMan › Directed Acyclic Graph Manager › DAGMan allows you to specify the dependencies between your Condor jobs, so it can manage them automatically for you. › (e.g., “Don’t run job “B” until job “A” has completed successfully.”)

57 What is a DAG? › A DAG is the data structure used by DAGMan to represent these dependencies. › Each job is a “node” in the DAG. › Each node can have any number of “parent” or “children” nodes – as long as there are no loops! Job A Job B Job C Job D

58 Defining a DAG › A DAG is defined by a.dag file, listing each of its nodes and their dependencies: # diamond.dag Job A a.sub Job B b.sub Job C c.sub Job D d.sub Parent A Child B C Parent B C Child D › each node will run the Condor job specified by its accompanying Condor submit file Job A Job BJob C Job D

59 Submitting a DAG › To start your DAG, just run condor_submit_dag with your.dag file, and Condor will start a personal DAGMan daemon which to begin running your jobs: % condor_submit_dag diamond.dag › condor_submit_dag is run by the schedd  DAGMan daemon itself is “watched” by Condor, so you don’t have to

60 DAGMan Running a DAG › DAGMan acts as a “meta-scheduler”, managing the submission of your jobs to Condor based on the DAG dependencies. Condor Job Queue B C D A A.dag File

61 DAGMan Running a DAG (cont’d) › DAGMan holds & submits jobs to the Condor queue at the appropriate times. Condor Job Queue D B C B A C

62 Running a DAG (cont’d) › In case of a job failure, DAGMan continues until it can no longer make progress, and then creates a “rescue” file with the current state of the DAG. Condor Job Queue DAGMan X D A B Rescue File

63 Recovering a DAG › Once the failed job is ready to be re-run, the rescue file can be used to restore the prior state of the DAG. Condor Job Queue Rescue File C DAGMan D A B C

64 DAGMan Recovering a DAG (cont’d) › Once that job completes, DAGMan will continue the DAG as if the failure never happened. Condor Job Queue C D A B D

65 DAGMan Finishing a DAG › Once the DAG is complete, the DAGMan job itself is finished, and exits. Condor Job Queue C D A B

66 Additional DAGMan Features › Provides other handy features for job management…  nodes can have PRE & POST scripts  failed nodes can be automatically re- tried a configurable number of times  job submission can be “throttled”

67 What about Licensed Jobs? › e.g. matlab  Site license?  matlab compiler  Octave

68 Chirp › condor_chirp get_file remote local › condor_chirp put_file local remote

69 General User Commands › condor_status View Pool Status › condor_qView Job Queue › condor_submitSubmit new Jobs › condor_rmRemove Jobs › condor_prioIntra-User Prios › condor_historyCompleted Job Info › condor_submit_dagSubmit new DAG › condor_checkpointForce a checkpoint › condor_compileLink Condor library

70 Statistical Bootstrap › Build up from the worker side out › The matlab/octave worker: › worker.m: #!/s/octave/bin/octave -q load "subset" subset; subset = subset(floor(rand(10,1).* 1000)); printf("%f ", mean(subset));

71 Run the worker alone › (won’t work – why?) ›./worker.m

72 Submit file universe = vanilla executable = worker.m should_transfer_files = true when_to_transfer_output = on_exit transfer_input_files = subset output = mean.$(PROCESS) error = foo log = log queue 10

73 Create the initial data driver.m #!/s/octave/bin/octave –q dist_size = ; d = rand(dist_size, 1).* 500; subset = d(floor(rand(1000,1).* )); save "subset" subset;

74 And submit the job… › condor_submit submit

75 Add the submission to the driver script… #!/s/octave/bin/octave –q dist_size = ; d = rand(dist_size, 1).* 500; subset = d(floor(rand(1000,1).* )); save "subset" subset; system("condor_submit submit"); system("condor_wait log");

76 And run the driver! ›./driver.m

77 Parallel convergence checking: Another DAGman example › Evaluating a function at many points › Check for convergence -> retry › Particle Swarm Optimization

78 Prepare Compute Converge? Done Yes! No

79 Any Guesses? › Who has thoughts? › Best to work from “inside out” 79

80 The job itself. #!/bin/sh ###### random.sh echo $RANDOM exit 0 80

81 The submit file › Any guesses? 81

82 The submit file # submitRandom universe = vanilla executable = random.sh Should_transfer_files = yes When_to_transfer_output = on_exit output = out log = log queue 82

83 Next step: the inner DAG 83 First Last Node Node0 Node1Node2Node3Node4 Node11

84 The DAG file › Any guesses? 84

85 The inner DAG file Job Node0 submitRandom Job Node1 submitRandom Job Node2 submitRandom Job Node3 submitRandom PARENT Node0 CHILD Node1 PARENT Node0 CHILD Node2 PARENT Node0 CHILD Node3 Job Node11 submitRandom PARENT Node1, Node2, Node3 CHILD Node11 85

86 Inner DAG › Does this work? › At least one iteration? 86

87 How to iterate › DAGman has simple control structures  (Makes it reliable) › SUBDAGs! › Remember what happens if post fails? 87

88 The Outer Dag › Another Degenerate Dag  (But Useful!) 88 Post Script (with exit value) SubDag (with retry) t

89 This one is easy! › Can you do it yourself? 89

90 The outer DAG file ####### Outer.dag ############# SUBDAG EXTERNAL A inner.dag SCRIPT POST A converge.sh RETRY A 10 #### converge.sh could look like #!/bin/sh echo "Checking convergence" >> converge exit 1 90

91 Let’s run that… › condor_submit_dag outer.dag › Does it work? How can you tell? 91

92 DAGman a bit verbose… $ condor_submit_dag outer.dag File for submitting this DAG to Condor : submit.dag.condor.sub Log of DAGMan debugging messages : submit.dag.dagman.out Log of Condor library output : submit.dag.lib.out Log of Condor library error messages : submit.dag.lib.err Log of the life of condor_dagman itself : submit.dag.dagman.log -no_submit given, not submitting DAG to Condor. You can do this with: "condor_submit submit.dag.condor.sub" File for submitting this DAG to Condor : outer.dag.condor.sub Log of DAGMan debugging messages : outer.dag.dagman.out Log of Condor library output : outer.dag.lib.out Log of Condor library error messages : outer.dag.lib.err Log of the life of condor_dagman itself : outer.dag.dagman.log Submitting job(s). Logging submit event(s). 1 job(s) submitted to cluster

93 Debugging helps › Look in the user log file, “log” › Look in the DAGman debugging log › “foo”.dagman.out 93

94 What does converge.sh need › Note the output files? › How to make them unique? › Add DAG variables to inner dag  And submitRandom file 94

95 The submit file (again) # submitRandom universe = vanilla executable = random.sh output = out log = log queue 95

96 The submit file # submitRandom universe = vanilla executable = random.sh output = out.$(NodeNumber) log = log queue 96

97 The inner DAG file (again) Job Node0 submit_pre Job Node1 submitRandom Job Node2 submitRandom Job Node3 submitRandom PARENT Node0 CHILD Node1 PARENT Node0 CHILD Node2 PARENT Node0 CHILD Node3 Job Node11 submit_post PARENT Node1 CHILD Node11 PARENT Node2 CHILD Node11 PARENT Node3 CHILD Node11 97

98 The inner DAG file (again) Job Node0 submit_pre Job Node1 submitRandom Job Node2 submitRandom Job Node3 submitRandom … VARS Node1 NodeNumber=“1” VARS Node2 NodeNumber=“2” VARS Node3 NodeNumber=“3” … 98

99 Then converge.sh sees: $ ls out.* out.1 out.10 out.2 out.3 out.4 out.5 out.6 out.7 out.8 out.9 $ › And can act accordingly… 99

100 Thank you! Check us out on the Web: