Skip to content

soichih/node-htcondor

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

94 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

node-htcondor contains various nodejs hooks for htcondor.

To install:

npm install htcondor

Submit jobs

You can submit a condor job and watch for joblog.

var htcondor = require('htcondor');

//create condor submit object
var submit_options = {
    universe: "vanilla",
    executable: "test.sh",
    arguments: "hello",
    notification: "never",

    //transfer_output_files: 'bogus',

    requirements: '(Arch == "INTEL") || (Arch == "X86_64") && (machine == "some_ce")',

    shouldtransferfiles: "yes",
    when_to_transfer_output: "ON_EXIT",
    output: "stdout.txt",
    error: "stderr.txt",
    queue: 1
};

//submit to start the job
htcondor.submit(submit_options).then(function(job) {
    console.log("Submitted");

    //you can dump the condor submit property for your new job
    //console.dir(job.props);

    //job.id contains cluster/proc ids
    //console.dir(job.id);

    //you can *watch* job log
    job.onevent(function(event) {
        switch(event.MyType) {

        //normal status type events (just display content)
        case "SubmitEvent":
        case "ExecuteEvent":
        case "JobImageSizeEvent":
            console.log(event.MyType);
            break;

        //critical events
        case "ShadowExceptionEvent":
            console.log(event.MyType);
            console.dir(event);

            //I now stop the watcher (ends this submission)
            job.unwatch();
            break;

        //job ended normally
        case "JobTerminatedEvent":
            console.log(event.MyType);
            console.dir(event);

            //Do something based on the ReturnValue (resubmit, submit different job, etc..)
            console.log("return value:"+event.ReturnValue);

            //If you submitted queue != 1, then you can look for Event.Proc to get the Process ID (0, 1, etc..)

            //If you are happy, just unwatch
            job.unwatch();
            break;

        default:
            console.log(event.MyType);
            console.log("unknown event type.. ignoring for now");
        }
    });
});

You can do the usual condor stuff.

job.remove(function(err) {
    if(err) throw err;
    console.log("successfully removed job");
});
job.hold(function(err) {
    if(err) throw err;
    console.log("successfully held job");
});
job.release(function(err) {
    if(err) throw err;
    console.log("successfully released job");
});

You can also call via htcondor object

htcondor.remove(job, function(err) {
    if(err) throw err;
    console.log("successfully removed job");
});

Also, if you are modern, you can use "then" .. on most htcondor methods.

htcondor.remove(job).then(function() {
    console.log("successfully removed job");
}).fail(function(err) {
    //handle err
});;

Query Jobs

Wrapper for condor_q

htcondor = require("htcondor");
htcondor.q({id: "59794891.0"},function(err, job) {
    console.log(JSON.stringify(job, null, 4));
});

Outputs job entry..

{
    "MATCH_EXP_JOB_GLIDEIN_Entry_Name": "CMS_T2_US_Caltech_cit",
    "MaxHosts": 1,
    "MemoryUsage": "expression:( ( ResidentSetSize + 1023 ) / 1024 )",
    "MATCH_EXP_JOBGLIDEIN_ResourceName": "CIT_CMS_T2",
    "MATCH_EXP_JOB_GLIDECLIENT_Name": "osg-flock-grid-iu-edu_OSG_gWMSFrontend.main",
    "AccountingGroup": "group_xsedehigh.psiders",
    "User": "[email protected]",
    "NumJobReconnects": 1,
    "OnExitHold": "expression:( ExitBySignal == true ) || ( ExitCode isnt 0 )",
    "MATCH_GLIDEIN_ClusterId": 1998341,
    "CoreSize": 0,
...
}

You can set various condor_q options.. like constraint, or attributes.

htcondor = require("htcondor");
htcondor.q({
    constraint: "JobStatus==5",
    attributes: ["Iwd", "Owner", "JobStatus"]
},
function(err, jobs) {
    console.log(JSON.stringify(jobs, null, 4));
}
);

In above, node-htcondor will "stream" output from htcondor q output - so callbacks will be called as soon as each job info is parsed - instead of waiting until all jobs info is received.

Alternatively, then() will not stream but wait until it receives all job entries in a single array.

htcondor = require("htcondor");
htcondor.q({constraint: "JobStatus==5"}).then(function(jobs) {
    console.log(JSON.stringify(jobs, null, 4));
});

eventlog watcher

This module allows you to subscribe to condor event log (usually at /var/log/condor/EventLog), and receive callbacks so that you can monitor job status or any other attribute changes.

If you are monitoring jobs that you submit, then you can just watch the job log instead(See above). eventlog watcher is to monitor the entire cluster.

Obviously though.. you need to have EventLog enabled on your condor installation. You need to have something like following in your condor config.d if you don't see your EventLog generated already.

[/etc/condor/config.d/90eventlog.config]

EVENT_LOG=$(LOG)/EventLog
EVENT_LOG_JOB_AD_INFORMATION_ATTRS=Owner,CurrentHosts,x509userproxysubject,AccountingGroup,GlobalJobId,QDate,JobStartDate,JobCurrentStartDate,JobFinishedHookDone,MATCH_EXP_JOBGLIDEIN_Site,RemoteHost
EVENT_LOG_MAX_SIZE = 1000000
EVENT_LOG_MAX_ROTATIONS = 5
var eventlog = require('htcondor').eventlog

//you can start watching on your htcondor eventlog
eventlog.watch("/var/log/condor/EventLog");

//and receive events
eventlog.on(function(ads) {
    console.dir(ads);
});

eventlog.on() will be called for each classads posted. ads will look like following.

Why didn't I implement this more like fs.watch()? 2 reasons... You often need to start watch before you know which job id to watch (see submit sample below). Another reason is that, you usually don't have more than 1 EventLog, so there is no point of having multiple watcher watching the same log.

{ _jobid: '49563264.000.000',
  _timestamp: '12/15 19:12:25',
  _updatetime: Sun Dec 15 2013 19:12:25 GMT+0000 (UTC),
  Proc: 0,
  EventTime: '2013-12-15T19:12:25',
  TriggerEventTypeName: 'ULOG_SUBMIT',
  SubmitHost: '<129.79.53.21:9615?sock=8287_a430_1068600>',
  QDate: 1387134745,
  TriggerEventTypeNumber: 0,
  MyType: 'SubmitEvent',
  Owner: 'donkri',
  CurrentHosts: 0,
  GlobalJobId: 'osg-xsede.grid.iu.edu#49563264.0#1387134745',
  Cluster: 49563264,
  AccountingGroup: 'group_xsedelow.donkri',
  Subproc: 0,
  EventTypeNumber: 28,
  CurrentTime: 'time()' }

Call unwatch() to stop watchin on eventlog

eventlog.unwatch()

dump condor config

(This is a prototype feature)

You can dump all condor configs (in key/value dictionary).

htcondor.dumpconfig().then(function(configs) {
    console.dir(configs);
}).catch(function(err) {
    console.log("error occured");
    console.dir(err);
});

Configuring the module

You may optionally configure the module by setting config variable if HTCondor binaries or configuration are located in a non-standard location. Current options are:

CondorLocation
The location of the HTCondor install. The directory that contains htcondor's `bin`, `sbin`... directories. This will be used when issuing commands by prepending the command with the full path.
CondorConfig
Location of the Condor configuration file.

Warning

node-htcondor watches job logsi for all jobs submitted using Tail - which uses inotify kernel hook. Don't submit too many jobs at once. You need to throttle it so that you won't exeed your inotify limit imposed by your OS. (See /proc/sys/fs/inotify/max_user_watches) I might implement such throttling capability built into this module in the future..

#License MIT. Please see License file for more details.

About

nodejs hook for htcondor CLI

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published