Setting up Condorflux

Tyson Jones MURPA student at UCSD, February 2016 Under the supervision Frank Wuerthwein, Jeffrey Dost and Edgar Fernandez

The Condorflux system combines Telegraf, InfluxDB? , Grafana and the Condor python bindings to probe Condor, collect usage metrics and display them.


Influx runs on the front-end.

Download and Install

cat <<EOF | sudo tee /etc/yum.repos.d/influxdb.repo
name = InfluxDB Repository - RHEL \$releasever
baseurl =\$releasever/\$basearch/stable
enabled = 1
gpgcheck = 1
gpgkey =
sudo yum install influxdb

Start InfluxDB?

sudo service influxdb start

Test you can access the server by visiting the admin panel at

http://[your domain]:8083/

If unreachable, you may need to...

Configure your firewall

nano /etc/sysconfig/iptables

Add the following rules (these may be publicly accessible since we'll enable InfluxDB? authentication)

# InfluxDB admin panel
-A INPUT -m state --state NEW -m tcp -p tcp --dport 8083 -j ACCEPT
# InfluxDB http GET/POSTs
-A INPUT -m state --state NEW -m tcp -p tcp --dport 8086 -j ACCEPT
# Grafana webpage
-A INPUT -m state --state NEW -m tcp -p tcp --dport 3000 -j ACCEPT


service iptables reload
service iptables start

Create an Admin User

Enter the InfluxDB? shell


Create a new username and password which must be supplied to query any databases through the HTTP interface. Here, we make username: admin with password: correct horse battery staple

> CREATE USER admin WITH PASSWORD 'correcthorsebatterystaple' WITH ALL PRIVILEGES
> exit

Enable Authentication

Edit the default InfluxDB? configuration

nano /etc/influxdb/influxdb.conf

Navigate to the [http] section, and adjust

auth-enabled = true

The default config can be regenerated via

influxd config > /etc/influxdb/influxdb.conf

After enabling, use of the InfluxDB? shell will require authentication.

> auth admin [influx password]

Restart InfluxDB?

sudo service influxdb restart

Test InfluxDB?

Create a database testdb through the HTTP interface, passing your credentials.

http://[your domain]:8086/query?q=CREATE%20DATABASE%20testdb&u=admin&p=[influx password]

If successful, this should yield response


Let's spoof some data we can see in Grafana later. Open

http://[your domain]:8083/

and enter the influx username and password (e.g. admin and correcthorsebatterystaple).

Ensure you're using the testdb database in the top-right corner. Select Write Data and submit

testmes value=1 1455824000000000000
testmes value=3 1455824100000000000
testmes value=2 1455824200000000000
testmes value=4 1455824300000000000

(these nanosecond since epoch timestamps correspond to 18 Feb 2016, 11:33am, GMT-8:00)

We'll see these later, though you can check they were written now by entering the query

SELECT * FROM testmes

and hitting Enter

log located at /var/log/influxdb/influxd.log

Setup Data Retention

See here _____________________________________________________________________________________


Grafana should run on the front end, though it is not necessary that it run on the same machine as Influx

Download and Install

cat <<EOF | sudo tee /etc/yum.repos.d/grafana.repo
sudo yum install grafana

If this fails, install version 2.6.0 directly.

sudo yum -y install

Start Grafana

sudo service grafana-server start

Configure Grafana

nano /etc/grafana/grafana.ini

Navigate to the Server section, uncomment (remove the ;) domain and replace localhost with your domain.

domain = [your domain]

Navigate to the Anonymous Auth section and set

enabled = true
org_name = [your organization name]
org_role = Viewer

[your organization name] should be that which you'll create in the proceeding section.

Navigate to the SMTP / Emailing section and set

enabled = true
host = [your domain]:25
from_address = admin@condorflux.[your domain]

Navigate to the Security section and set

admin_user = admin

Restart Grafana

sudo service grafana-server restart

Create the Admin User


http://[your domain]:3000

and select Sign up, then enter your details with username admin (may require email confirmation).

Create an Organization

A Grafana organization groups users and the dashboards they can view.

  • View the side menubar by pressing the top left button.

  • In the side menubar, click Main Org., select + New Organization and enter your organization's name (e.g. UCSD)

Invite Users

In the side menubar, click your organization, select Users then + Add or inivite , and enter their details.

If a SMTP server is running on port 25, this will send an invitation email. If not, no error will be reported but no email will be sent. You can manually send the email by visiting Pending Invitations, clicking on Details of the invitation and copying the address (e.g. http://[your domain]:300/invite/abcdefg) into an email you deliver yourself.

Test Grafana

  • Click Data Sources from the side menubar, then Add new at the top right.

  • Enter Name: test source Type: InfluxDB 0.9.x Url: http://[your domain]:8086 Access: direct Basic Auth: [unchecked] With Credentials: [unchecked] Database: testdb User: admin Password: [influx password]

and click Test Connection, then Save

  • Click Dashboards in the side menubar, Home in the top left and select + New in the dropdown menu to create a new dashboard.

  • Click the green rectangle, and select Add Panel > Graph

  • Under the Metrics tab, set the bottom-right dropdown box to test source, then press + Query

  • Adjust the query to be FROM testmas SELECT field(value) and the time display (top right corner) to be 18 Feb 2016 11:30am to 18 Feb 2016 11:40am, and the spoof data should be visible.

If you're satisfied with your tests, discard your test graph by navigating away (e.g. clicking on Dashboards in the side menubar) and selecting Ignore in the Unsaved changes! alert.

Discard your test database with the query

http://[your domain]:8086/query?q=DROP%20DATABASE%20testdb&u=admin&p=[your password]


Telegraf runs on each machine of which to monitor the system

Condorflux Daemon

The daemon runs on each GlideInWMS? submit site, or any machine which can query a Condor collector

Place the Daemon

The daemon requires no special priveleges or location to run, though must make and edit files in its working directory.

Run the Daemon


The daemon should report a missing metrics file and a configuration error and exit, though has now created a default configuration file and metrics file.

Configure the Daemon

nano config.json

Update the follow fields

"INFLUX PASSWORD": "[influx password]",
"DATABASE URL": "http://[your domain]:8086",

Note that if keeping this username and password private is important, then keep in mind they're in the URLs opened by the daemon which are printed in debug mode (DEBUG_MODE = True), which may be logged by the CRON job, so make the log private.

You may optionally change the field JOB CONSTRAINT which constrains which jobs are collected from the Condor binaries, in the Condor expression syntax. Note classad string literals must be in quotations, which must be escaped to be valid JSON. For example

"JOB CONSTRAINT": "Owner=?=\"jdost\""

If the daemon is to use a non-local collector (e.g. it is running on a gateway though collects on behalf of other gateways under the same collector), then the COLLECTOR ADDRESS should also be changed. For example


Note also if you intend to group and tag metrics by BATCH_JOB_SITE, which employs the LastRemoteHost of the job, you should fill "NODE RENAMES" with "[regex]": "[name]" pairs. If a job's host matches a regex, it will use the associated name in place of the host name, and aggregate with other metrics of the same name. For example,

    ".*\\.t2\\.ucsd\\.edu": "UCSD"

will result in jobs with LastRemoteHosts of, and all being treated as running on the same BATCH_JOB_SITE of UCSD.

Add Metrics

Please see the proceeding section

Setup a CRON

For example

crontab -e
*/15 * * * * ./

where reads as

cd [path to daemon directory]

Looking Into the Past

By editing the NEXT INITIAL BIN START TIME field in the daemon's cache (cache.json), one can set the daemon to look at arbitrarily old jobs (those which started or ended since that time). NEXT INITIAL BIN START TIME must be a seconds since epoch time-stamp and must be earlier than the current time. The field will be located at the very top or very bottom of cache.json.

Note that doing this may cause metrics to be re-calculated at times which causes conflicts in InfluxDB? data. Make sure to clear all metrics from the database (or just drop the database) before looking into the past.

Creating Custom Metrics

Creating a custom metric involves editing the daemon's auxiliary file (to have the daemon collect and push the metrics to influx) and adding a Grafana graph (to pull metrics from influx and display them).

Daemon Operation

The daemon remembers the last time it run, and only ever considers jobs which are currently running, or started/ended since the daemon last run (this includes jobs which started then ended between daemon runs). The daemon splits the time since the previous and now into configurably sized time bins (the final bin may end before the current time, if the time since the previous run is not even divisible by the bin size), and metrics are calculated at each bin. Custom metrics specify how they're to be calculated for a time bin.

The daemon wraps classads in a Job object with many convenient methods, and provides a Bin object which abstracts the tedium of separating metric calculations between tagged values. These are elaborated upon a few sections down.

Create Metric in Daemon

Custom metrics are specified in (in the same directory as as a class (of an arbitrary, but requiredly unique class name) with attributes db, mes, tags, fields, cache and a non-static method calculate_at_bin(time_bin, jobs).


  • The name of the influxDB in which to store this metric.
  • This can be an existing or an original (will be automatically created) database. If this is a new database, it will need to be manually added to Grafana as a datasource later, and its exact name is required memorising.


  • The prefix of the influxDB measurement name of this metric; this is suffixed with the metrics tags list comma-separated in the full measurement name. This identifier is used when adding this metric to Grafana graphs and is discoverable there.
  • Different daemon instances (i.e. running on different machines) will contribute to the same metric if they share the same mes and tags (though tags must include something site specific, like SUBMIT_SITE, to ensure the daemon's don't overwrite eachother's data; see PITFALLS)


  • A list of Condor classad fields by which to split the metric's aggregation, which become InfluxDB? tags with the job's classad's values. That is, jobs with differing classad values (corresponding to the fields in tags) will be included in separate calculations of the metric.
  • The list may also include special fields "BATCH_JOB_SITE" and "BATCH_SUBMIT_SITE" which aren't actual condor classad fields, but are transformed to correspond to the sites of where the job ran and was submitted in the BATCH system.
  • If a job's classad doesn't include all fields in tags (which aren't a "special field"), it won't be passed to the metric / included in its calculation.


  • A list of any additional Condor classad fields the metric needs for its calculation. This must be declared so that the stripped classads passed to the metric's calculate_at_bin method actually contain the needed fields.
  • If a job doesn't contain the declared fields, it will be excluded from the list passed to calculate_at_bin.


  • A list of any Condor classad fields which need to be cached between daemon executions; these are fields which changes in are sought.


  • A non-static method called by the daemon to calculate the metric at a particular time bin.
  • The time bin is passed as a Bin object. Also passed is a list of all jobs (as Job objects) which contain all fields in the metric's fields and tags in the job's classad ( E.g. if tags = ["Owner"] and fields = ["DiskUsage", "RemoteUserCpu"], then any job known by the daemon which doesn't contain all of "Owner", "DiskUsage" and "RemoteUserCpu" in its classad will be excluded from the jobs passed to calculate_at_bin.

For example,

class ExampleMetric:
    db = "example metric database"
    mes = "num running jobs"
    tags = ["SUBMIT_SITE", "MATCH_EXP_JOB_Site"],
    fields = []

    def calculate_at_bin(self, time_bin, jobs):
        for job in jobs:
            if job.is_running_during(time_bin.start_time, time_bin.end_time):
                time_bin.add_to_sum(1, job.get_values(self.tags))
        return time_bin.get_sum()

might result in a push to influxDB "example metric database" for time "12345" with body

num\ running\ jobs,SUBMIT_SITE=UCSD,MATCH_EXP_JOB_SITE=UCI value=10 12345
num\ running\ jobs,SUBMIT_SITE=UCI,MATCH_EXP_JOB_SITE=UCI value=2 12345
num\ running\ jobs,SUBMIT_SITE=UCSD,MATCH_EXP_JOB_SITE=UCR value=14 12345
num\ running\ jobs,SUBMIT_SITE=UCI,MATCH_EXP_JOB_SITE=UCR value=1 12345

For convenience, when performing the same calculation (e.g. counting idle jobs) in multiple metrics, the operation function can be specified globally and attached to a metric.

For example, could read

def count_idle_jobs(self, time_bin, jobs):
    for job in jobs:
        if job.is_idle_during(time_bin.start_time, time_bin.end_time):
            time_bin.add_to_sum(1, job.get_values(self.tags))
    return time_bin.get_sum()

class IdlePerOwnerAndSubmitMetric:
    db = "GlideInMetrics"
    mes = "idle jobs"
    tags = ["SUBMIT_SITE", "Owner"]
    fields = []
    calculate_at_bin = count_idle_jobs

class TotalIdlePerSubmitMetric:
    db = "GlideInMetrics"
    mes = "idle jobs"
    tags = ["SUBMIT_SITE"]
    fields = []
    calculate_at_bin = count_idle_jobs can be extended with custom functions and use external libraries, but must not define any classes which aren't to be interpreted as metrics.

Metrics are most easily temporarily removed by "commenting" them out as a multiline string.


class Metric:



Conflicting Metrics from Different Daemons

Different daemon instances on different submit sites may overwrite each other's measurement data if not tagged by some unique feature, like SUBMIT_SITE or Owner (when site specific). For example, a metric of the same name counting running jobs tagged by MATCH_EXP_JOB_Site which is running on both the UCSD and UCI submit sites will each have access only to a subset of the total job pool (jobs which were submitted from them), and each under-report how many jobs are running at a particular site. To combat this, tag by SUBMIT_SITE and if non-submit-site-specific data is required, aggregate in Grafana.

Only Recent Job States are Reliably Known

Due to limitations in the Condor classads, only a job's most recent session of being in a particular state (like RUNNING or IDLE) can be known by the daemon and thus any metric. For example, if between consecutive executions of the daemon, a job becomes IDLE then RUNNING then IDLE then RUNNING, only the time span that the job spends in its latter idle and running stages can be retrieved. This is only really relevant when the time between daemon execution's is set to be really large (such as when looking into the past).

Class Names Must Be Unique

A metric's python class name is arbitrary and not used anywhere, though must be unique from the class names of all other metrics in

Add Data Source to Grafana

If a new metric in features an original database name (db), then the new influx database (created when the daemon next runs, though there's no need to wait) must be added to Grafana as a Data Source.

In Grafana (http://[your domain]:3000), go to Data Sources in the side menubar, and select Add new at the top of the window.

Fill out (leaving all checkboxes unchecked)... field | value ------|------ Name | A name for the datasource. This is used only within Grafana when adding measurements to a graph. Type | InfluxDB 0.9.x (as of writing this, the latest InfluxDB? version is v0.10 and works with Grafana 2.6.0) Url | http://[your domain]:8086 Note this is your influxDB site, which isn't necessarily where Grafana is running. Access | direct Database | The database name. This is that specified in the metric as db. User | admin Password | [influx password]

then click Add. Afterwards click Test Connection to verify your influx credentials.

Add Graph to Grafana

select the corresponding database

select the corresponding measurement name

If stack, set to individual

do staircase

INTEGRATE THIS _______________________________
SELECT field(value) mean() GROUP BY time(15m) fill(0)
time() should contain a duration equal to (preferred) or larger than (preferably a multiple of) the duration between daemon runs.
fill(0) is used to replace null values at times (there were no timestamps sent to database within the time interval) with zero. This will give dud zeroes if time() doesn't contains too small a duration (less than duration between daemon runs)
When displaying specifically integer information (not to be interpolated) like the number of jobs running, ensure staircase line in Display Styles is ticked (interpolation should only erroneously occur if grafana's interval grouping is misaligned to the daemons bins, which is uncontrollable. I think Grafana will align if it can).
Editing / Deleting Custom Metrics ====================
Solely changing the daemon's will leave old measurements in the front-end databases, the lack of re-population of which may be confused as actual data. Measurements which are no longer being updated should be removed from the databases by...
> Remove from
> Remove measurements from influxDB > > http://[your domain]:8086/query?q=DROP%20MEASUREMENT%20[measurement-name]&u=admin&p=[influx password] >
> Remove from any grafana graphs
> (Optional) Delete database, datasource, dashboard
Daemon Object Documentation =========================


Topic revision: r2 - 2016/02/26 - 20:35:26 - TysonJones
This site is powered by the TWiki collaboration platformCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback