Informational Files

Workflow Metrics

For every DAG, a JSON formatted metrics file named <DAG input file>.metrics is created when DAGMan exits. In a workflow with nested DAGs, each nested DAG will create its own metrics file. The metrics file will contain the following information:

  • client: The name of the client workflow software (condor_dagman).

  • version: The version of the client workflow software (condor_dagman).

  • type: The type of data, "metrics".

  • start_time: The start time of the client, in epoch seconds, with millisecond precision.

  • end_time: The end time of the client, in epoch seconds, with millisecond precision.

  • duration: The duration of the client, in seconds, with millisecond precision.

  • exitcode: The condor_dagman exit code.

  • dagman_id: The condor_dagman instances ClusterId value.

  • parent_dagman_id: The ClusterId value of this DAGs parent condor_dagman instance; empty if this DAG is not a SUBDAG.

  • rescue_dag_number: The number of the Rescue DAG being run; 0 if not running a Rescue DAG.

  • jobs: The number of nodes in the DAG input file, not including SUBDAG nodes.

  • jobs_failed: The number of failed nodes in the workflow, not including SUBDAG nodes.

  • jobs_succeeded: The number of successful nodes in the workflow, not including SUBDAG nodes; this includes jobs that succeeded after retries.

  • dag_jobs: The number of SUBDAG nodes in the DAG input file.

  • dag_jobs_failed: The number of SUBDAG nodes that failed.

  • dag_jobs_succeeded: The number of SUBDAG nodes that succeeded.

  • total_jobs: The total number of jobs in the DAG input file.

  • total_jobs_run: The total number of nodes executed in a DAG. It should be equal to jobs_succeeded + jobs_failed + dag_jobs_succeeded + dag_jobs_failed.

  • dag_status: The final DAG_Status of the DAG.

If DAGMAN_REPORT_GRAPH_METRICS is set to True then the additionally following metrics will be recorded:

  • graph_height: The height of the DAG.

  • graph_width: The width of the DAG.

  • graph_num_edges: The number of edges (connections) in the DAG.

  • graph_num_vertices: The number of vertices (nodes) in the DAG.

Current Node Status File

DAGMan has the option to write the DAG and its node statuses to a file periodically. This is intended for a user or script to use for monitoring the DAG. To have DAGMan write the node status file simply use the NODE_STATUS_FILE commands syntax as follows:

NODE_STATUS_FILE filename [minimumUpdateTime] [ALWAYS-UPDATE]

The node status file is a collection of ClassAds in New ClassAd format. There is one ClassAd for the overall status of the DAG, one ClassAd for the status of each node, and one ClassAd with the time at which the node status file was completed as well as the time of the next update.

The status file may be updated once per DAGMAN_USER_LOG_SCAN_INTERVAL in combination with the optional minimumUpdateTime value which defaults to 60 seconds. The status file is also updated a final time when the DAG completes either successfully or not.

Normally the node status file is only updated if the status of some node has changed since the last file update. If provided the optional ALWAYS-UPDATE keyword then DAGMan will always update the status file even if no nodes have changed status.

The following example would result the file my.dag.status that will be rewritten with the current DAG status information at intervals of 30 seconds or more:

NODE_STATUS_FILE my.dag.status 30

Possible DagStatus and NodeStatus attribute values are:

  • 0 (STATUS_NOT_READY): At least one parent has not yet finished or the node is a FINAL node.

  • 1 (STATUS_READY): All parents have finished, but the node is not yet running.

  • 2 (STATUS_PRERUN): The node’s PRE script is running.

  • 3 (STATUS_SUBMITTED): The node’s HTCondor job(s) are in the queue.

  • 4 (STATUS_POSTRUN): The node’s POST script is running.

  • 5 (STATUS_DONE): The node has completed successfully.

  • 6 (STATUS_ERROR): The node has failed.

  • 7 (STATUS_FUTILE): The node will never run because an ancestor node failed.

An ancestor is a node that a another node depends on either directly or indirectly through a chain of PARENT/CHILD relationships. Provided the DAG visualized below, node G’s ancestors are nodes A, B, D, and F.

flowchart LR A & B --> C & D D --> E & F F --> G

DAG Ancestor Tree Visualized


A NODE_STATUS_FILE command inside any splice is ignored, and if multiple DAG files are specified then the first specification takes precedence.

Machine-Readable Event History

DAGMan can produce a machine-readable history of events called the job state log. This log was designed for use by the Pegasus Workflow Management System which operates as a layer on top of DAGMan. The job state log can be used to monitor the state of the DAGMan workflow. The job state log is produced when the JOBSTATE_LOG command is declared with the following syntax:


The job state log is a filtered and easily machine-readable version of the *.dagman.out debug log file. It contains all the node events and some additional meta information. Unlike the node status file, the job state log is appended to. Meaning it contains the entire DAG history rather than just the current snapshot.

There are 5 line types in the job state log. Each line begins with a Unix timestamp in the form of seconds since the Epoch. Fields within each line are separated by a single space character.

  1. DAGMan Start:

    A meta-event identifying the condor_dagman job start. Where DAGJobId is the ClusterId and ProcId of the DAGMan job.

    timestamp INTERNAL *** DAGMAN_STARTED DAGJobID ***
  2. DAGMan Exit:

    A meta-event identifying the condor_dagman job exit. Where ExitCode is the DAGMan jobs exit code.

    timestamp INTERNAL *** DAGMAN_FINISHED ExitCode ***
  3. Recovery Started:

    A meta-event identifying DAGMan has entered recovery mode. While in recovery, node events are only printed if they were not already printed prior to recovery mode start.

    timestamp INTERNAL *** RECOVERY_STARTED ***
  4. Recovery Finish/Failure:

    A meta-event identifying DAGMan recovery mode completion or failure.

    timestamp INTERNAL *** RECOVERY_FINISHED ***
    timestamp INTERNAL *** RECOVERY_FAILURE ***
  5. Node Events:

    A meta-event identifying job and script events of a specified node.

    timestamp NodeName EventName CondorID JobTag - SequenceNumber

    The NodeName is the DAG identifier for the node as specified by the JOB command.

    The EventName is one of the many defined event or meta-events as listed below:










    The CondorId is the node job’s ClusterId and ProcId. Meta-events that take prior to successful job submission will not have an assigned CondorId.

    The JobTag is an externally defined tag to assist any workflow managers built on top of the job state log. JobTag defaults to the dash character (-) when no tag is specified. This is defined by setting the following custom job ad attributes in the job’s submit description:

    +job_tag_name = "+job_tag_value"
    +job_tag_value = "<JobTag>"

    If utilizing Pegasus this can be bypassed by setting:

    +pegasus_site = "<JobTag>"

    The SequenceNumber is a monotonically-increasing number that represents each node run attempt due to retries or if the DAG is rerun from a rescue file.

Below is example contents of a job state log assuming JobTag was set to local:

1292620511 INTERNAL *** DAGMAN_STARTED 4972.0 ***
1292620523 NodeA PRE_SCRIPT_STARTED - local - 1
1292620523 NodeA PRE_SCRIPT_SUCCESS - local - 1
1292620525 NodeA SUBMIT 4973.0 local - 1
1292620525 NodeA EXECUTE 4973.0 local - 1
1292620526 NodeA JOB_TERMINATED 4973.0 local - 1
1292620526 NodeA JOB_SUCCESS 0 local - 1
1292620526 NodeA POST_SCRIPT_STARTED 4973.0 local - 1
1292620531 NodeA POST_SCRIPT_TERMINATED 4973.0 local - 1
1292620531 NodeA POST_SCRIPT_SUCCESS 4973.0 local - 1
1292620535 INTERNAL *** DAGMAN_FINISHED 0 ***


Only one job state log can exist per DAGMan process. If multiple are declared then the first one found will take effect and the remainder will output a warning at parse time.