An asynchronous deployment and operations platform with the aim to better enable simplicity, faster time to delivery, and consistency.

This project is maintained by oshied

Directord Components

Directord is a single application which consists of three parts:


Cluster Messaging

The cluster messaging uses a router format which ensures it has bi-directional communication to and from the nodes. The router allows us to create a low latency mesh which monitors node health and ensures is highly responsive instruction delivery network.


The application flow has been built to best enable asynchronous operations, easily scaling to hundreds of clients without impacting throughput. While the User is expected to interact with the system via CLI, Directord does provide bindings for programable interfaces.


The User CLI provides for cluster management and insight into operations. These functions allow operators to see and manipulate job and node status within the cluster.

Data storage and persistance

Directord has two modes of operation for data-storage and persistence.

If the datastore option is set to memory, the server will spawn a manager thread to facilitate the document store.

Storing information persistently introduces a dependency on an external system and creates latency. While the latency should be minimal, and have nearly no impact on task execution, it is something that needs to be considered when constructing the cluster topology. Inversely, using an external datastore will lower the memory utilization of Directord and can have a profound effect on deployment node requirements; this is especially true at hyper-scale.

The datastore option has three potential drivers, memory, file, or redis.

Datastore Configuration (/etc/directord/config.yaml)
disc file:///var/cache/directord
redis redis://
memory (default) memory

Client caching and persistence

Directord clients make use of a POSIX friendly cache, which allows the server and client interactions make use of stored values and ensure client level idempotency during execution. Cached task values are stored for 72 hours unless otherwise configured differently. Cached ARGS are stored indefinitely but can be EVICTED through client interactions as needed.


Every Directord task is profiled. The execution and the round trip time are stored and made available when inspecting jobs. This builtin profiling allows operators to better understand their deployment workloads.

To further understand deployment characteristics, Directord also provides an analyze function which will allow operators to dig deeper into their data. The job and parent analyze functions will highlight outliers, node discrepancies, failures, and performance for entire orchestrations.

$ sudo /opt/directord/bin/directord orchestrate ~/directord/tests/comparison-orchestration.yaml --target directord-{0..5} --wait

# Note the ID used in this command is the "Parent" UUID for a given orchestration
$ sudo /opt/directord/bin/directord manage --analyze-parent fd9c7387-de10-420f-92fe-2ac0a4716c8d
KEY                       VALUE
------------------------  ------------------------------------
ID                        fd9c7387-de10-420f-92fe-2ac0a4716c8d
ACTUAL_RUNTIME            18.774806261062622
COMBINED_EXECUTION_TIME   61.2486093044281
TOTAL_AVG_EXECUTION_TIME  0.0612486093044281
TOTAL_FAILURES            0
TOTAL_JOBS                1000
TOTAL_SUCCESSES           6000

Total Items: 12