Inventory Documentation



Inventory is simply a registry of "things" that Hawkular knows about. It contains info about your applications, servers, etc. It also keeps track of their relationships with each other.

As such inventory is a graph where nodes are the "things" or "resources" you monitor and manage in Hawkular and edges are relationships between those resources.

Types of Entities

To help with classification of the resources you want to monitor, inventory understands a couple of other types of entities.


Tenants are units of data separation. There may be no relationships between entities in different tenants.

From the point of view of the REST API user, tenants are invisible, because they are directly deduced from the login credentials (+ optionally a persona) passed in the request.


Environments correspond to the common deployment policies in the enterprise where for example applications are developed in an "development" environment and then are promoted to "test", "staging" and finally to "production".


A resource, as already hinted in the Introduction, is the representation of the "thing" in your infrastructure that is being monitored and managed by Hawkular.

Most importantly, a resource references metrics that are relevant to it. Notice that metrics are standalone entities that can exist even without a resource referencing them though and that there is essentially an M to N mapping between resources and metrics (meaning that a single metric can be referenced by multiple resources).

The idea behind this separation is that metrics represent concrete qualities or measurements being collected while resources are essentially only a logical concept that may or may not correspond to an actual piece of software deployed on some machine.

This is best imagined if you consider the resources as things you yourself can compose on your own. For example you might want to create a resource "Health Indicators" for which you would pick and choose several metrics collected possibly on several machines representing the health of your infrastructure.

To make things slightly more complicated though, metrics CAN also be contained within a resource. This is mainly useful for feeds that want to model the fact that a metric cannot exist without a resource in question. Such metrics still can be incorporated into othe resources but as soon as the containing resource is deleted, the resources that incorporate the metric "lose" it.

Resources can have configuration. A configuration is JSON document that the feed can take and apply in some manner to the target resource that it is managing.

A resource can also have a connection configuration. This describes to the feed how to connect to the managed resource.

The two types of configurations are described by data entities with possible IDs configuration or connectionConfiguration that are contained in the resource entities.


Each resource can have a number of metrics associated with it. As an example, we can monitor the response time on an URL. While the actual data, i.e. the individual measurements of the response time, are being stored in Hawkular Metrics, the definition of the metric is held in inventory, too. This is to keep track of what resources are composed of.

Notice though that resources do not have to "own" their metrics. Rather, an m-to-n relationship can exist between the two, meaning that a single metric can be incorporated into several resources (or none). Please refer to Resource chapter for more detailed discussion of this.

Resource Type

A resource type provides metadata about resources - i.e. what metrics they support what are the operations one can execute on them and with what kind of parameters, etc.

Each resource (i.e. thing being monitored and managed by Hawkular) has to have a resource type.

As with resources and metrics, resource types can reference metric types which define the metadata of the metrics to be referenced by the resources. As with resources and metrics, resource types and metric types are separate standalone entities with m-to-n relationships.

Resource types can also contain 2 kinds of data entities - either configurationSchema or connectionConfigurationSchema. The former is a JSON schema document describing the format of the configuration of the resources of this type, while the latter describes the format of the connection configuration of the resources. See Resource chapter for what those are in more detail.

Metric Type

Metric type defines metadata of a metric. Namely it is its unit and data type (gauge, counter, …​).

Operation Type

Resource types can define types of operations that can be executed on resources of those types. The operation types define the format of the return values as well as the format of the parameters accepted.


A feed is a representation of an "agent" that collects monitoring data and performs management operations. Feeds also report on the metrics and resources (along with their types) it found in the place they are running.

In other words, the feeds are supposed to "own" both their data and metadata. Inventory is clever enough to figure out if 2 feeds declare the same metadata and consider it the same. On the other hand, the 2 feeds are not required to synchronize their data and metadata in any way.

Data Entity

A data entity holds a JSON value. It can be contained in a variety of other entities (resource types, resource, operation types) and the ID of the data entity defines its "purpose" in the containing entity. E.g. operation types can have data entities with ID either returnType or parameterTypes.

Metadata Pack

A metadata pack is a means of grouping resource types and metric types together and ensuring they do not change. If a metadata pack with a certain "identity hash" exists, the users can be sure that a set of resource types and metric types with the exact configuration schemas, units, operation types, etc. are also defined. As such a metadata pack is a means of quickly making sure that other metadata exists in a way the user expects.

Inventory Organization

Inventory Organization
Figure 1. Inventory Organization

For the textually inclined, here’s a description of the above diagram as a tree (frankly, this is now more readable than the above automatically generated diagram):

  • Tenant

  • Environment

    • 1 contains 0..N Resource, Metric

      Resources and metrics can exist at 2 places. Either directly under an environment, which is where custom user-defined resources and metric can live. Resources and metrics can also live under feed entities where they are placed by the connecting feeds when they report the data they have discovered.

    • 0..1 incorporates 0..N Feed

      Environment can express that it is composed of a number of feeds. A feed on the other hand can only be incorporated in at most 1 environment at a time.

  • Feed

    • 1 contains 0..N Resource Type, Metric Type, Resource, Metric

      Feeds represent the "things" that report data and metadata. Feeds can be associated with (at most 1 at a time) environment but they don’t have to. To move a feed from one environment to another, simply remove the "incorporates" association with the first enviroment and create a new one with another environment.

  • Resource Type

    • 1 contains 0..2 Data Entity

      A resource type can contain a data entity with ID configurationSchema defining a JSON schema for configuration of resources that are defined by this resource type or connectionConfigurationSchema which is a JSON schema for connectionConfiguration of the resources (which defines how the feed connects to the resource in question, whatever that may mean for a given resource and feed).

    • 1 contains 0..N Operation Type

      Resource types define the types of operations that should be executable on the resources of the type.

    • 1 defines 0..N Resource

      A resource type defines resources. This means that the users can reasonably assume that the data on the resource correspond to the metadata defined at the resource type (i.e. that the configurations on the resource correspond to the schemas defined at the resource type and that the resource incorporates the metrics with the metric types that are incorporated by the resource type.

      A resource MUST be defined by exactly 1 resource type. It is not possible to "re-declare" a resource type of a resource.
      Conformance of the resource data to the schemas is currently not enforced, so it’s only informational.
    • 1 incorporates 0..N Metric Type

      By incorporating a metric type, the resource type declares that the users can expect the resources of this type to incorporate metrics of the metric type.

      This is currently not enforced, so its only informational.
  • Metric Type

    • 1 defines 0..N Metric

      Similarly to how a resource must have a resource type, each metric must have a single metric type. The metric type defines the unit of the metric as well as default collection interval and other metadata about the metric.

  • Operation Type

    • 1 contains 0..2 Data Entity

      Similarly to how resource type defines the configurationSchema and connectionConfigurationSchema, the operation type defines 2 schemas for a return type (the data entity has ID returnType) and parameter types (the data entity has ID parameterTypes).

  • Resource

    • 0..1 contains 0..N Resource

      A resource can contain other resources. This is a "existential" relationship meaning that if a parent resource is deleted, so are its child resources.

    • 0..M isParentOf 0..N Resource

      In addition to defining the resource hierarchy using the existential containment of resources in each other, there can exist other more free form hierarchies realized using the isParentOf relationship (each contained resource is also "parented" using this relationship automagically). This enables the users to create "alternative hierarchies" across feeds or environments, etc.

      As opposed to contains relationship which forms a strict tree in the Inventory, isParentOf is allowed to create "diamonds", i.e. 1 resource is allowed to have 2 or more parents. It is not allowed to form loops using the isParentOf relationship though.

    • 0..M incorporates 0..N Metric

      A resource can be associated with metrics. This essentially means that there exists a "causal" relationship between the existence of the resource and the metric. I.e. the metric is emmitted because it "tells something about" the resource or the resource logically represents some quality monitored by a metric, etc.

    • 0..1 contains 0..N Metric

      In addition to incorporating a metric to a resource, a resource can also outright contain a metric. As explained in the resource description in the previous chapter, this can be used by the feeds to declare that a metric cannot exist without a resource. I.e. this is a stronger form of the incorporates between a resource and a metric. Each contained resource is automatically incorporated by the resource, too.

  • Metadata Pack

    Metadata packs group and "freeze" metadata so that users can be sure they exists in a state they expect. Metadata packs can associate GLOBAL resource types and metric types only. By incorporating a resource/metric type in a metadata pack it essentially becomes read-only. A membership in a metadata pack ensures that the resource type cannot be modified or deleted. This can be used to ensure the users that a certain resource/metric type is present in the inventory.

Basic Principles

Data is Pushed

All data, including identifiers of the entities, is generated in the clients of inventory.

This means that, generally speaking, an entity cannot be uniquely identified by its, client-generated, ID. To uniquely identify an entity, one has to use its canonical path.

Canonical Paths

A canonical path follows the contains relationships from a tenant down to the entity in question.

The canonical path has a form illustrated by the following example:


The above example is a canonical path to a resource with ID resource-id which is located in environment env-id which is inside a tenant tenant-id.

The type specifiers in the individual path segments can be these:

  • t - tenant

  • e - environment

  • rt - resource type

  • mt - metric type

  • f - feed

  • r - resource

  • m - metric

  • ot - operation type

  • d - data entity

  • mp - metadata pack

Identity Hashing

Inventory is able to establish "identity hashes" of several types of entities (namely resource, metric, resource type, metric type, operation type, data entity and feed) to enable automatic "linking" of identical entities across the tenant. This enables the inventory to tell that 2 resource types in 2 feeds are identical or that 2 feeds report the same resources.

Generically, an identity hash is a Merkle tree-hash composed of the ID of the entity (i.e. NOT its canonical path but only its ID) and the hashes of its child entities (as established by the contains relationship). In addition, data entities include their ID and the contents of their JSON data in their identity hash.

Note that because of the inclusion of the entity’s ID in the hash an identity hash is more restrictive than a "content hash". A content hash would be able to tell if 2 otherwise different resources have the same configurations but an identity hash will tell that the 2 resources have the same ID and the configurations. This is based on an assumption that like resources should be likely detected and therefore should have the same ID.


Inventory is configurable using several means.

  • There are built-in defaults.

  • Configuration can be read from a configuration file

  • Several configuration properties can be overriden using java system properties and environment variables.

Inventory uses a single configuration file even though several different and independent subsystems are configured using it.

First it can be used to override the choice of inventory implementation in case there are more of them on the classpath.


Locating Configuration File

  1. If there is a system property called hawkular-inventory.conf the value is supposed to be a path to file from which the configuration will be loaded.

  2. If such system property is not defined, the system checks for existence of a file called .hawkular-inventory.conf in the home directory of the user running the server.

  3. If no such file exists, the default configuration is used.

Configuration Properties

Table 1. Available Configuration Properties
Property Name Availability Environment Variable Default Value Description

This is the property to be used in the configuration file. Also this is the name of the system property to override the configured value with (if not specified otherwise)

Some properties are only available for certain components inside inventory that might or might not be present during the runtime

This is the name of the environment variable to override the value





The fully qualified class name of the org.hawkular.inventory.api.Inventory interface implementation that is accessible on the runtime classpath.

If this property is not present, the first implementation available using Java service loading mechanism is used.


Inventory implementation inheriting from org.hawkular.inventory.base.BaseInventory (this is true by default)



The base implementation assumes that the backend storage uses some kind of optimistic locking for transaction handling. This property defines the number of retries of transactions if they fail due to locking or concurrent access situations.


Inventory implementation based on Tinkerpop3 API (the default)



The fully qualified class name of an implementation of the org.hawkular.inventory.impl.tinkerpop.spi.GraphProvider interface.

Tinkerpop is an API that is implemented by multiple graph databases. This property can be used to override the default selection mechanism that is to use the first implementation loaded using the Java services mechanism. Hawkular is by default packaged with Sqlg.

sql.jdbc.url (system property hawkular.inventory.sql.jdbc.url)

The connection string to the database to store inventory data to.



The connection string to the database that will be used to store inventory data to.

If the connection string starts with jndi:, the remainder of the connection string is understood to be the object name of a datasource accessible in the JNDI tree of the Hawkular server (e.g. the default jndi:java:/jboss/datasources/HawkularInventoryDS_hsqldb). Otherwise this can be a connection string to any database supported by Sqlg. As of the time of writing, this is Postgresql, H2 and HSQLDB (e.g. jdbc:postgresql://localhost/hawkular). In the case of a "normal" JDBC connection, it will usually be necessary to also provide the username and password. If the connection is obtained from JNDI there is no need to provide the credentials.

sql.jdbc.username (system property hawkular.inventory.sql.jdbc.username)

The user name to use when connecting using a normal JDBC connection.



The user name to use when connecting using a normal JDBC connection.

sql.jdbc.password (system property hawkular.inventory.sql.jdbc.password)

The password to use when connecting using a normal JDBC connection.



The password to use when connecting using a normal JDBC connection.


Comma separated list of addresses (including schema) allowed to interact with the inventory REST API from the browser.



This configuration is shared with Hawkular Metrics and Hawkular Alerts and sets up the addresses allowed to interact with the inventory from the browser in cross-origin manner. By default any address is allowed.


Comma separated additional headers allowed to be passed in CORS requests from browser



Again, this is shared with Hawkular Metrics and Hawkular Alerts and specifies additional headers (apart from the default ones required for inventory to function properly) that can be passed to Hawkular Inventory in CORS requests. This is an advanced configuration that you usually don’t have alter.


© 2016 | Hawkular is released under Apache License v2.0