Diagnostics Selectors

Overview

The diagnostics platform within Fuchsia contains multiple services. Each of these services (such as data exfiltration, metric polling, and error-state analysis) share a common need to describe the specific properties of their diagnostics data.

We have created a domain specific language (DSL), called Diagnostics Selectors, which provides the ability to describe diagnostics properties exposed by components. Selectors are designed to act on diagnostics schemas, and are used whenever:

  1. Diagnostics data is encoded in a "data hierarchy” in which named nodes host both child nodes and named diagnostics properties.
  2. Diagnostics hierarchies are attributable to specific components which expose that data, by their monikers.

Diagnostics Selectors have the following high-level syntax:

<component_selector>:<hierarchy_path_selector>:<property_selector>

The three parts of the syntax above serve to progressively index the Fuchsia diagnostics data source:

  • component_selector: Specify the producer of the diagnostics data in terms of the producer's moniker.
  • hierarchy_path_selector: Specifies a path through the data hierarchy exposed (by the producer) to a specific node of interest.
  • property_selector: Specifies specific properties of interest on the node you specified in the hierarchy.

Some tools that are built on top of the diagnostics platform (such as Triage and Detect) need to differentiate between data types with their selectors. These tools extend the selector DSL with an additional syntax specifying data type:

INSPECT|LOG|LIFECYCLE:<component_selector>:<hierarchy_path_selector>:<property_selector>

Each of these parts of the syntax are described in detail below.

As Files

Comments

Diagnostics selectors can be written into cfg files in a flat directory. Within these files, comments can be written using //. For example,

// a comment
core/sessions/foo

core2/session:foo  // inline comment

Component selector

Syntax

The component selector defines a pattern which describes the moniker of one or more components in the component topology. The component selector is a collection of forward-slash (/) delimited strings describing the path from a root component to the component of interest.

The following component topology is used to demonstrate the component selector syntax:

A visual tree representation for the selectors explained below

Given this topology, consider the following component selector:

core/sessions/foo

This component selector unambiguously identifies the component named foo relative to its grandparent core.

Each segment (the sections delimited by forward slashes) of the component selector describes exactly one "level” of the component topology.

Component selectors segments can contain only characters that are accepted in a component moniker. Each component selector segment is an instance name or a collection name with an instance name which has a restricted set of allowed characters and length. The colon : in between a collection name and instance name must be escaped with a backslash (\).

Wildcarding

Component selectors support wildcarding, which will match a single "level" of a component selector. Consider the below component_selector, as applied to the above example topology:

core/other_comp/*

This selector matches all components on the system which are running under the parent named other_comp, which is itself running under a parent core. These are their monikers:

  • core/other_comp/foo
  • core/other_comp/bar

Wildcards can also be used as string-completion regular-expressions for a single "level” of a component selector. Consider the following component selector, relative to the above topology figure:

core/*_comp

This matches all components on the system that are running under a parent called core, and which names end in _runner. These include the following monikers:

  • core/some_comp
  • core/other_comp

Component selectors may end in the recursive wildcard **, which matches all components under the given realm:

core/**

This matches all components on the system running under core or any subrealm of core.

The component selector ** alone matches all components on the system.

Hierarchy path selector

Syntax

The hierarchy path selector defines a pattern which describes a path through a structured data hierarchy, to one or many named nodes. The syntax of this sub-selector is nearly identical to that of the component selector, since they both describe paths through a tree of named nodes. The only difference is the optional tree name filter discussed below.

Consider the following JSON-encoding of a diagnostics data hierarchy. In this case, the hierarchy comes from Inspect.

"root": {
    "reverser_service": {
        "connection-0x0": {
            "request_count": 1,
        },
        connection_validity: {
            "is_valid": true
        },
        "connection_count": 1,
        "connection_validity": "connection_xyz"

    },
    "version": "part1"
}

Given this data hierarchy, consider the following hierarchy path selector:

root/reverser_service/connection-0x0

This hierarchy path selector unambiguously describes a path from the root of the data hierarchy to a specific node within the data hierarchy.

Each segment (the sections delimited by forward slashes) of the selector describes exactly one level, or node, of the data hierarchy. Hierarchy path selector segments may contain any characters, however if a segment needs to contain asterisk (*), forward slashes (/), back slashes (\), whitespace (tabs \t or ), or colons (:) they must be escaped.

One thing to note that is unique to hierarchy path selectors and not component selectors, is the case in which a given node shares both a child and property of the same name. Consider the following selector:

root/reverser_service/connection_validity

This path hierarchy selector describes the path from root to the connection_validity node. It is completely unrelated to the connection_validity property on the reverser_service node, which can be selected using a Property Selector: root/reverser_service:connection_validity.

Wildcarding

Hierarchy path selectors support wildcarding, which will match a single "level" of a component selector. The following example will match all nodes in the data hierarchy which are children of a node reverser_service under root:

root/reverser_service/*

Wildcards can also be used as string-completion regular-expressions for a single "level” of a component selector. The following example will match all nodes under reverse_service that start with connection-.

core/reverser_service/connection-*

In the example above, the only matching node is connection-0x0, but if more connection nodes existed, they’d match as well.

Tree name filters

When a component publishes multiple fucshia.inspect.Tree protocols, the selector syntax supports filtering those trees by the name value in the protocol's metadata.

Suppose you had an Inspect hierarchy that looked like this when you did ffx inspect show core/my_component:

core/my_component:
  metadata:
    name = root
    component_url = fuchsia-boot:///my_component#meta/my_component.cm
    timestamp = 70863435581892
  payload:
    root:
      connections:
        connections_closed = 7
core/my_component:
  metadata:
    name = second_tree
    component_url = fuchsia-boot:///my_component#meta/my_component.cm
    timestamp = 70863435581892
  payload:
    root:
      data:
        values = [0, 1, 2]

If you know that you want the property root/data:values, you can use a tree name filter to avoid the overhead of snapshotting both trees with the following syntax (leaving out the property selector portion for now):

[name=second_tree]root/data

If you don't know which tree you want to select against, or you know you want to select from all the trees, you can use this hierarchy selector:

[...]root

This is equivalent to listing all of the names:

[name=root, name=second_tree]root

Omitting a name filter list is equivalent to the filter [name=root].

If a component doesn't specify a name when publishing Inspect, it will implicitly be root.

There is no character restriction on the values in a name filter list, but * must be escaped. If the name contains values other than [a-zA-Z0-9-_], the name must be quoted. In a quoted name, internal quotes must be escaped.

Name filters are case-sensitive.

Property selector

Syntax

The property selector is the simplest of all sub-selectors. It defines a pattern to match a single string, which is a property name on a diagnostics hierarchy. All properties in diagnostics hierarchies have string names. Omitting the property selector is effectively a hierarchy path selector

Like the previous selector segments, if you wish to match against asterisks (*), forward slashes (/), back slashes (\), whitespace (tabs \t or ), or colons (:) they must be escaped with a backslash (\).

Wildcarding

Wildcards can be used to match entire property strings, or can be used as string-completion globs.

eg: abc will match any string with the exact name "abc".
eg: a\* will match any string with the exact name "a*".
eg: a\\* will match any that starts with exactly "a\".
eg: a* will match any string that starts with "a".
eg: a*b will match any string that starts with a and ends with b.
eg: a*b*c will match any string that starts with a and ends with c, with b in the middle.

Full Selector Examples

The following selector will select data from any echo.cm instance on the system that exists in any realm that itself is under realm1. The data retrieved will be the active_connections property on the node at a/b/c.

realm1/*/echo:a/b/c:active_connections

The following selector will select inspect data from any echo.cm instance on the system that exists in any realm that itself is directly under realm1. The inspect data will be the memory_usage property on the node at a/b/c/d.

realm1/echo:a/b/c/d:*