Fuchsia has several observability systems that help you perform diagnostics on a component, such as:
Logs
Logs are time-ordered streams of diagnostic data emitted by Fuchsia programs. They most commonly take the form of low frequency human-oriented text strings describing changes to state in a subsystem.
Contents
A log contains several pieces of metadata that is mostly self-reported by each component that generates a specific log. At a minimum, a log message contains a timestamp and string contents.
Additionally, if a log message is written to the LogSink
protocol using the
supported libraries it includes a severity, a PID, a TID, a count of prior
dropped logs, and a list of string tags. Not all of the metadata is required,
but the supported libraries ensure that every field has a value when writing
records to the socket obtained through LogSink.
Archivist inserts additional metadata to the data that a component writes to a socket. For more information about the fields that Archivist provides for a Fuchsia device log, see Log records.
Storage
Currently all log stores are rotated on a first-in-first-out (FIFO) basis, with newer messages overwriting older ones. Messages from any component can roll out messages from any other component.
Logs in Fuchsia can be stored in the following types of memory:
Volatile
There are two in-memory stores for logs on a Fuchsia device:
- The
klog
or debuglog, which is a 128kb buffer in the kernel. - The
syslog
which is a configurable buffer in the Archivist.
As you work with logs, it is important to understand how Fuchsia writes logs, so that you can better locate where a log may be stored as it explains which of these buffers is the intended point of transit for your message.
Persistent
The feedback data component maintains a persistent_disk_store of messages from the previous boot of the system.
You can see these messages when targeting your Fuchsia device with
ffx target snapshot
.
Work with Logs
To begin working with logging, you may want to:
Inspect
Inspect provides components with the ability to expose structured, hierarchical information about their state at runtime. Components host a mapped Virtual Memory Object (VMO) using the Inspect Format to expose an Inspect Hierarchy containing this internal state. Note: For more information about the Inspect VMO file format, see Inspect VMO file format.
Work with Inspect
To begin working with Inspect, you may want to:
Archivist
Archivist ingests data from Inspect and Fuchsia logs, making them available over
the fuchsia.diagnostics/ArchiveAccessor
protocol.
The Archivist ingests lifecycle events from the Component
Framework through fuchsia.component.EventStream
:
- Capability requested: The Archivist receives
Capability requested
events for connections tofuchsia.logger.LogSink
andfuchsia.inspect.InspectSink
which allows it to attribute Inspect and logs.
CapabilityRequested
event
CapabilityRequested
event is the mechanisms that provide logs and Inspect
with attribution metadata such as the moniker and component URL.
Archivist-manifest expose
s fuchsia.logger.LogSink
just like any other
protocol capability, but it also use
s an event from the framework, binding it
to a protocol in its namespace:
To better understand how this works, take a look at the Archivist's CML file:
// Copyright 2022 The Fuchsia Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
// This shard is meant to contain stuff that is meant to be shared across all flavors of the
// archivist.
{
include: [ "//src/diagnostics/archivist/meta/config.shard.cml" ],
program: {
runner: "elf",
lifecycle: { stop_event: "notify" },
},
capabilities: [
{
dictionary: "diagnostics-accessors",
path: "/svc/fuchsia.component.sandbox.DictionaryRouter",
},
{
protocol: [
"fuchsia.diagnostics.LogSettings",
"fuchsia.diagnostics.LogStream",
"fuchsia.inspect.InspectSink",
"fuchsia.logger.Log",
"fuchsia.logger.LogSink",
],
},
],
use: [
{
event_stream: [ "capability_requested" ],
from: "parent",
path: "/events/log_sink_requested_event_stream",
filter: { name: "fuchsia.logger.LogSink" },
},
{
event_stream: [ "capability_requested" ],
from: "parent",
path: "/events/inspect_sink_requested_event_stream",
filter: { name: "fuchsia.inspect.InspectSink" },
},
{
protocol: [ "fuchsia.tracing.provider.Registry" ],
availability: "optional",
},
{
protocol: [ "fuchsia.inspect.InspectSink" ],
from: "self",
},
{
protocol: [ "fuchsia.component.sandbox.CapabilityStore" ],
from: "framework",
},
],
expose: [
{
protocol: [
"fuchsia.diagnostics.LogSettings",
"fuchsia.diagnostics.LogStream",
"fuchsia.inspect.InspectSink",
"fuchsia.logger.Log",
"fuchsia.logger.LogSink",
],
from: "self",
},
{
dictionary: "diagnostics-accessors",
from: "self",
},
],
}
This causes Component Manager to redirect incoming requests from the default
fuchsia.io
namespace protocol to the
fuchsia.component.EventStream
protocol. Archivist receives
events on this protocol, retrieving attribution metadata from the EventHeader
sent by Component Manager along with the LogSink
channel's handle. The moniker
included in the descriptor is constructed in the event header.
Configuring a capability_requested
event for LogSink
does not affect
capability routing itself, only the delivery of the channel to the Archivist as
an Event instead of as an Open()
. This means that the CML for passing the
attributed LogSink
remains the same for the rest of the component topology.
For more information, see Life of a protocol open and the events documentation.
Attributing LogSink connections
When a Fuchsia component wants to write logs, it must obtain a connection to a
fuchsia.logger.LogSink
in its incoming directory,
typically provided by an instance of the Archivist.
Typical Fuchsia service connections are anonymous such that the server and
client have no identifying information about each other. The client only sees
the service in its namespace, e.g. /svc/fuchsia.logger.LogSink
, and the
server sees an anonymous Open()
request to its incoming namespace.
Attributed logging provides trustworthy attribution metadata to allow for
better monitoring, storage, querying, and presentation of logs. This attribution
metadata identifies the source of an incoming LogSink
connection, which is
accomplish through the CapabilityRequested
event.