The LogDog Butler runs on log-producing clients and manages the ingestion of log data from a read-forward stream, formatting and bundling it into log protobufs, and the transmission of those protobufs through the Transport Layer to Collector instances.
The Butler is fairly configurable, and is capable of being used in multiple scenarios via its subcommands:
stream, which uploads the contents of a single log stream.
serve, which runs a Butler stream server instance.
run, which runs another command, instrumenting its environment and standard file descriptors for log streaming.
The Butler, based on its configuration, may configure default streams. For example, when using the
run subcommand, it has the option to connect the bootstrapped process'
STDERR streams to LogDog streams.
However, the Butler can also enable additional stream registration by instantiating a local Stream Server and exposing that to other applications. Those applications can then connect to the Stream Server, perform a minimal handshake, and connect an additional stream to the Butler.
Clients can use the streamclient package to connect to a stream server and create Butler streams. Additional stream client packages are available for other languages:
Stream Servers are created by specifying the
-streamserver-uri parameter to LogDog. Stream server options vary by operating system.
POSIX systems (Linux, Mac) support the following stream servers:
pathis a filesystem path to a named UNIX domain socket to create.
Windows systems support the following stream servers:
nameis a valid Windows named pipe name.
In production, each Butler instance will begin by registering a unique log stream Prefix with its Coordinator instance. The exchange will ensure that only that specific Butler instance may create log streams underneath of that Prefix.
Production streaming can be requested by specifying the
logdog Output option and associated parameters:
$ logdog_butler -output logdog,host=<coordinator-host> ...
The Butler instance must be configured to use a service account that has WRITE permission for the target project.
This will cause the Butler to perform prefix registration during its Output initialization, prior to any bootstrapping or streaming.