By far the most commonly used data from a run are its artifacts. These are HTTP entities containing output data from the task execution.

Unlike most API methods which return a JSON body, requesting an artifact from the Queue service returns the artifact itself, possibly via one or more HTTP redirects. This means that -- at least for public artifacts which require no authentication -- any sufficiently robust HTTP client can download an artifact directly.

NOTE: Not all clients are "sufficiently robust"! The artifact interface makes heavy use of redirects, and artifacts may make use of other web-standard features such as content encoding. Like any distributed system, requests may fail, too, and a robust client should retry. Out of the box, curl and wget do not handle most of these cases.

Taskcluster's Queue service supports a number of artifact types, including several cloud data-storage back-ends as well as two special types: errors and references. Error artifacts will always return an HTTP 403 (Forbidden), with message and details supplied by the task. Reference artifacts return a 303 (See Other) redirecting the client to another URL.

Public and Private Artifacts

While it is possible to create artifacts which require authorization to download, most artifacts are public. These are easily identified by the prefix public/ in the artifact name. All other artifacts are private, and authorization will be required to read them.

That authorization is by artifact name, not by task. So a particular user with access to artifacts named projects/mrsfields/ can access such an artifact on any task - whether that task is related to the mrsfields project or not.

Log Artifacts

By convention, workers record the output of a task -- a command's output to stdout and stderr, for example -- in an artifact named public/logs/live.log. The details of this artifact, such as its storage type and encoding, can differ from worker to worker. You should not assume anything about it, as the implementation may change dynamically to optimize performance.

While a task is running, some workers can relay data to the client as it is produced, in a single HTTP request. Clients can parse and display this data as it arrives to present a "live log" to the user.