Documentation
¶
Overview ¶
Package sentry reports workflow failures to Sentry through github.com/getsentry/sentry-go: a workflow.Reporter that sends one Sentry event per failure it is handed.
r := sentry.New(nil, sentry.WithLabels("tenant"))
client, err := workflow.Use(client, middleware.Report(r, middleware.ReportConfig{}))
middleware.Report (or ReportFailures alone) hands it one Failure per run that ended failed, and with ReportConfig.Attempts one per failed attempt that will be retried; ReportConfig.Discarded adds discarded runs. The Reporter itself sends whatever it is given.
What is sent ¶
Each event carries the error (its message and type chain, as sentry-go's CaptureException builds them), level LevelError for a final failure and LevelWarning for an attempt (WithLevels), the tags workflow.name, workflow.step, workflow.status and workflow.final, and a "workflow" context with the run ID, scheduled ID, attempt and time. The payload is never sent: workflow.Failure does not carry it. Job labels are not sent by default, because they can carry tenant or user ids into a third-party service; WithLabels (chosen keys) or WithAllLabels opts in, as tags named workflow.label.<key>. The default fingerprint groups events by workflow, step and error type (WithFingerprint replaces it), so one recurring failure is one Sentry issue rather than one per run.
Hub ¶
New takes the hub to send through. With a nil hub each Report uses the hub in its ctx (sentry.GetHubFromContext) or else sentry.CurrentHub. Every Report works on a clone of that hub, so concurrent reports never share a scope and the app's own scope is never changed. WithEvent is the escape hatch: it sees each event and the Failure just before sending, and may change or drop the event.
Sending is sentry-go's: its transport batches in the background, so call sentry.Flush (or the hub client's Flush) before the process exits.
Index ¶
Constants ¶
const ( TagWorkflow = "workflow.name" TagStep = "workflow.step" TagStatus = "workflow.status" TagFinal = "workflow.final" TagLabelPrefix = "workflow.label." ContextName = "workflow" )
Tag and context names set on every event. They are fixed so a Sentry search or alert rule written against them keeps working.
Variables ¶
This section is empty.
Functions ¶
func DefaultFingerprint ¶
DefaultFingerprint groups events by workflow, step and the type of the innermost error, so a recurring failure is one issue whatever the run or the error text (which often holds ids).
Types ¶
type Option ¶
type Option func(*Reporter)
Option configures a Reporter.
func WithAllLabels ¶
func WithAllLabels() Option
WithAllLabels forwards every job label as a tag. Prefer WithLabels: a label added later is then sent without anyone deciding it.
func WithEvent ¶
WithEvent runs fn on every event just before it is sent, with the Failure it came from; fn returns the event to send, or nil to drop it. It is the one escape hatch for anything the options do not cover.
func WithFingerprint ¶
WithFingerprint replaces the grouping of events into Sentry issues. The default is DefaultFingerprint.
func WithLabels ¶
WithLabels forwards the job labels with these keys as tags (workflow.label.<key>). Labels are not sent unless chosen, because they can carry tenant or user ids.
func WithLevels ¶
WithLevels sets the event level of a final failure and of an attempt failure (LevelError and LevelWarning by default).
type Reporter ¶
type Reporter struct {
// contains filtered or unexported fields
}
Reporter is a workflow.Reporter that sends each Failure to Sentry as one event. Build it with New; it is safe for concurrent use.