Documentation
¶
Overview ¶
Package workflows hosts the deterministic go-workflows workflow code for the durable shadow pilot (Phase 3). Workflow code must stay deterministic: no time.Now, no randomness, no I/O, no goroutines, no map-iteration-order leaks. All side effects run inside activities.
The shared DTOs (AutomationRunWorkflowInput, ShadowRunTrace) are defined in the activities package - the WriteShadowComparison activity consumes the trace, so workflows imports activities (the one clean direction, no import cycle) and re-exports the names via type aliases.
Index ¶
Constants ¶
This section is empty.
Variables ¶
var ErrChainShadowOnlyRequired = errors.New("durable chain shadow workflow requires shadow_only=true")
var ErrShadowOnlyRequired = errors.New("durable shadow workflow requires shadow_only=true")
ErrShadowOnlyRequired rejects non-shadow execution: Phase 3 may observe only, never execute real runs.
Functions ¶
This section is empty.
Types ¶
type AutomationRunWorkflowInput ¶
type AutomationRunWorkflowInput = activities.AutomationRunWorkflowInput
AutomationRunWorkflowInput is the metadata-only workflow input. See the activities package for the definition.
type ChainShadowTrace ¶
type ChainShadowTrace = activities.ChainShadowTrace
func MiviaWorkflowChainShadowWorkflow ¶
func MiviaWorkflowChainShadowWorkflow(ctx workflow.Context, input ChainShadowWorkflowInput) (ChainShadowTrace, error)
MiviaWorkflowChainShadowWorkflow is the Phase 5 multi-workflow chain shadow. It drives the current services only through activities/ports and records bounded metadata. Stage N+1 cannot compile until stage N reaches done. Comparison always runs last.
type ChainShadowWorkflowInput ¶
type ChainShadowWorkflowInput = activities.ChainShadowWorkflowInput
type ExecutionTrace ¶
type ExecutionTrace = activities.ExecutionTrace
ExecutionTrace is the bounded, safe workflow result. See the activities package for the definition.
func MiviaAutomationRunTestExecutionWorkflow ¶
func MiviaAutomationRunTestExecutionWorkflow(ctx workflow.Context, input TestExecutionWorkflowInput) (ExecutionTrace, error)
MiviaAutomationRunTestExecutionWorkflow is the Phase 4 test-only execution workflow. It is deterministic and follows the same rules as the Phase 3 shadow workflow: no wall clock, no randomness, no I/O in workflow code; all side effects live in activities.
Sequence: load snapshot -> claim run (activity) -> complete attempt (activity) -> observe review queue -> observe verifier state -> write execution comparison.
Step semantics (mirroring Phase 3):
- A step that classifies BLOCKED does not short-circuit: the trace keeps observing so the full picture is captured.
- A step that FAILS (activity error or unsafe metadata) short-circuits the remaining steps, which are appended as skipped.
- WriteExecutionComparison ALWAYS runs last, even after a failure.
- FinalStatus mirrors the most recent snapshot run status ("unknown" when the load failed); FailureCategory is the first non-none category any step produced.
- Synthetic (workflow-built) step results carry a zero CompletedAt: workflow code must not read the wall clock.
- The attempt outcome is NOT pre-validated in workflow code beyond pure ref checks on the run ref: the fail-closed unsafe-ref gate lives in the CompleteRunAttempt activity, BEFORE the completion port is invoked, so the guard is exercised end-to-end.
type ShadowRunTrace ¶
type ShadowRunTrace = activities.ShadowRunTrace
ShadowRunTrace is the bounded, safe workflow result. See the activities package for the definition.
func MiviaAutomationRunShadowWorkflow ¶
func MiviaAutomationRunShadowWorkflow(ctx workflow.Context, input AutomationRunWorkflowInput) (ShadowRunTrace, error)
MiviaAutomationRunShadowWorkflow is the Phase 3 shadow workflow. It is observe-only and deterministic. Sequence: load snapshot -> validate executable -> observe work task (only when a task id is present) -> observe codex attempt -> observe governed closeout -> observe review queue -> observe verifier state -> write shadow comparison.
Step semantics (documented decisions):
- A step that classifies BLOCKED does not short-circuit: shadow mode keeps observing so the trace captures the full picture (the spec allows but does not require skipping after blocked).
- A step that FAILS (activity error or unsafe metadata) short-circuits the remaining observe steps, which are appended as skipped.
- WriteShadowComparison ALWAYS runs last, even after a failure.
- FinalStatus mirrors the snapshot run status ("unknown" when the load failed); FailureCategory is the first non-none category any step set.
- Synthetic (workflow-built) skipped/failed step results carry a zero CompletedAt: workflow code must not read the wall clock.
type StageTraceEntry ¶
type StageTraceEntry = activities.StageTraceEntry
type TestExecutionWorkflowInput ¶
type TestExecutionWorkflowInput = activities.TestExecutionWorkflowInput
TestExecutionWorkflowInput is the metadata-only workflow input. See the activities package for the definition.
Source Files