ic_backup/ops/persistence/execution_workflow/
mod.rs

1//! Fixed immutable stage admission using the existing layout, plan and settlement owners.
2
3mod ancestry;
4mod progress;
5mod stage;
6use ancestry::{
7    acquire_stage, read_admitted_binding, read_binding, read_current_binding, stage_path,
8    validate_predecessors,
9};
10pub use progress::{ExecutionWorkflowProgressError, read_execution_workflow_progress};
11pub use stage::ExecutionStageGuard;
12
13use super::json::read_record;
14use super::{
15    AttemptJournalError, AttemptJournalGuard, BackupLayoutGuard, ExecutionProgressPersistenceError,
16    ExecutionSettlementCheckpointError, ExecutionSettlementPersistenceError, JournalLock,
17    JournalLockError, OperationPlanPersistenceError, PersistenceError, read_execution_progress,
18    read_execution_settlement, read_operation_plan, reconcile_execution_settlement,
19    replay_execution_settlement,
20};
21use crate::model::{
22    artifacts::ArtifactChecksumRecord,
23    execution_workflow::{
24        ExecutionStageBindingRecord, ExecutionStagePredecessorRecord, ExecutionWorkflowError,
25        ExecutionWorkflowRecord, MAX_EXECUTION_STAGE_BYTES, MAX_EXECUTION_WORKFLOW_BYTES,
26    },
27    operation_plan::{OperationPlanError, OperationPlanRecord},
28};
29use crate::policy::execution_progress::ExecutionProgressView;
30use std::{
31    collections::{BTreeMap, BTreeSet, btree_map::Entry},
32    fs,
33    path::PathBuf,
34};
35use thiserror::Error;
36
37const WORKFLOW_FILE: &str = "execution-workflow.json";
38const BINDING_FILE: &str = "stage-binding.json";
39
40/// Exact local identities of every original stage's retained checkpoint.
41///
42/// This is a sequential local projection, not an immutable product completion
43/// receipt. It authenticates no learned evidence, remote effect, application
44/// safety or command/byte custody and grants no terminal or reference release.
45#[derive(Clone, Debug, Eq, PartialEq)]
46pub struct ExecutionWorkflowSettlementView {
47    workflow: ArtifactChecksumRecord,
48    stages: Vec<ExecutionStageSettlementView>,
49}
50impl ExecutionWorkflowSettlementView {
51    /// Read the exact original workflow declaration identity.
52    #[must_use]
53    pub const fn workflow(&self) -> &ArtifactChecksumRecord {
54        &self.workflow
55    }
56    /// Read one row per original stage in canonical allocation order.
57    #[must_use]
58    pub fn stages(&self) -> &[ExecutionStageSettlementView] {
59        &self.stages
60    }
61}
62
63/// One original stage's bound plan and complete chronological local checkpoint.
64#[derive(Clone, Debug, Eq, PartialEq)]
65pub struct ExecutionStageSettlementView {
66    stage_sequence: u64,
67    binding: ArtifactChecksumRecord,
68    settlement: ArtifactChecksumRecord,
69}
70impl ExecutionStageSettlementView {
71    /// Read the original opaque stage identity.
72    #[must_use]
73    pub const fn stage_sequence(&self) -> u64 {
74        self.stage_sequence
75    }
76    /// Read the exact retained binding identity.
77    #[must_use]
78    pub const fn binding(&self) -> &ArtifactChecksumRecord {
79        &self.binding
80    }
81    /// Read the complete original journal-history checkpoint identity.
82    #[must_use]
83    pub const fn settlement(&self) -> &ArtifactChecksumRecord {
84        &self.settlement
85    }
86}
87
88/// Durably create the full original workflow allocation without replacement.
89/// # Errors
90/// Rejects occupied/unsafe paths, bounds, ownership and publication failures.
91pub fn create_execution_workflow(
92    layout: &BackupLayoutGuard,
93    record: &ExecutionWorkflowRecord,
94) -> Result<(), ExecutionWorkflowPersistenceError> {
95    layout.check_root()?;
96    let bytes = super::json::encode_json(record, MAX_EXECUTION_WORKFLOW_BYTES)?;
97    super::immutable_record::create_encoded_record(layout, WORKFLOW_FILE, &bytes)
98}
99/// Read exact original workflow allocation; missing evidence never supplies a new budget.
100/// # Errors
101/// Rejects missing/changed/unsafe/oversized records and ownership failures.
102pub fn read_execution_workflow(
103    layout: &BackupLayoutGuard,
104    expected: &ArtifactChecksumRecord,
105) -> Result<ExecutionWorkflowRecord, ExecutionWorkflowPersistenceError> {
106    super::immutable_record::read_with(
107        layout,
108        WORKFLOW_FILE,
109        MAX_EXECUTION_WORKFLOW_BYTES,
110        |record: &ExecutionWorkflowRecord| {
111            if &record.digest() != expected {
112                return Err(ExecutionWorkflowPersistenceError::DigestMismatch);
113            }
114            Ok(())
115        },
116    )
117}
118
119/// Read which allocated stage paths are already occupied under the held layout.
120///
121/// This is a namespace preflight for a retained run, not evidence that an absent
122/// stage has never dispatched. An occupied partial stage remains occupied and is
123/// never repaired. Derived symlinks and non-directories refuse admission.
124/// # Errors
125/// Rejects changed workflow/layout, unsafe stage entries and IO failures.
126pub(crate) fn read_execution_stage_occupancy(
127    layout: &BackupLayoutGuard,
128    expected: &ArtifactChecksumRecord,
129) -> Result<BTreeSet<u64>, ExecutionWorkflowPersistenceError> {
130    let workflow = read_execution_workflow(layout, expected)?;
131    let mut occupied = BTreeSet::new();
132    for stage in workflow.stages() {
133        let sequence = stage.operation_sequence();
134        match fs::symlink_metadata(stage_path(layout, sequence)) {
135            Ok(metadata) if metadata.is_dir() => {
136                occupied.insert(sequence);
137            }
138            Ok(_) => return Err(ExecutionWorkflowPersistenceError::UnsafeStage),
139            Err(error) if error.kind() == std::io::ErrorKind::NotFound => {}
140            Err(error) => return Err(PersistenceError::from(error).into()),
141        }
142    }
143    read_execution_workflow(layout, expected)?;
144    Ok(occupied)
145}
146
147/// Replay every allocated stage's exact retained binding and all-Applied checkpoint.
148///
149/// The original workflow determines the complete stage set. Every fixed stage must
150/// exist with a valid child plan, immutable binding, complete original journal set
151/// and exact published settlement. Every binding's predecessor identities must
152/// match this complete original set. Stage locks are held one at a time under
153/// the caller's workflow layout guard. No absent stage means zero spending, and no
154/// occupied checkpoint is repaired or republished. The resulting digest rows
155/// are local evidence only; integration-owned authenticity, complete transfer,
156/// application safety, command quiescence and terminal/release proof remain separate.
157/// # Errors
158/// Rejects missing, unsafe, changed, held or unsettled original evidence without effects.
159pub fn read_complete_execution_workflow(
160    layout: &BackupLayoutGuard,
161    expected: &ArtifactChecksumRecord,
162) -> Result<ExecutionWorkflowSettlementView, ExecutionStageCheckpointError> {
163    read_complete_with(layout, expected, || {
164        read_execution_workflow(layout, expected)
165    })
166}
167
168/// Replay complete original settlement while retaining the exact workflow journal lock.
169///
170/// Terminal-evidence providers may use this instead of reacquiring their held journal
171/// lock. The guard must protect this layout's original `execution-workflow.json`.
172/// Both opening and closing reads retain the ordinary bounds, strict decoding and
173/// expected original digest checks; child locks and checkpoint admission are unchanged.
174/// This supplies local settlement only, never application or reference-release authority.
175/// # Errors
176/// Rejects a guard for another journal, changed originals, or incomplete settlement.
177pub fn read_complete_execution_workflow_locked(
178    layout: &BackupLayoutGuard,
179    journal: &JournalLock,
180    expected: &ArtifactChecksumRecord,
181) -> Result<ExecutionWorkflowSettlementView, ExecutionStageCheckpointError> {
182    if journal.journal() != layout.root().join(WORKFLOW_FILE) {
183        return Err(ExecutionWorkflowPersistenceError::DigestMismatch.into());
184    }
185    read_complete_with(layout, expected, || {
186        layout.check_root()?;
187        let workflow: ExecutionWorkflowRecord =
188            read_record(journal.journal(), MAX_EXECUTION_WORKFLOW_BYTES)?;
189        if &workflow.digest() != expected {
190            return Err(ExecutionWorkflowPersistenceError::DigestMismatch);
191        }
192        Ok(workflow)
193    })
194}
195
196fn read_complete_with(
197    layout: &BackupLayoutGuard,
198    expected: &ArtifactChecksumRecord,
199    read_original: impl Fn() -> Result<ExecutionWorkflowRecord, ExecutionWorkflowPersistenceError>,
200) -> Result<ExecutionWorkflowSettlementView, ExecutionStageCheckpointError> {
201    let workflow = read_original()?;
202    let mut stages = Vec::with_capacity(workflow.stages().len());
203    for allocated in workflow.stages() {
204        let sequence = allocated.operation_sequence();
205        let stage = acquire_stage(layout, sequence)?;
206        let (binding, plan) = read_current_binding(&stage, &workflow, sequence)?;
207        let settlement = replay_execution_settlement(&stage, &plan.digest())?;
208        read_binding(&stage, &workflow, sequence, &binding.digest())?;
209        let view = ExecutionStageSettlementView {
210            stage_sequence: sequence,
211            binding: binding.digest(),
212            settlement: settlement.digest(),
213        };
214        stages.push((binding, view));
215    }
216    for (binding, _) in &stages {
217        for predecessor in binding.predecessors() {
218            // The original catalog supplies canonical sequence order.
219            let index = stages
220                .binary_search_by_key(&predecessor.stage_sequence(), |(_, view)| {
221                    view.stage_sequence()
222                })
223                .map_err(|_| ExecutionWorkflowPersistenceError::DigestMismatch)?;
224            let actual = &stages[index].1;
225            if predecessor.binding() != actual.binding()
226                || predecessor.settlement() != actual.settlement()
227            {
228                return Err(ExecutionWorkflowPersistenceError::DigestMismatch.into());
229            }
230        }
231    }
232    read_original()?;
233    Ok(ExecutionWorkflowSettlementView {
234        workflow: expected.clone(),
235        stages: stages.into_iter().map(|(_, view)| view).collect(),
236    })
237}
238
239/// Stage-bound checkpoint publication/replay refusal; originals remain retained.
240#[derive(Debug, Error)]
241pub enum ExecutionStageCheckpointError {
242    /// Original workflow/stage or complete ancestor history cannot be admitted.
243    #[error(transparent)]
244    Stage(#[from] ExecutionWorkflowPersistenceError),
245    /// Complete original journal checkpoint derivation/publication failed.
246    #[error(transparent)]
247    Settlement(#[from] ExecutionSettlementCheckpointError),
248}
249
250/// Failed create-only preparation; retained records never grant repair or extra spending.
251#[derive(Debug, Error)]
252pub enum ExecutionStagePreparationError {
253    /// Original stage records, predecessor settlement or layout were rejected.
254    #[error(transparent)]
255    Stage(#[from] ExecutionWorkflowPersistenceError),
256    /// Complete original child authority derivation failed before allocation.
257    #[error(transparent)]
258    Authority(#[from] OperationPlanError),
259    /// Existing attempt-journal locking or durable create-only publication failed.
260    #[error(transparent)]
261    Journal(#[from] AttemptJournalError),
262}
263
264/// Complete retained stage-resume refusal; original evidence and spending remain unchanged.
265#[derive(Debug, Error)]
266pub enum ExecutionStageResumeError {
267    /// Original stage, ancestor history or layout cannot be admitted.
268    #[error(transparent)]
269    Stage(#[from] ExecutionWorkflowPersistenceError),
270    /// Complete original child journals, accounting or causality cannot be admitted.
271    #[error(transparent)]
272    Progress(#[from] ExecutionProgressPersistenceError),
273}
274/// Typed immutable workflow/stage admission failure; no recovery evidence is removed.
275#[derive(Debug, Error)]
276pub enum ExecutionWorkflowPersistenceError {
277    /// Workflow/stage identity differs from the retained expected digest.
278    #[error("execution workflow/stage digest mismatch")]
279    DigestMismatch,
280    /// A derived stage directory is a symlink or another unsafe entry.
281    #[error("unsafe execution stage directory")]
282    UnsafeStage,
283    /// Original model allocation or learned binding was rejected.
284    #[error(transparent)]
285    Model(#[from] ExecutionWorkflowError),
286    /// Existing immutable operation-plan admission failed.
287    #[error(transparent)]
288    Plan(#[from] OperationPlanPersistenceError),
289    /// Existing complete chronological Applied settlement admission failed.
290    #[error(transparent)]
291    Settlement(#[from] ExecutionSettlementPersistenceError),
292    /// Cooperating layout or journal exclusion failed.
293    #[error(transparent)]
294    Lock(#[from] JournalLockError),
295    /// Existing bounded JSON/durable filesystem owner failed.
296    #[error(transparent)]
297    Persistence(#[from] PersistenceError),
298}
299
300#[cfg(all(test, unix))]
301mod tests;
ic_backup/ops/persistence/execution_workflow/mod.rs

ic_backup/ops/persistence/execution_workflow/
mod.rs
