fix: integrate safe Workdir removal recovery
Merge approved T-730 source b1dc8f9f into develop. Resolve the test-registration conflict by preserving both Drive and Workdir-removal tests. Integration root check, Runtime/Server/Worker suites, real Web Drive adapter, generated Workdir contract, Web check/unit/components/build/browser, fmt and diff checks passed.
This commit is contained in:
@@ -11927,10 +11927,6 @@ fn cleanup_workdir_cleanliness_from_runtime(value: Option<&str>) -> CleanupWorkd
|
||||
.unwrap_or(CleanupWorkdirCleanliness::Unknown)
|
||||
}
|
||||
|
||||
fn cleanup_workdir_cleanliness_is_clean(value: CleanupWorkdirCleanliness) -> bool {
|
||||
matches!(value, CleanupWorkdirCleanliness::Clean)
|
||||
}
|
||||
|
||||
#[derive(Debug, Deserialize)]
|
||||
struct LogQuery {
|
||||
limit: Option<usize>,
|
||||
@@ -17787,9 +17783,15 @@ async fn scoped_attach_current_worker_workdir(
|
||||
code: "working_directory_not_found".to_string(),
|
||||
message: format!("unknown Workdir `{workdir_id}`"),
|
||||
})?;
|
||||
if workdir.materialization_status != "present" {
|
||||
// A failed removal must not prevent its owner from attaching a still usable
|
||||
// repository checkout to resolve current blockers. Runtime binding validates
|
||||
// the actual checkout; durable pending-removal reservations still fence this
|
||||
// mutation. Other sources and missing/corrupted materializations stay denied.
|
||||
let repository_retry = workdir.materialization_status == "pending"
|
||||
&& matches!(workdir.source, WorkdirRegistrySource::Repository { .. });
|
||||
if workdir.materialization_status != "present" && !repository_retry {
|
||||
return Err(Error::WorkdirAttachmentConflict(format!(
|
||||
"Workdir `{workdir_id}` is not active ({})",
|
||||
"Workdir `{workdir_id}` is not available for attachment ({})",
|
||||
workdir.materialization_status
|
||||
))
|
||||
.into());
|
||||
@@ -24527,18 +24529,88 @@ fn runtime_reports_workdir_not_found(result: &crate::hosts::RuntimeWorkingDirect
|
||||
fn classify_workdir_provider_error(error: &RuntimeRegistryError) -> (&'static str, bool) {
|
||||
match error {
|
||||
RuntimeRegistryError::UnknownRuntime(_) => ("runtime_unavailable", true),
|
||||
RuntimeRegistryError::RuntimeOperationFailed { code, .. }
|
||||
if code == "working_directory_unsupported" =>
|
||||
{
|
||||
("unsupported_target", false)
|
||||
RuntimeRegistryError::RuntimeOperationFailed { code, .. } => {
|
||||
let category =
|
||||
workdir_cleanup_diagnostic_category(code).unwrap_or("provider_unavailable");
|
||||
(category, category != "unsupported_target")
|
||||
}
|
||||
RuntimeRegistryError::RuntimeOperationFailed { .. } => ("provider_unavailable", true),
|
||||
RuntimeRegistryError::InvalidIdentifier { .. }
|
||||
| RuntimeRegistryError::UnknownHost(_)
|
||||
| RuntimeRegistryError::UnknownWorker { .. } => ("authority_invalid", false),
|
||||
}
|
||||
}
|
||||
|
||||
// Only bounded classifications cross the public removal boundary. Provider
|
||||
// messages may contain OS paths or credentials and are never persisted here.
|
||||
fn workdir_cleanup_failure_category(
|
||||
result: &crate::hosts::RuntimeWorkingDirectoryResult,
|
||||
) -> &'static str {
|
||||
result
|
||||
.diagnostics
|
||||
.iter()
|
||||
.find_map(|diagnostic| workdir_cleanup_diagnostic_category(&diagnostic.code))
|
||||
.unwrap_or("provider_cleanup_failed")
|
||||
}
|
||||
|
||||
fn workdir_cleanup_diagnostic_category(code: &str) -> Option<&'static str> {
|
||||
Some(match code {
|
||||
"working_directory_unsupported" => "unsupported_target",
|
||||
"working_directory_cleanup_mount_present" => "mount_present",
|
||||
"working_directory_cleanup_mount_check_unavailable" => "mount_check_unavailable",
|
||||
"working_directory_cleanup_permission_denied" => "permission_denied",
|
||||
"working_directory_cleanup_resource_busy" => "resource_busy",
|
||||
"working_directory_cleanup_storage_unavailable" => "storage_unavailable",
|
||||
"working_directory_cleanup_ownership_unknown" => "ownership_unknown",
|
||||
"working_directory_cleanup_changes_present" => "changes_present",
|
||||
"working_directory_cleanup_changes_unknown" => "changes_unknown",
|
||||
"working_directory_cleanup_target_invalid"
|
||||
| "working_directory_cleanup_escape_rejected" => "ownership_unknown",
|
||||
_ => return None,
|
||||
})
|
||||
}
|
||||
|
||||
fn workdir_removal_failure_guidance(category: &str) -> &'static str {
|
||||
match category {
|
||||
"mount_present" => {
|
||||
"A mount remains inside the Workdir. Confirm its owner and usage, have that owner safely release it, then retry deletion. Yoi does not unmount external resources."
|
||||
}
|
||||
"mount_check_unavailable" => {
|
||||
"Runtime cannot inspect mount boundaries. Restore mount inspection before retrying deletion."
|
||||
}
|
||||
"permission_denied" => {
|
||||
"Runtime lacks permission to remove the Workdir. Have its owner resolve access safely, then retry; do not force-clean or change shared resource permissions."
|
||||
}
|
||||
"resource_busy" => {
|
||||
"A resource is busy. Have its owner stop or release it safely, then retry deletion."
|
||||
}
|
||||
"storage_unavailable" => {
|
||||
"Runtime storage is unavailable. Restore storage access, then retry deletion."
|
||||
}
|
||||
"ownership_unknown" | "provider_identity_changed" | "authority_changed" => {
|
||||
"Workdir identity or ownership could not be confirmed. Verify the target with its owner before retrying."
|
||||
}
|
||||
"changes_present" | "dirty_or_unknown" => {
|
||||
"Current Workdir changes are protected. Preserve or resolve them before retrying deletion."
|
||||
}
|
||||
"changes_unknown" => {
|
||||
"Runtime cannot verify remaining Workdir changes. Restore repository inspection or preserve remaining files before retrying."
|
||||
}
|
||||
"blocked_by_live_authority" => {
|
||||
"Workdir has a current attachment, reservation or retention hold. Release the owning authority before retrying deletion."
|
||||
}
|
||||
"runtime_unavailable" | "provider_unavailable" => {
|
||||
"Runtime is unavailable. Reconnect it, then retry deletion; the Workdir registry is retained."
|
||||
}
|
||||
"provider_observation_unknown" | "provider_cleanup_outcome_unknown" => {
|
||||
"Runtime has not confirmed deletion. Retry to re-observe the target; the Workdir registry is retained."
|
||||
}
|
||||
"unsupported_target" => "This provider does not support Workdir deletion.",
|
||||
_ => {
|
||||
"Runtime cleanup failed. Inspect correlated Runtime diagnostics, resolve the cause, then retry deletion."
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
|
||||
enum WorkdirRemovalOwnerObservation {
|
||||
Running { process_start_marker: u64 },
|
||||
@@ -24741,17 +24813,46 @@ fn execute_reserved_workdir_removal_with_provider(
|
||||
return api.config_store.commit_workdir_removal_removed(&operation);
|
||||
}
|
||||
let Some(status) = observed.working_directory.as_ref() else {
|
||||
let category = workdir_cleanup_failure_category(&observed);
|
||||
let category = if category == "provider_cleanup_failed" {
|
||||
"provider_observation_unknown"
|
||||
} else {
|
||||
category
|
||||
};
|
||||
return api.config_store.fail_workdir_removal_operation(
|
||||
&operation,
|
||||
"provider_observation_unknown",
|
||||
true,
|
||||
category,
|
||||
category != "unsupported_target",
|
||||
);
|
||||
};
|
||||
let corrupted = status.summary.status == WorkingDirectoryStatusKind::Corrupted;
|
||||
if !corrupted
|
||||
&& (status.summary.cleanliness.as_deref() != Some("clean")
|
||||
|| status.summary.status != WorkingDirectoryStatusKind::Active)
|
||||
// A provider summary is not permission to remove a different materialization.
|
||||
// Unknown/corrupted metadata is delegated to Runtime's guarded cleanup, never
|
||||
// repaired into a fabricated identity here.
|
||||
let summary = &status.summary;
|
||||
if summary.occupied_by.is_some() {
|
||||
return api.config_store.complete_workdir_removal_retained(
|
||||
&operation,
|
||||
WorkdirRemovalDisposition::Retained,
|
||||
"blocked_by_live_authority",
|
||||
);
|
||||
}
|
||||
if summary.working_directory_id != operation.working_directory_id
|
||||
|| (summary.repository_id == "unknown"
|
||||
&& summary.status != WorkingDirectoryStatusKind::Corrupted)
|
||||
|| (summary.repository_id != "unknown"
|
||||
&& crate::workdir_removal::workdir_materialization_fingerprint(
|
||||
&workdir_record_from_summary(api, &operation.runtime_id, summary),
|
||||
) != operation.materialization_fingerprint)
|
||||
{
|
||||
return api.config_store.fail_workdir_removal_operation(
|
||||
&operation,
|
||||
"provider_identity_changed",
|
||||
true,
|
||||
);
|
||||
}
|
||||
// Past lifecycle failure is not a current dirtiness observation. Runtime
|
||||
// must revalidate unknown/partial checkouts at the actual deletion boundary.
|
||||
if summary.cleanliness.as_deref() == Some("dirty") {
|
||||
return api.config_store.complete_workdir_removal_retained(
|
||||
&operation,
|
||||
WorkdirRemovalDisposition::Retained,
|
||||
@@ -24769,18 +24870,25 @@ fn execute_reserved_workdir_removal_with_provider(
|
||||
.fail_workdir_removal_operation(&operation, category, retryable);
|
||||
}
|
||||
};
|
||||
if deleted.state != InternalWorkerOperationState::Accepted
|
||||
&& !runtime_reports_workdir_not_found(&deleted)
|
||||
{
|
||||
let category = if deleted
|
||||
.diagnostics
|
||||
.iter()
|
||||
.any(|diagnostic| diagnostic.code == "working_directory_unsupported")
|
||||
{
|
||||
"unsupported_target"
|
||||
let deletion_confirmed = runtime_reports_workdir_not_found(&deleted)
|
||||
|| (deleted.state == InternalWorkerOperationState::Accepted
|
||||
&& deleted.working_directory.as_ref().is_some_and(|status| {
|
||||
status.summary.working_directory_id == operation.working_directory_id
|
||||
&& status.summary.status == WorkingDirectoryStatusKind::NotFound
|
||||
}));
|
||||
if !deletion_confirmed {
|
||||
let category = if deleted.state == InternalWorkerOperationState::Accepted {
|
||||
"provider_cleanup_outcome_unknown"
|
||||
} else {
|
||||
"provider_cleanup_failed"
|
||||
workdir_cleanup_failure_category(&deleted)
|
||||
};
|
||||
tracing::warn!(
|
||||
workdir_id = %operation.working_directory_id,
|
||||
operation_id = %operation.operation_id,
|
||||
runtime_id = %operation.runtime_id,
|
||||
category,
|
||||
"Runtime did not confirm Workdir removal; registry retained"
|
||||
);
|
||||
return api.config_store.fail_workdir_removal_operation(
|
||||
&operation,
|
||||
category,
|
||||
@@ -25066,9 +25174,7 @@ fn build_runtime_cleanup_plan(
|
||||
.unwrap_or_else(|| cleanup_workdir_cleanliness_from_registry(&record.cleanliness));
|
||||
let action = if cleanup_workdir_file_status_is_record_only(file_status) {
|
||||
CleanupTargetKind::WorkdirRecordDelete
|
||||
} else if cleanup_workdir_file_status_is_corrupted(file_status)
|
||||
|| cleanup_workdir_cleanliness_is_clean(cleanliness)
|
||||
{
|
||||
} else if !matches!(cleanliness, CleanupWorkdirCleanliness::Dirty) {
|
||||
CleanupTargetKind::WorkdirCleanCleanup
|
||||
} else {
|
||||
CleanupTargetKind::WorkdirDirtyDiscard
|
||||
@@ -25096,12 +25202,12 @@ fn build_runtime_cleanup_plan(
|
||||
} else if cleanup_workdir_file_status_is_record_only(file_status) {
|
||||
"Not-found Workdir record can be deleted from the Backend registry".to_string()
|
||||
} else if cleanup_workdir_file_status_is_corrupted(file_status) {
|
||||
"Corrupted Workdir can be deleted from Runtime storage and Backend registry"
|
||||
"Runtime will verify remaining files and removal authority for this incomplete Workdir"
|
||||
.to_string()
|
||||
} else if matches!(cleanliness, CleanupWorkdirCleanliness::Dirty) {
|
||||
"Dirty Workdir requires explicit discard confirmation before cleanup".to_string()
|
||||
"Dirty Workdir is protected; preserve or resolve current changes before retrying".to_string()
|
||||
} else if matches!(cleanliness, CleanupWorkdirCleanliness::Unknown) {
|
||||
"Workdir clean state is unknown; explicit discard confirmation is required"
|
||||
"Runtime will recheck current changes, identity and removal blockers before deleting"
|
||||
.to_string()
|
||||
} else {
|
||||
"Clean Workdir can be manually cleaned up".to_string()
|
||||
@@ -25303,7 +25409,12 @@ async fn execute_runtime_cleanup_with_context(
|
||||
target_id: candidate.target_id.clone(),
|
||||
action: candidate.action.clone(),
|
||||
status: status.to_string(),
|
||||
message: message.to_string(),
|
||||
message: removal.failure_category.as_deref().map_or_else(
|
||||
|| message.to_string(),
|
||||
|category| {
|
||||
format!("{category}: {}", workdir_removal_failure_guidance(category))
|
||||
},
|
||||
),
|
||||
});
|
||||
}
|
||||
CleanupTargetKind::WorkerDelete => {
|
||||
@@ -53019,7 +53130,7 @@ mod tests {
|
||||
};
|
||||
let source_fingerprint = repository_source_fingerprint(&repository_source);
|
||||
for (workdir_id, _, _) in &fixtures {
|
||||
runtime
|
||||
let created = runtime
|
||||
.create_working_directory(WorkingDirectoryRequest {
|
||||
repository: WorkingDirectoryRepository {
|
||||
id: "repo-test".to_string(),
|
||||
@@ -53042,6 +53153,13 @@ mod tests {
|
||||
"present",
|
||||
"clean",
|
||||
);
|
||||
api.store
|
||||
.upsert_workdir_registry(&workdir_record_from_summary(
|
||||
&api,
|
||||
provider_runtime_id,
|
||||
&created.summary,
|
||||
))
|
||||
.unwrap();
|
||||
}
|
||||
let listener = tokio::net::TcpListener::bind("127.0.0.1:0").await.unwrap();
|
||||
let provider_base_url = format!("http://{}", listener.local_addr().unwrap());
|
||||
@@ -53591,6 +53709,91 @@ mod tests {
|
||||
assert_eq!(response.status(), StatusCode::FORBIDDEN);
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn current_worker_can_reattach_usable_pending_checkout_after_failed_removal_but_not_during_attempt()
|
||||
{
|
||||
let mut fixture = manual_worker_assignment_fixture().await;
|
||||
let identity =
|
||||
worker_runtime::auth::RuntimeIdentityMaterial::generate(&fixture.worker.runtime_id)
|
||||
.unwrap();
|
||||
configure_runtime_request_auth(&mut fixture.api, &identity, &fixture.worker.runtime_id);
|
||||
let api = &fixture.api;
|
||||
api.store
|
||||
.detach_worker_workdir(
|
||||
TEST_WORKSPACE_ID,
|
||||
&fixture.worker,
|
||||
Some(&fixture.docs_workdir_id),
|
||||
"release for removal",
|
||||
)
|
||||
.unwrap();
|
||||
let mut record = api
|
||||
.store
|
||||
.get_workdir_registry(TEST_WORKSPACE_ID, &fixture.docs_workdir_id)
|
||||
.unwrap()
|
||||
.unwrap();
|
||||
record.materialization_status = "pending".into();
|
||||
api.store.upsert_workdir_registry(&record).unwrap();
|
||||
let intent =
|
||||
workdir_removal_intent(&record, "account:owner", "remove docs checkout").unwrap();
|
||||
let pending = api
|
||||
.config_store
|
||||
.reserve_workdir_removal_operation(&intent)
|
||||
.unwrap();
|
||||
let path = format!("/api/w/{TEST_WORKSPACE_ID}/workers/self/workdir-attachments");
|
||||
let request = || {
|
||||
runtime_source_request(
|
||||
&identity,
|
||||
Some(&fixture.worker.worker_id),
|
||||
"POST",
|
||||
&path,
|
||||
serde_json::to_vec(
|
||||
&json!({ "alias": "docs", "working_directory_id": fixture.docs_workdir_id }),
|
||||
)
|
||||
.unwrap(),
|
||||
)
|
||||
};
|
||||
let blocked = build_router(api.clone()).oneshot(request()).await.unwrap();
|
||||
let blocked_status = blocked.status();
|
||||
let blocked_body = to_bytes(blocked.into_body(), usize::MAX).await.unwrap();
|
||||
assert_eq!(
|
||||
blocked_status,
|
||||
StatusCode::CONFLICT,
|
||||
"{}",
|
||||
String::from_utf8_lossy(&blocked_body)
|
||||
);
|
||||
api.config_store
|
||||
.fail_workdir_removal_operation(&pending, "provider_unavailable", true)
|
||||
.unwrap();
|
||||
let attached = build_router(api.clone()).oneshot(request()).await.unwrap();
|
||||
let attached_status = attached.status();
|
||||
let attached_body = to_bytes(attached.into_body(), usize::MAX).await.unwrap();
|
||||
assert_eq!(
|
||||
attached_status,
|
||||
StatusCode::OK,
|
||||
"{}",
|
||||
String::from_utf8_lossy(&attached_body)
|
||||
);
|
||||
let attached: server_api::CurrentWorkerWorkdirAttachmentResponse =
|
||||
serde_json::from_slice(&attached_body).unwrap();
|
||||
assert!(attached.attached);
|
||||
assert_eq!(attached.working_directory_id, fixture.docs_workdir_id);
|
||||
assert_eq!(
|
||||
api.store
|
||||
.get_workdir_registry(TEST_WORKSPACE_ID, &fixture.docs_workdir_id)
|
||||
.unwrap()
|
||||
.unwrap()
|
||||
.materialization_status,
|
||||
"pending"
|
||||
);
|
||||
assert!(
|
||||
api.config_store
|
||||
.workdir_removal_guards(&pending)
|
||||
.unwrap()
|
||||
.iter()
|
||||
.any(|guard| guard.category == "active_attachment")
|
||||
);
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn current_worker_attachment_list_and_stale_detach_use_signed_identity() {
|
||||
let mut fixture = manual_worker_assignment_fixture().await;
|
||||
@@ -54272,6 +54475,18 @@ mod tests {
|
||||
}
|
||||
}
|
||||
|
||||
fn confirmed_workdir_cleanup_result() -> crate::hosts::RuntimeWorkingDirectoryResult {
|
||||
workdir_removal_result(
|
||||
InternalWorkerOperationState::Rejected,
|
||||
None,
|
||||
vec![RuntimeDiagnostic {
|
||||
code: "working_directory_not_found".into(),
|
||||
severity: HostDiagnosticSeverity::Warning,
|
||||
message: "authoritatively absent".into(),
|
||||
}],
|
||||
)
|
||||
}
|
||||
|
||||
fn reserve_removal_fixture(
|
||||
api: &WorkspaceApi,
|
||||
working_directory_id: &str,
|
||||
@@ -54312,7 +54527,7 @@ mod tests {
|
||||
Some(clean_summary),
|
||||
Vec::new(),
|
||||
),
|
||||
workdir_removal_result(InternalWorkerOperationState::Accepted, None, Vec::new()),
|
||||
confirmed_workdir_cleanup_result(),
|
||||
);
|
||||
let removed = execute_reserved_workdir_removal_with_provider(
|
||||
&api,
|
||||
@@ -54378,7 +54593,7 @@ mod tests {
|
||||
message: "timeout".to_string(),
|
||||
}],
|
||||
),
|
||||
workdir_removal_result(InternalWorkerOperationState::Accepted, None, Vec::new()),
|
||||
confirmed_workdir_cleanup_result(),
|
||||
);
|
||||
let unknown = execute_reserved_workdir_removal_with_provider(
|
||||
&api,
|
||||
@@ -54400,7 +54615,7 @@ mod tests {
|
||||
Some(dirty_summary),
|
||||
Vec::new(),
|
||||
),
|
||||
workdir_removal_result(InternalWorkerOperationState::Accepted, None, Vec::new()),
|
||||
confirmed_workdir_cleanup_result(),
|
||||
);
|
||||
let dirty = execute_reserved_workdir_removal_with_provider(
|
||||
&api,
|
||||
@@ -54418,7 +54633,7 @@ mod tests {
|
||||
Some(clean_retry_summary),
|
||||
Vec::new(),
|
||||
),
|
||||
workdir_removal_result(InternalWorkerOperationState::Accepted, None, Vec::new()),
|
||||
confirmed_workdir_cleanup_result(),
|
||||
);
|
||||
let removed_after_retry = execute_reserved_workdir_removal_with_provider(
|
||||
&api,
|
||||
@@ -54444,7 +54659,7 @@ mod tests {
|
||||
Some(corrupted_summary),
|
||||
Vec::new(),
|
||||
),
|
||||
workdir_removal_result(InternalWorkerOperationState::Accepted, None, Vec::new()),
|
||||
confirmed_workdir_cleanup_result(),
|
||||
);
|
||||
let corrupted = execute_reserved_workdir_removal_with_provider(
|
||||
&api,
|
||||
@@ -54493,6 +54708,297 @@ mod tests {
|
||||
assert_eq!(unsupported_provider.cleanup_calls(), 1);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn workdir_cleanup_failure_classification_is_bounded_and_guides_normal_retry() {
|
||||
for (code, category) in [
|
||||
("working_directory_cleanup_mount_present", "mount_present"),
|
||||
(
|
||||
"working_directory_cleanup_mount_check_unavailable",
|
||||
"mount_check_unavailable",
|
||||
),
|
||||
(
|
||||
"working_directory_cleanup_permission_denied",
|
||||
"permission_denied",
|
||||
),
|
||||
("working_directory_cleanup_resource_busy", "resource_busy"),
|
||||
(
|
||||
"working_directory_cleanup_storage_unavailable",
|
||||
"storage_unavailable",
|
||||
),
|
||||
(
|
||||
"working_directory_cleanup_ownership_unknown",
|
||||
"ownership_unknown",
|
||||
),
|
||||
(
|
||||
"working_directory_cleanup_changes_present",
|
||||
"changes_present",
|
||||
),
|
||||
(
|
||||
"working_directory_cleanup_changes_unknown",
|
||||
"changes_unknown",
|
||||
),
|
||||
(
|
||||
"arbitrary provider text /private/path",
|
||||
"provider_cleanup_failed",
|
||||
),
|
||||
] {
|
||||
let result = workdir_removal_result(
|
||||
InternalWorkerOperationState::Rejected,
|
||||
None,
|
||||
vec![RuntimeDiagnostic {
|
||||
code: code.into(),
|
||||
severity: HostDiagnosticSeverity::Error,
|
||||
message: "secret /private/path".into(),
|
||||
}],
|
||||
);
|
||||
assert_eq!(workdir_cleanup_failure_category(&result), category);
|
||||
let guidance = workdir_removal_failure_guidance(category);
|
||||
assert!(!guidance.contains("/private/path"));
|
||||
assert!(!guidance.contains("secret"));
|
||||
assert!(guidance.len() <= 500);
|
||||
assert!(guidance.contains("retry"));
|
||||
}
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn workdir_removal_retry_rechecks_pending_materialization_changes_and_attachments() {
|
||||
let workspace = tempfile::tempdir().unwrap();
|
||||
init_clean_git_workspace(workspace.path());
|
||||
let api = test_api(workspace.path()).await;
|
||||
let (operation, mut summary) = reserve_removal_fixture(&api, "cleanup-retry");
|
||||
let failed_provider = FakeWorkdirRemovalProvider::new(
|
||||
workdir_removal_result(
|
||||
InternalWorkerOperationState::Accepted,
|
||||
Some(summary.clone()),
|
||||
Vec::new(),
|
||||
),
|
||||
workdir_removal_result(
|
||||
InternalWorkerOperationState::Rejected,
|
||||
None,
|
||||
vec![RuntimeDiagnostic {
|
||||
code: "working_directory_cleanup_mount_present".into(),
|
||||
severity: HostDiagnosticSeverity::Error,
|
||||
message: "/private/mount credentials must not escape".into(),
|
||||
}],
|
||||
),
|
||||
);
|
||||
let failed = execute_reserved_workdir_removal_with_provider(
|
||||
&api,
|
||||
operation,
|
||||
false,
|
||||
&failed_provider,
|
||||
)
|
||||
.unwrap();
|
||||
assert!(failed.retryable);
|
||||
assert_eq!(failed.failure_category.as_deref(), Some("mount_present"));
|
||||
assert!(
|
||||
api.store
|
||||
.get_workdir_registry(TEST_WORKSPACE_ID, "cleanup-retry")
|
||||
.unwrap()
|
||||
.is_some()
|
||||
);
|
||||
assert!(!serde_json::to_string(&failed).unwrap().contains("/private"));
|
||||
|
||||
summary.status = WorkingDirectoryStatusKind::CleanupPending;
|
||||
summary.cleanliness = Some("dirty".into());
|
||||
let dirty_provider = FakeWorkdirRemovalProvider::new(
|
||||
workdir_removal_result(
|
||||
InternalWorkerOperationState::Accepted,
|
||||
Some(summary.clone()),
|
||||
Vec::new(),
|
||||
),
|
||||
confirmed_workdir_cleanup_result(),
|
||||
);
|
||||
let dirty =
|
||||
execute_reserved_workdir_removal_with_provider(&api, failed, false, &dirty_provider)
|
||||
.unwrap();
|
||||
assert_eq!(dirty.disposition, Some(WorkdirRemovalDisposition::Retained));
|
||||
assert!(dirty.retryable);
|
||||
assert_eq!(dirty_provider.cleanup_calls(), 0);
|
||||
|
||||
let worker = seed_cleanup_worker(&api, 731, "normal");
|
||||
seed_cleanup_link(&api, &worker, "cleanup-retry");
|
||||
let occupied_provider = FakeWorkdirRemovalProvider::new(
|
||||
workdir_removal_result(
|
||||
InternalWorkerOperationState::Accepted,
|
||||
Some(summary.clone()),
|
||||
Vec::new(),
|
||||
),
|
||||
confirmed_workdir_cleanup_result(),
|
||||
);
|
||||
let occupied =
|
||||
execute_reserved_workdir_removal_with_provider(&api, dirty, false, &occupied_provider)
|
||||
.unwrap();
|
||||
assert_eq!(
|
||||
occupied.failure_category.as_deref(),
|
||||
Some("blocked_by_live_authority")
|
||||
);
|
||||
assert!(occupied.retryable);
|
||||
assert_eq!(occupied_provider.cleanup_calls(), 0);
|
||||
api.store
|
||||
.detach_worker_workdir(
|
||||
TEST_WORKSPACE_ID,
|
||||
&RuntimeWorkerRef::new("runtime-test", worker),
|
||||
Some("cleanup-retry"),
|
||||
"resolve retry attachment",
|
||||
)
|
||||
.unwrap();
|
||||
|
||||
// Unknown because checkout was partially removed is delegated to Runtime
|
||||
// for current change/identity protection, not denied by the old state.
|
||||
summary.cleanliness = Some("unknown".into());
|
||||
let retry_provider = FakeWorkdirRemovalProvider::new(
|
||||
workdir_removal_result(
|
||||
InternalWorkerOperationState::Accepted,
|
||||
Some(summary),
|
||||
Vec::new(),
|
||||
),
|
||||
confirmed_workdir_cleanup_result(),
|
||||
);
|
||||
let removed = execute_reserved_workdir_removal_with_provider(
|
||||
&api,
|
||||
occupied.clone(),
|
||||
false,
|
||||
&retry_provider,
|
||||
)
|
||||
.unwrap();
|
||||
assert_eq!(removed.operation_id, occupied.operation_id);
|
||||
assert_eq!(removed.attempt_count, 4);
|
||||
assert_eq!(
|
||||
removed.disposition,
|
||||
Some(WorkdirRemovalDisposition::Removed)
|
||||
);
|
||||
assert_eq!(retry_provider.cleanup_calls(), 1);
|
||||
assert!(
|
||||
api.store
|
||||
.get_workdir_registry(TEST_WORKSPACE_ID, "cleanup-retry")
|
||||
.unwrap()
|
||||
.is_none()
|
||||
);
|
||||
assert_eq!(
|
||||
execute_reserved_workdir_removal_with_provider(
|
||||
&api,
|
||||
removed.clone(),
|
||||
false,
|
||||
&retry_provider
|
||||
)
|
||||
.unwrap(),
|
||||
removed
|
||||
);
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn workdir_observation_failure_preserves_bounded_os_cause_without_removing_registry() {
|
||||
let workspace = tempfile::tempdir().unwrap();
|
||||
init_clean_git_workspace(workspace.path());
|
||||
let api = test_api(workspace.path()).await;
|
||||
for (id, transport_error) in [
|
||||
("observation-denied", false),
|
||||
("rpc-observation-denied", true),
|
||||
] {
|
||||
let (operation, _) = reserve_removal_fixture(&api, id);
|
||||
let provider = FakeWorkdirRemovalProvider::new(
|
||||
workdir_removal_result(
|
||||
InternalWorkerOperationState::Rejected,
|
||||
None,
|
||||
vec![RuntimeDiagnostic {
|
||||
code: "working_directory_cleanup_permission_denied".into(),
|
||||
severity: HostDiagnosticSeverity::Error,
|
||||
message: "permission denied /private/path".into(),
|
||||
}],
|
||||
),
|
||||
confirmed_workdir_cleanup_result(),
|
||||
);
|
||||
if transport_error {
|
||||
*provider.observation.lock().unwrap() =
|
||||
Some(Err(RuntimeRegistryError::RuntimeOperationFailed {
|
||||
runtime_id: "runtime-test".into(),
|
||||
code: "working_directory_cleanup_permission_denied".into(),
|
||||
message: "permission denied /private/path".into(),
|
||||
}));
|
||||
}
|
||||
let result =
|
||||
execute_reserved_workdir_removal_with_provider(&api, operation, false, &provider)
|
||||
.unwrap();
|
||||
assert_eq!(
|
||||
result.failure_category.as_deref(),
|
||||
Some("permission_denied")
|
||||
);
|
||||
assert!(result.retryable);
|
||||
assert_eq!(provider.cleanup_calls(), 0);
|
||||
assert!(
|
||||
!serde_json::to_string(&result)
|
||||
.unwrap()
|
||||
.contains("/private/path")
|
||||
);
|
||||
assert!(
|
||||
api.store
|
||||
.get_workdir_registry(TEST_WORKSPACE_ID, id)
|
||||
.unwrap()
|
||||
.is_some()
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn workdir_removal_keeps_registry_for_changed_identity_and_unknown_cleanup_outcome() {
|
||||
let workspace = tempfile::tempdir().unwrap();
|
||||
init_clean_git_workspace(workspace.path());
|
||||
let api = test_api(workspace.path()).await;
|
||||
for (id, changed_identity) in [("identity-mismatch", true), ("unknown-cleanup", false)] {
|
||||
let (operation, mut summary) = reserve_removal_fixture(&api, id);
|
||||
if changed_identity {
|
||||
summary.repository_id = "another-repository".into();
|
||||
}
|
||||
let provider = FakeWorkdirRemovalProvider::new(
|
||||
workdir_removal_result(
|
||||
InternalWorkerOperationState::Accepted,
|
||||
Some(summary),
|
||||
Vec::new(),
|
||||
),
|
||||
workdir_removal_result(InternalWorkerOperationState::Accepted, None, Vec::new()),
|
||||
);
|
||||
let result =
|
||||
execute_reserved_workdir_removal_with_provider(&api, operation, false, &provider)
|
||||
.unwrap();
|
||||
assert!(result.retryable);
|
||||
assert_eq!(
|
||||
result.failure_category.as_deref(),
|
||||
Some(if changed_identity {
|
||||
"provider_identity_changed"
|
||||
} else {
|
||||
"provider_cleanup_outcome_unknown"
|
||||
})
|
||||
);
|
||||
assert_eq!(provider.cleanup_calls(), usize::from(!changed_identity));
|
||||
assert!(
|
||||
api.store
|
||||
.get_workdir_registry(TEST_WORKSPACE_ID, id)
|
||||
.unwrap()
|
||||
.is_some()
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn cleanup_plan_offers_normal_revalidation_for_prior_failure_and_unknown() {
|
||||
let workspace = tempfile::tempdir().unwrap();
|
||||
init_clean_git_workspace(workspace.path());
|
||||
let api = test_api(workspace.path()).await;
|
||||
for (id, state) in [
|
||||
("pending-retry", "pending"),
|
||||
("unknown-retry", "unknown"),
|
||||
("corrupt-retry", "corrupted"),
|
||||
] {
|
||||
seed_cleanup_workdir(&api, id, state, "unknown");
|
||||
}
|
||||
let plan = build_runtime_cleanup_plan(&api, "runtime-test").unwrap();
|
||||
for candidate in plan.workdirs {
|
||||
assert_eq!(candidate.action, CleanupTargetKind::WorkdirCleanCleanup);
|
||||
assert!(candidate.blocking_reason.is_none());
|
||||
}
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn concurrent_orphan_recovery_runs_delayed_provider_cleanup_once() {
|
||||
let workspace = tempfile::tempdir().unwrap();
|
||||
@@ -54518,7 +55024,7 @@ mod tests {
|
||||
Some(clean_summary),
|
||||
Vec::new(),
|
||||
),
|
||||
workdir_removal_result(InternalWorkerOperationState::Accepted, None, Vec::new()),
|
||||
confirmed_workdir_cleanup_result(),
|
||||
)
|
||||
.with_observation_delay(std::time::Duration::from_millis(100)),
|
||||
);
|
||||
|
||||
@@ -10254,6 +10254,19 @@ impl ControlPlaneStore for SqliteWorkspaceStore {
|
||||
record.workdir_id, active.worker.runtime_id, active.worker.worker_id
|
||||
)));
|
||||
}
|
||||
// Direct attachments (including ordinary recovery/reconnection)
|
||||
// must share the same atomic removal fence as reserved attachments.
|
||||
let removal_pending: bool = tx.query_row(
|
||||
"SELECT EXISTS(SELECT 1 FROM workdir_removal_operations WHERE workspace_id=?1 AND workdir_id=?2 AND state='pending')",
|
||||
params![record.workspace_id, record.workdir_id],
|
||||
|row| row.get(0),
|
||||
)?;
|
||||
if removal_pending {
|
||||
return Err(Error::WorkdirAttachmentConflict(format!(
|
||||
"Workdir {} has a pending durable removal operation",
|
||||
record.workdir_id
|
||||
)));
|
||||
}
|
||||
let write = tx.execute(
|
||||
r#"INSERT INTO worker_workdir_links (
|
||||
workspace_id, runtime_id, worker_id, workdir_id, alias, capabilities, linked_at, unlinked_at, connection_id
|
||||
|
||||
@@ -438,7 +438,9 @@ impl SqliteWorkspaceStore {
|
||||
"retained completion cannot use removed disposition".to_string(),
|
||||
));
|
||||
}
|
||||
self.finish_workdir_removal_operation(operation, disposition, false, Some(category), false)
|
||||
// Retention describes this attempt's current blockers, not a permanent
|
||||
// ban. A subsequent normal request rechecks all live authority.
|
||||
self.finish_workdir_removal_operation(operation, disposition, true, Some(category), false)
|
||||
}
|
||||
|
||||
pub fn fail_workdir_removal_operation(
|
||||
|
||||
Reference in New Issue
Block a user