863 lines
33 KiB
Rust
863 lines
33 KiB
Rust
use std::collections::VecDeque;
|
|
|
|
use crate::current_repo_index::{CurrentRepoIndex, CurrentRepoIndexHandle};
|
|
use crate::parallel::config::ParallelPhase1Config;
|
|
use crate::parallel::dead_repo_blacklist::{
|
|
DeadRepoBlacklist, DeadRepoFailureOutcome, DeadRepoSuccessOutcome,
|
|
};
|
|
use crate::parallel::repo_scheduler::{
|
|
InFlightRepoTable, RepoCompletion, RepoRequestAction, TransportCompletion,
|
|
TransportRequestAction, TransportStateTables,
|
|
};
|
|
use crate::parallel::stats::ParallelRunStats;
|
|
use crate::parallel::types::{
|
|
RepoDedupKey, RepoIdentity, RepoKey, RepoRequester, RepoSyncResultEnvelope, RepoSyncTask,
|
|
RepoTransportErrorClass, RepoTransportMode, RepoTransportResultEnvelope,
|
|
RepoTransportResultKind, RepoTransportTask, TalInputSpec,
|
|
};
|
|
use crate::policy::SyncPreference;
|
|
|
|
pub struct GlobalRunCoordinator {
|
|
pub config: ParallelPhase1Config,
|
|
pub tal_inputs: Vec<TalInputSpec>,
|
|
pub current_repo_index: CurrentRepoIndexHandle,
|
|
pub in_flight_repos: InFlightRepoTable,
|
|
pub transport_tables: TransportStateTables,
|
|
pub pending_repo_tasks: VecDeque<RepoSyncTask>,
|
|
pub pending_transport_tasks: VecDeque<RepoTransportTask>,
|
|
pub stats: ParallelRunStats,
|
|
/// Mutable working copy of the dead-repo blacklist (#141): failure/success
|
|
/// counters update during the run and persist at run end. Skip decisions
|
|
/// inside `transport_tables` use the frozen run-start snapshot.
|
|
pub dead_repo_blacklist: Option<DeadRepoBlacklist>,
|
|
}
|
|
|
|
impl GlobalRunCoordinator {
|
|
pub fn new(config: ParallelPhase1Config, tal_inputs: Vec<TalInputSpec>) -> Self {
|
|
let dead_repo_blacklist = config
|
|
.dead_repo_blacklist
|
|
.as_ref()
|
|
.map(|blacklist_config| {
|
|
let (blacklist, warning) = DeadRepoBlacklist::load(&blacklist_config.path);
|
|
if let Some(warning) = warning {
|
|
crate::progress_log::emit(
|
|
"dead_repo_blacklist_load_warning",
|
|
serde_json::json!({ "warning": warning }),
|
|
);
|
|
}
|
|
blacklist
|
|
});
|
|
let mut transport_tables = TransportStateTables::new();
|
|
if let Some(blacklist) = dead_repo_blacklist.as_ref() {
|
|
// Frozen run-start snapshot for skip decisions.
|
|
transport_tables.set_dead_repo_blacklist(blacklist.clone());
|
|
}
|
|
Self {
|
|
config,
|
|
tal_inputs,
|
|
current_repo_index: CurrentRepoIndex::shared(),
|
|
in_flight_repos: InFlightRepoTable::new(),
|
|
transport_tables,
|
|
pending_repo_tasks: VecDeque::new(),
|
|
pending_transport_tasks: VecDeque::new(),
|
|
stats: ParallelRunStats::default(),
|
|
dead_repo_blacklist,
|
|
}
|
|
}
|
|
|
|
pub fn current_repo_index_handle(&self) -> CurrentRepoIndexHandle {
|
|
self.current_repo_index.clone()
|
|
}
|
|
|
|
pub fn register_repo_request(
|
|
&mut self,
|
|
repo_key: RepoKey,
|
|
requester: RepoRequester,
|
|
validation_time: time::OffsetDateTime,
|
|
sync_preference: SyncPreference,
|
|
priority: u8,
|
|
) -> RepoRequestAction {
|
|
let action = self.in_flight_repos.register_request(
|
|
repo_key,
|
|
requester,
|
|
validation_time,
|
|
sync_preference,
|
|
priority,
|
|
);
|
|
|
|
match &action {
|
|
RepoRequestAction::Enqueued(task) => {
|
|
self.stats.repo_tasks_total += 1;
|
|
self.pending_repo_tasks.push_back(task.clone());
|
|
self.stats.repo_queue_depth = self.pending_repo_tasks.len();
|
|
}
|
|
RepoRequestAction::Reused(_) | RepoRequestAction::FailedReuse { .. } => {
|
|
self.stats.repo_tasks_reused += 1;
|
|
}
|
|
RepoRequestAction::Waiting => {}
|
|
}
|
|
|
|
action
|
|
}
|
|
|
|
pub fn pop_next_repo_task(&mut self) -> Option<RepoSyncTask> {
|
|
let next = self.pending_repo_tasks.pop_front();
|
|
self.stats.repo_queue_depth = self.pending_repo_tasks.len();
|
|
next
|
|
}
|
|
|
|
pub fn mark_repo_running(
|
|
&mut self,
|
|
repo_key: &RepoKey,
|
|
started_at: time::OffsetDateTime,
|
|
) -> Result<(), String> {
|
|
self.in_flight_repos.mark_running(repo_key, started_at)?;
|
|
self.stats.repo_tasks_running += 1;
|
|
Ok(())
|
|
}
|
|
|
|
pub fn complete_repo_success(
|
|
&mut self,
|
|
result: RepoSyncResultEnvelope,
|
|
finished_at: time::OffsetDateTime,
|
|
) -> Result<RepoCompletion, String> {
|
|
let repo_key = result.repo_key.clone();
|
|
let completion = self
|
|
.in_flight_repos
|
|
.complete_success(&repo_key, result, finished_at)?;
|
|
self.stats.repo_tasks_running = self.stats.repo_tasks_running.saturating_sub(1);
|
|
Ok(completion)
|
|
}
|
|
|
|
pub fn complete_repo_failure(
|
|
&mut self,
|
|
result: RepoSyncResultEnvelope,
|
|
finished_at: time::OffsetDateTime,
|
|
) -> Result<RepoCompletion, String> {
|
|
let repo_key = result.repo_key.clone();
|
|
let completion = self
|
|
.in_flight_repos
|
|
.complete_failure(&repo_key, result, finished_at)?;
|
|
self.stats.repo_tasks_running = self.stats.repo_tasks_running.saturating_sub(1);
|
|
self.stats.repo_tasks_failed += 1;
|
|
Ok(completion)
|
|
}
|
|
|
|
pub fn register_transport_request(
|
|
&mut self,
|
|
identity: RepoIdentity,
|
|
requester: RepoRequester,
|
|
validation_time: time::OffsetDateTime,
|
|
priority: u8,
|
|
rsync_scope_uri: String,
|
|
rsync_failure_scope_uri: Option<String>,
|
|
sync_preference: SyncPreference,
|
|
retry_short_timeout: bool,
|
|
) -> TransportRequestAction {
|
|
let mut action = self.transport_tables.register_transport_request(
|
|
identity,
|
|
requester,
|
|
validation_time,
|
|
priority,
|
|
rsync_scope_uri,
|
|
rsync_failure_scope_uri,
|
|
sync_preference,
|
|
);
|
|
match &mut action {
|
|
TransportRequestAction::Enqueue(task) => {
|
|
if retry_short_timeout {
|
|
task.retry_short_timeout = true;
|
|
}
|
|
self.stats.repo_tasks_total += 1;
|
|
self.pending_transport_tasks.push_back(task.clone());
|
|
self.stats.repo_queue_depth = self.pending_transport_tasks.len();
|
|
}
|
|
TransportRequestAction::ReusedSuccess(_)
|
|
| TransportRequestAction::ReusedTerminalFailure(_) => {
|
|
self.stats.repo_tasks_reused += 1;
|
|
}
|
|
TransportRequestAction::Waiting { .. } => {}
|
|
}
|
|
action
|
|
}
|
|
|
|
pub fn push_transport_task(&mut self, task: RepoTransportTask) {
|
|
self.stats.repo_tasks_total += 1;
|
|
self.pending_transport_tasks.push_back(task);
|
|
self.stats.repo_queue_depth = self.pending_transport_tasks.len();
|
|
}
|
|
|
|
pub fn pop_next_transport_task(&mut self) -> Option<RepoTransportTask> {
|
|
let next = self.pending_transport_tasks.pop_front();
|
|
self.stats.repo_queue_depth = self.pending_transport_tasks.len();
|
|
next
|
|
}
|
|
|
|
pub fn mark_transport_running(
|
|
&mut self,
|
|
dedup_key: &crate::parallel::types::RepoDedupKey,
|
|
started_at: time::OffsetDateTime,
|
|
) -> Result<(), String> {
|
|
self.transport_tables
|
|
.mark_transport_running(dedup_key, started_at)?;
|
|
self.stats.repo_tasks_running += 1;
|
|
Ok(())
|
|
}
|
|
|
|
pub fn complete_transport_result(
|
|
&mut self,
|
|
result: RepoTransportResultEnvelope,
|
|
finished_at: time::OffsetDateTime,
|
|
) -> Result<TransportCompletion, String> {
|
|
let completion = self
|
|
.transport_tables
|
|
.complete_transport_result(result.clone(), finished_at)?;
|
|
self.stats.repo_tasks_running = self.stats.repo_tasks_running.saturating_sub(1);
|
|
if matches!(
|
|
result.result,
|
|
crate::parallel::types::RepoTransportResultKind::Failed { .. }
|
|
) && result.mode == crate::parallel::types::RepoTransportMode::Rsync
|
|
{
|
|
self.stats.repo_tasks_failed += 1;
|
|
}
|
|
self.update_dead_repo_blacklist(&result, finished_at);
|
|
Ok(completion)
|
|
}
|
|
|
|
/// Update dead-repo blacklist counters (#141) from a real worker result.
|
|
/// Runs on the single pump thread; synthesized terminal envelopes never
|
|
/// pass through here, so no double counting is possible.
|
|
fn update_dead_repo_blacklist(
|
|
&mut self,
|
|
result: &RepoTransportResultEnvelope,
|
|
finished_at: time::OffsetDateTime,
|
|
) {
|
|
let Some(blacklist_config) = self.config.dead_repo_blacklist.clone() else {
|
|
return;
|
|
};
|
|
let Some(blacklist) = self.dead_repo_blacklist.as_mut() else {
|
|
return;
|
|
};
|
|
let now_unix = finished_at.unix_timestamp().max(0) as u64;
|
|
let key: Option<(RepoTransportMode, String)> = match (&result.dedup_key, &result.result) {
|
|
(RepoDedupKey::RrdpNotify { notification_uri }, RepoTransportResultKind::Success { .. }) => {
|
|
Some((RepoTransportMode::Rrdp, notification_uri.clone()))
|
|
}
|
|
(RepoDedupKey::RsyncScope { .. }, RepoTransportResultKind::Success { .. }) => {
|
|
Some((RepoTransportMode::Rsync, result.repo_identity.rsync_base_uri.clone()))
|
|
}
|
|
(
|
|
RepoDedupKey::RrdpNotify { notification_uri },
|
|
RepoTransportResultKind::Failed {
|
|
error_class: RepoTransportErrorClass::TransportFetch,
|
|
..
|
|
},
|
|
) => Some((RepoTransportMode::Rrdp, notification_uri.clone())),
|
|
(
|
|
RepoDedupKey::RsyncScope { .. },
|
|
RepoTransportResultKind::Failed {
|
|
error_class: RepoTransportErrorClass::TransportFetch,
|
|
..
|
|
},
|
|
) => Some((RepoTransportMode::Rsync, result.repo_identity.rsync_base_uri.clone())),
|
|
_ => None,
|
|
};
|
|
let Some((transport, uri)) = key else {
|
|
return;
|
|
};
|
|
match &result.result {
|
|
RepoTransportResultKind::Success { .. } => {
|
|
if let Some(outcome) = blacklist.record_transport_success(transport, &uri) {
|
|
crate::progress_log::emit(
|
|
"dead_repo_blacklist_remove",
|
|
serde_json::json!({
|
|
"transport": transport.as_str(),
|
|
"uri": uri,
|
|
"reason": match outcome {
|
|
DeadRepoSuccessOutcome::CounterReset => "counter_reset",
|
|
DeadRepoSuccessOutcome::BlacklistRemoved => "fetch_success",
|
|
},
|
|
}),
|
|
);
|
|
}
|
|
}
|
|
RepoTransportResultKind::Failed { .. } => {
|
|
if let DeadRepoFailureOutcome::Admitted = blacklist.record_transport_failure(
|
|
transport,
|
|
&uri,
|
|
now_unix,
|
|
blacklist_config.fail_threshold,
|
|
blacklist_config.capacity,
|
|
) {
|
|
crate::progress_log::emit(
|
|
"dead_repo_blacklist_admit",
|
|
serde_json::json!({
|
|
"transport": transport.as_str(),
|
|
"uri": uri,
|
|
"fail_threshold": blacklist_config.fail_threshold,
|
|
}),
|
|
);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
pub fn dead_repo_blacklist(&self) -> Option<&DeadRepoBlacklist> {
|
|
self.dead_repo_blacklist.as_ref()
|
|
}
|
|
|
|
pub fn runtime_record(
|
|
&self,
|
|
identity: &RepoIdentity,
|
|
) -> Option<&crate::parallel::repo_scheduler::RepoRuntimeRecord> {
|
|
self.transport_tables.runtime_record(identity)
|
|
}
|
|
|
|
pub fn finalized_runtime_records_for_transport(
|
|
&self,
|
|
dedup_key: &RepoDedupKey,
|
|
) -> Vec<crate::parallel::repo_scheduler::RepoRuntimeRecord> {
|
|
self.transport_tables
|
|
.finalized_runtime_records_for_transport(dedup_key)
|
|
}
|
|
|
|
pub fn finalized_runtime_records_for_transport_result(
|
|
&self,
|
|
result: &RepoTransportResultEnvelope,
|
|
) -> Vec<crate::parallel::repo_scheduler::RepoRuntimeRecord> {
|
|
self.transport_tables
|
|
.finalized_runtime_records_for_transport_result(result)
|
|
}
|
|
|
|
pub fn reset_run_state(&mut self) {
|
|
self.in_flight_repos.reset_run_state();
|
|
self.transport_tables.reset_run_state();
|
|
self.pending_repo_tasks.clear();
|
|
self.pending_transport_tasks.clear();
|
|
self.stats = ParallelRunStats::default();
|
|
if let Ok(mut index) = self.current_repo_index.write() {
|
|
index.clear();
|
|
}
|
|
}
|
|
}
|
|
|
|
#[cfg(test)]
|
|
mod tests {
|
|
use crate::parallel::config::ParallelPhase1Config;
|
|
use crate::parallel::repo_scheduler::RepoRequestAction;
|
|
use crate::parallel::run_coordinator::GlobalRunCoordinator;
|
|
use crate::parallel::types::{
|
|
RepoIdentity, RepoKey, RepoRequester, RepoSyncResultEnvelope, RepoSyncResultKind,
|
|
RepoSyncResultRef, TalInputSpec,
|
|
};
|
|
use crate::policy::SyncPreference;
|
|
use crate::storage::{RepositoryViewEntry, RepositoryViewState};
|
|
|
|
fn requester(tal_id: &str, rir_id: &str, manifest: &str) -> RepoRequester {
|
|
RepoRequester {
|
|
tal_id: tal_id.to_string(),
|
|
rir_id: rir_id.to_string(),
|
|
parent_node_id: None,
|
|
ca_instance_handle_id: format!("{tal_id}:{manifest}"),
|
|
publication_point_rsync_uri: "rsync://example.test/repo/".to_string(),
|
|
manifest_rsync_uri: manifest.to_string(),
|
|
}
|
|
}
|
|
|
|
#[test]
|
|
fn coordinator_holds_tal_inputs_and_default_stats() {
|
|
let coordinator = GlobalRunCoordinator::new(
|
|
ParallelPhase1Config::default(),
|
|
vec![TalInputSpec::from_url("https://example.test/arin.tal")],
|
|
);
|
|
assert_eq!(coordinator.tal_inputs.len(), 1);
|
|
assert_eq!(coordinator.tal_inputs[0].tal_id, "arin");
|
|
assert_eq!(coordinator.stats.repo_tasks_total, 0);
|
|
assert!(coordinator.in_flight_repos.is_empty());
|
|
}
|
|
|
|
#[test]
|
|
fn coordinator_enqueues_new_repo_task_once() {
|
|
let mut coordinator = GlobalRunCoordinator::new(
|
|
ParallelPhase1Config::default(),
|
|
vec![TalInputSpec::from_url("https://example.test/arin.tal")],
|
|
);
|
|
let key = RepoKey::new("rsync://example.test/repo/", None);
|
|
let action = coordinator.register_repo_request(
|
|
key.clone(),
|
|
requester("arin", "arin", "rsync://example.test/repo/root.mft"),
|
|
time::OffsetDateTime::UNIX_EPOCH,
|
|
SyncPreference::RrdpThenRsync,
|
|
7,
|
|
);
|
|
let task = match action {
|
|
RepoRequestAction::Enqueued(task) => task,
|
|
other => panic!("expected enqueue, got {other:?}"),
|
|
};
|
|
assert_eq!(task.priority, 7);
|
|
assert_eq!(coordinator.stats.repo_tasks_total, 1);
|
|
assert_eq!(coordinator.stats.repo_queue_depth, 1);
|
|
|
|
let popped = coordinator.pop_next_repo_task().expect("task queued");
|
|
assert_eq!(popped.repo_key, key);
|
|
assert_eq!(coordinator.stats.repo_queue_depth, 0);
|
|
}
|
|
|
|
#[test]
|
|
fn coordinator_reset_run_state_clears_runtime_only_state() {
|
|
let mut coordinator = GlobalRunCoordinator::new(
|
|
ParallelPhase1Config::default(),
|
|
vec![TalInputSpec::from_url("https://example.test/arin.tal")],
|
|
);
|
|
let identity = RepoIdentity::new(
|
|
Some("https://example.test/notify.xml".to_string()),
|
|
"rsync://example.test/repo/",
|
|
);
|
|
let requester = requester("arin", "arin", "rsync://example.test/repo/root.mft");
|
|
let action = coordinator.register_transport_request(
|
|
identity.clone(),
|
|
requester,
|
|
time::OffsetDateTime::UNIX_EPOCH,
|
|
0,
|
|
"rsync://example.test/repo/".to_string(),
|
|
None,
|
|
SyncPreference::RrdpThenRsync,
|
|
false,
|
|
);
|
|
assert!(matches!(
|
|
action,
|
|
crate::parallel::repo_scheduler::TransportRequestAction::Enqueue(_)
|
|
));
|
|
assert!(coordinator.runtime_record(&identity).is_some());
|
|
assert_eq!(coordinator.pending_transport_tasks.len(), 1);
|
|
|
|
{
|
|
let mut index = coordinator.current_repo_index.write().expect("index write lock");
|
|
index
|
|
.apply_repository_view_entries(&[RepositoryViewEntry {
|
|
rsync_uri: "rsync://example.test/repo/a.roa".to_string(),
|
|
current_hash: Some("aa".repeat(32)),
|
|
repository_source: Some("rsync://example.test/repo/".to_string()),
|
|
object_type: Some("roa".to_string()),
|
|
state: RepositoryViewState::Present,
|
|
}])
|
|
.expect("apply current object");
|
|
assert_eq!(index.active_uri_count(), 1);
|
|
}
|
|
|
|
coordinator.reset_run_state();
|
|
|
|
assert!(coordinator.runtime_record(&identity).is_none());
|
|
assert!(coordinator.pending_transport_tasks.is_empty());
|
|
assert!(coordinator.pending_repo_tasks.is_empty());
|
|
assert_eq!(
|
|
coordinator
|
|
.current_repo_index
|
|
.read()
|
|
.expect("index read lock")
|
|
.active_uri_count(),
|
|
0
|
|
);
|
|
assert_eq!(coordinator.stats.repo_tasks_total, 0);
|
|
}
|
|
|
|
#[test]
|
|
fn coordinator_merges_waiting_requesters_and_reuses_success() {
|
|
let mut coordinator = GlobalRunCoordinator::new(
|
|
ParallelPhase1Config::default(),
|
|
vec![
|
|
TalInputSpec::from_url("https://example.test/arin.tal"),
|
|
TalInputSpec::from_url("https://example.test/apnic.tal"),
|
|
],
|
|
);
|
|
let key = RepoKey::new("rsync://shared.example/repo/", None);
|
|
let _ = coordinator.register_repo_request(
|
|
key.clone(),
|
|
requester("arin", "arin", "rsync://shared.example/repo/root.mft"),
|
|
time::OffsetDateTime::UNIX_EPOCH,
|
|
SyncPreference::RrdpThenRsync,
|
|
0,
|
|
);
|
|
let _ = coordinator.pop_next_repo_task();
|
|
coordinator
|
|
.mark_repo_running(&key, time::OffsetDateTime::UNIX_EPOCH)
|
|
.expect("running");
|
|
|
|
let wait_action = coordinator.register_repo_request(
|
|
key.clone(),
|
|
requester("apnic", "apnic", "rsync://shared.example/repo/child.mft"),
|
|
time::OffsetDateTime::UNIX_EPOCH,
|
|
SyncPreference::RrdpThenRsync,
|
|
0,
|
|
);
|
|
assert_eq!(wait_action, RepoRequestAction::Waiting);
|
|
|
|
let completion = coordinator
|
|
.complete_repo_success(
|
|
RepoSyncResultEnvelope {
|
|
repo_key: key.clone(),
|
|
tal_id: "arin".to_string(),
|
|
rir_id: "arin".to_string(),
|
|
result: RepoSyncResultKind::Success(RepoSyncResultRef {
|
|
repo_key: key.clone(),
|
|
source: "rrdp".to_string(),
|
|
}),
|
|
phase: Some("rrdp_ok".to_string()),
|
|
timing_ms: 12,
|
|
warnings: Vec::new(),
|
|
},
|
|
time::OffsetDateTime::UNIX_EPOCH,
|
|
)
|
|
.expect("success");
|
|
assert_eq!(completion.released_requesters.len(), 1);
|
|
assert_eq!(coordinator.stats.repo_tasks_running, 0);
|
|
|
|
let reuse_action = coordinator.register_repo_request(
|
|
key,
|
|
requester("ripe", "ripe", "rsync://shared.example/repo/again.mft"),
|
|
time::OffsetDateTime::UNIX_EPOCH,
|
|
SyncPreference::RrdpThenRsync,
|
|
0,
|
|
);
|
|
assert!(matches!(reuse_action, RepoRequestAction::Reused(_)));
|
|
assert_eq!(coordinator.stats.repo_tasks_reused, 1);
|
|
}
|
|
|
|
#[test]
|
|
fn coordinator_reuses_failures_without_enqueuing_again() {
|
|
let mut coordinator = GlobalRunCoordinator::new(
|
|
ParallelPhase1Config::default(),
|
|
vec![TalInputSpec::from_url("https://example.test/arin.tal")],
|
|
);
|
|
let key = RepoKey::new("rsync://shared.example/repo/", None);
|
|
let _ = coordinator.register_repo_request(
|
|
key.clone(),
|
|
requester("arin", "arin", "rsync://shared.example/repo/root.mft"),
|
|
time::OffsetDateTime::UNIX_EPOCH,
|
|
SyncPreference::RrdpThenRsync,
|
|
0,
|
|
);
|
|
let _ = coordinator.pop_next_repo_task();
|
|
coordinator
|
|
.mark_repo_running(&key, time::OffsetDateTime::UNIX_EPOCH)
|
|
.expect("running");
|
|
coordinator
|
|
.complete_repo_failure(
|
|
RepoSyncResultEnvelope {
|
|
repo_key: key.clone(),
|
|
tal_id: "arin".to_string(),
|
|
rir_id: "arin".to_string(),
|
|
result: RepoSyncResultKind::Failed {
|
|
detail: "timeout".to_string(),
|
|
},
|
|
phase: Some("rrdp_failed_rsync_failed".to_string()),
|
|
timing_ms: 12,
|
|
warnings: Vec::new(),
|
|
},
|
|
time::OffsetDateTime::UNIX_EPOCH,
|
|
)
|
|
.expect("failure");
|
|
assert_eq!(coordinator.stats.repo_tasks_failed, 1);
|
|
|
|
let action = coordinator.register_repo_request(
|
|
key,
|
|
requester("ripe", "ripe", "rsync://shared.example/repo/child.mft"),
|
|
time::OffsetDateTime::UNIX_EPOCH,
|
|
SyncPreference::RrdpThenRsync,
|
|
0,
|
|
);
|
|
assert_eq!(
|
|
action,
|
|
RepoRequestAction::FailedReuse {
|
|
detail: "timeout".to_string()
|
|
}
|
|
);
|
|
assert_eq!(coordinator.stats.repo_tasks_total, 1);
|
|
assert_eq!(coordinator.stats.repo_tasks_reused, 1);
|
|
}
|
|
|
|
// ---- Dead-repo blacklist (#141) integration tests ----
|
|
|
|
use crate::parallel::dead_repo_blacklist::{DeadRepoBlacklist, DeadRepoBlacklistConfig};
|
|
use crate::parallel::types::{
|
|
RepoDedupKey, RepoTransportErrorClass, RepoTransportMode, RepoTransportResultEnvelope,
|
|
RepoTransportResultKind,
|
|
};
|
|
|
|
const DEAD_NOTIFY: &str = "https://dead.example/notification.xml";
|
|
const DEAD_BASE: &str = "rsync://dead.example/repo/";
|
|
|
|
fn blacklist_config(fail_threshold: u32) -> DeadRepoBlacklistConfig {
|
|
DeadRepoBlacklistConfig {
|
|
path: std::env::temp_dir().join(format!(
|
|
"dead_repo_blacklist_coordinator_test_{}.json",
|
|
std::process::id()
|
|
)),
|
|
fail_threshold,
|
|
capacity: 256,
|
|
}
|
|
}
|
|
|
|
fn coordinator_with_blacklist(
|
|
fail_threshold: u32,
|
|
seed: DeadRepoBlacklist,
|
|
) -> GlobalRunCoordinator {
|
|
let mut config = ParallelPhase1Config::default();
|
|
config.dead_repo_blacklist = Some(blacklist_config(fail_threshold));
|
|
let mut coordinator = GlobalRunCoordinator::new(
|
|
config,
|
|
vec![TalInputSpec::from_url("https://example.test/arin.tal")],
|
|
);
|
|
coordinator.dead_repo_blacklist = Some(seed.clone());
|
|
coordinator.transport_tables.set_dead_repo_blacklist(seed);
|
|
coordinator
|
|
}
|
|
|
|
fn dead_identity() -> RepoIdentity {
|
|
RepoIdentity::new(Some(DEAD_NOTIFY.to_string()), DEAD_BASE)
|
|
}
|
|
|
|
fn dead_requester() -> RepoRequester {
|
|
requester("arin", "arin", "rsync://dead.example/repo/root.mft")
|
|
}
|
|
|
|
fn transport_envelope(
|
|
mode: RepoTransportMode,
|
|
error_class: RepoTransportErrorClass,
|
|
) -> RepoTransportResultEnvelope {
|
|
let (dedup_key, detail) = match mode {
|
|
RepoTransportMode::Rrdp => (
|
|
RepoDedupKey::RrdpNotify {
|
|
notification_uri: DEAD_NOTIFY.to_string(),
|
|
},
|
|
"http request failed: connect timeout".to_string(),
|
|
),
|
|
RepoTransportMode::Rsync => (
|
|
RepoDedupKey::RsyncScope {
|
|
rsync_scope_uri: DEAD_BASE.to_string(),
|
|
},
|
|
"rsync error: timeout waiting for daemon connection".to_string(),
|
|
),
|
|
};
|
|
RepoTransportResultEnvelope {
|
|
dedup_key,
|
|
rsync_failure_scope_uri: None,
|
|
repo_identity: dead_identity(),
|
|
mode,
|
|
tal_id: "arin".to_string(),
|
|
rir_id: "arin".to_string(),
|
|
timing_ms: 1,
|
|
result: RepoTransportResultKind::Failed {
|
|
detail,
|
|
warnings: Vec::new(),
|
|
error_class,
|
|
},
|
|
}
|
|
}
|
|
|
|
fn register_dead(
|
|
coordinator: &mut GlobalRunCoordinator,
|
|
) -> crate::parallel::repo_scheduler::TransportRequestAction {
|
|
coordinator.register_transport_request(
|
|
dead_identity(),
|
|
dead_requester(),
|
|
time::OffsetDateTime::UNIX_EPOCH,
|
|
0,
|
|
DEAD_BASE.to_string(),
|
|
None,
|
|
SyncPreference::RrdpThenRsync,
|
|
false,
|
|
)
|
|
}
|
|
|
|
#[test]
|
|
fn blacklist_admission_after_threshold_transport_failures() {
|
|
let mut coordinator = coordinator_with_blacklist(2, DeadRepoBlacklist::new());
|
|
for _ in 0..2 {
|
|
let action = register_dead(&mut coordinator);
|
|
assert!(matches!(
|
|
action,
|
|
crate::parallel::repo_scheduler::TransportRequestAction::Enqueue(_)
|
|
));
|
|
coordinator
|
|
.complete_transport_result(
|
|
transport_envelope(
|
|
RepoTransportMode::Rrdp,
|
|
RepoTransportErrorClass::TransportFetch,
|
|
),
|
|
time::OffsetDateTime::UNIX_EPOCH,
|
|
)
|
|
.expect("complete");
|
|
// Simulate the next run: run-local scheduler state resets, the
|
|
// persistent blacklist working copy carries over.
|
|
coordinator.reset_run_state();
|
|
}
|
|
let blacklist = coordinator.dead_repo_blacklist().expect("blacklist");
|
|
assert!(blacklist.is_blacklisted(RepoTransportMode::Rrdp, DEAD_NOTIFY));
|
|
// Admission mid-run does not affect this run's frozen snapshot: the
|
|
// scheduler still enqueues rrdp for a fresh identity registration.
|
|
coordinator.reset_run_state();
|
|
let action = register_dead(&mut coordinator);
|
|
assert!(matches!(
|
|
action,
|
|
crate::parallel::repo_scheduler::TransportRequestAction::Enqueue(_)
|
|
));
|
|
}
|
|
|
|
#[test]
|
|
fn blacklist_ignores_protocol_and_unknown_failures() {
|
|
let mut coordinator = coordinator_with_blacklist(1, DeadRepoBlacklist::new());
|
|
for error_class in [
|
|
RepoTransportErrorClass::Protocol,
|
|
RepoTransportErrorClass::Unknown,
|
|
RepoTransportErrorClass::Storage,
|
|
] {
|
|
let action = register_dead(&mut coordinator);
|
|
assert!(matches!(
|
|
action,
|
|
crate::parallel::repo_scheduler::TransportRequestAction::Enqueue(_)
|
|
));
|
|
coordinator
|
|
.complete_transport_result(
|
|
transport_envelope(RepoTransportMode::Rrdp, error_class),
|
|
time::OffsetDateTime::UNIX_EPOCH,
|
|
)
|
|
.expect("complete");
|
|
coordinator.reset_run_state();
|
|
}
|
|
assert!(
|
|
coordinator
|
|
.dead_repo_blacklist()
|
|
.expect("blacklist")
|
|
.is_empty()
|
|
);
|
|
}
|
|
|
|
#[test]
|
|
fn blacklist_skip_rrdp_routes_straight_to_rsync() {
|
|
let mut seed = DeadRepoBlacklist::new();
|
|
seed.record_transport_failure(RepoTransportMode::Rrdp, DEAD_NOTIFY, 100, 1, 256);
|
|
let mut coordinator = coordinator_with_blacklist(3, seed);
|
|
let action = register_dead(&mut coordinator);
|
|
let task = match action {
|
|
crate::parallel::repo_scheduler::TransportRequestAction::Enqueue(task) => task,
|
|
other => panic!("expected rsync enqueue, got {other:?}"),
|
|
};
|
|
assert_eq!(task.mode, RepoTransportMode::Rsync);
|
|
// register_dead passes retry_short_timeout=false (mirroring a prefetch
|
|
// snapshot that recorded no real rrdp failure for skipped runs); the
|
|
// skip path must still keep the short retry profile (#141).
|
|
assert!(task.retry_short_timeout);
|
|
}
|
|
|
|
#[test]
|
|
fn blacklist_dual_dead_terminates_immediately() {
|
|
let mut seed = DeadRepoBlacklist::new();
|
|
seed.record_transport_failure(RepoTransportMode::Rrdp, DEAD_NOTIFY, 100, 1, 256);
|
|
seed.record_transport_failure(RepoTransportMode::Rsync, DEAD_BASE, 100, 1, 256);
|
|
let mut coordinator = coordinator_with_blacklist(3, seed);
|
|
let action = register_dead(&mut coordinator);
|
|
let envelope = match action {
|
|
crate::parallel::repo_scheduler::TransportRequestAction::ReusedTerminalFailure(
|
|
envelope,
|
|
) => envelope,
|
|
other => panic!("expected terminal failure, got {other:?}"),
|
|
};
|
|
match &envelope.result {
|
|
RepoTransportResultKind::Failed {
|
|
detail, warnings, ..
|
|
} => {
|
|
assert!(detail.contains("dead repo blacklist"));
|
|
assert!(warnings
|
|
.iter()
|
|
.any(|warning| warning.message.contains("dead_repo_blacklist_skip_all")));
|
|
}
|
|
other => panic!("expected failed result, got {other:?}"),
|
|
}
|
|
let record = coordinator
|
|
.runtime_record(&dead_identity())
|
|
.expect("record");
|
|
assert_eq!(
|
|
record.state,
|
|
crate::parallel::types::RepoRuntimeState::FailedTerminal
|
|
);
|
|
assert_eq!(coordinator.stats.repo_tasks_total, 0);
|
|
assert_eq!(coordinator.stats.repo_tasks_reused, 1);
|
|
}
|
|
|
|
#[test]
|
|
fn blacklist_rsync_dead_terminates_after_rrdp_failure() {
|
|
// Only rsync blacklisted: rrdp is attempted; when it fails, the rsync
|
|
// fallback is short-circuited into an instant terminal failure.
|
|
let mut seed = DeadRepoBlacklist::new();
|
|
seed.record_transport_failure(RepoTransportMode::Rsync, DEAD_BASE, 100, 1, 256);
|
|
let mut coordinator = coordinator_with_blacklist(3, seed);
|
|
let action = register_dead(&mut coordinator);
|
|
assert!(matches!(
|
|
action,
|
|
crate::parallel::repo_scheduler::TransportRequestAction::Enqueue(_)
|
|
));
|
|
let completion = coordinator
|
|
.complete_transport_result(
|
|
transport_envelope(RepoTransportMode::Rrdp, RepoTransportErrorClass::TransportFetch),
|
|
time::OffsetDateTime::UNIX_EPOCH,
|
|
)
|
|
.expect("complete");
|
|
assert!(completion.follow_up_tasks.is_empty());
|
|
let record = coordinator
|
|
.runtime_record(&dead_identity())
|
|
.expect("record");
|
|
assert_eq!(
|
|
record.state,
|
|
crate::parallel::types::RepoRuntimeState::FailedTerminal
|
|
);
|
|
let terminal = record.terminal_failure.as_ref().expect("terminal failure");
|
|
match &terminal.result {
|
|
RepoTransportResultKind::Failed { detail, .. } => {
|
|
assert!(detail.contains("dead repo blacklist"));
|
|
}
|
|
other => panic!("expected failed result, got {other:?}"),
|
|
}
|
|
}
|
|
|
|
#[test]
|
|
fn blacklist_success_resets_failure_counter() {
|
|
let mut coordinator = coordinator_with_blacklist(3, DeadRepoBlacklist::new());
|
|
// One transport failure starts counting.
|
|
let action = register_dead(&mut coordinator);
|
|
assert!(matches!(
|
|
action,
|
|
crate::parallel::repo_scheduler::TransportRequestAction::Enqueue(_)
|
|
));
|
|
coordinator
|
|
.complete_transport_result(
|
|
transport_envelope(RepoTransportMode::Rrdp, RepoTransportErrorClass::TransportFetch),
|
|
time::OffsetDateTime::UNIX_EPOCH,
|
|
)
|
|
.expect("complete");
|
|
assert_eq!(coordinator.dead_repo_blacklist().expect("blacklist").len(), 1);
|
|
|
|
// A later success clears the entry.
|
|
coordinator.reset_run_state();
|
|
let action = register_dead(&mut coordinator);
|
|
assert!(matches!(
|
|
action,
|
|
crate::parallel::repo_scheduler::TransportRequestAction::Enqueue(_)
|
|
));
|
|
let mut success = transport_envelope(RepoTransportMode::Rrdp, RepoTransportErrorClass::Unknown);
|
|
success.result = RepoTransportResultKind::Success {
|
|
source: "rrdp".to_string(),
|
|
warnings: Vec::new(),
|
|
};
|
|
coordinator
|
|
.complete_transport_result(success, time::OffsetDateTime::UNIX_EPOCH)
|
|
.expect("complete");
|
|
assert!(
|
|
coordinator
|
|
.dead_repo_blacklist()
|
|
.expect("blacklist")
|
|
.is_empty()
|
|
);
|
|
}
|
|
}
|