Complete M7.5 durable batch outcomes and commit resolution
ADR-0063: manifest v2 (360 bytes) binds a per-tick validation trace object and a canonical decision block by SHA-256. Publication replays the trace from the committed predecessor's verified payload and requires every report and the candidate bytes, derives deaths deterministically, uploads payload and trace with verified read-back, then sends manifest and decisions to one fenced statement. RegionBatch seals at a terminal transition or 64 ticks and refuses ticks until durable confirmation. Schema v5 records the checkpoint, its decisions and outbox rows atomically; a unique index keeps one terminal outcome per actor (EVL08), triggers make commits/decisions immutable (EVL09), an exact retry of a committed ID returns the original receipt, changed bytes conflict, and read-only resolve_region_commit answers committed/absent/ conflict. Outbox pending/acknowledge give idempotent at-least-once delivery. Recovery reconciles durable deaths with restored lifecycle (and archive deaths for the lineage profile). Validated: Windows CTest 58/58; Linux GCC Debug/Release, clang-tidy and ASan/UBSan CTest 58/58 each; PostgreSQL suites Release, analysis, ASan (lost response, committed-after-timeout, terminated orphan retry, changed bytes, repeated/partial/reordered decisions, gap, outbox redelivery, history quarantine); five mutations each caught. No Docker/Gitea CI, process kill, broker or deployment. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
@@ -7,10 +7,12 @@
|
||||
#include <cstdlib>
|
||||
#include <filesystem>
|
||||
#include <fstream>
|
||||
#include <optional>
|
||||
#include <source_location>
|
||||
#include <stdexcept>
|
||||
#include <string>
|
||||
#include <sys/stat.h>
|
||||
#include <thread>
|
||||
#include <unistd.h>
|
||||
|
||||
void run_region_checkpoint_store_tests();
|
||||
@@ -172,6 +174,15 @@ std::size_t count_objects(const fs::path& dir) {
|
||||
return n;
|
||||
}
|
||||
|
||||
using EnvBatch = RegionBatch<evo::processes::EnvironmentalCommunity>;
|
||||
using LineageBatch = RegionBatch<evo::processes::LineageRegion>;
|
||||
// Runs up to `ticks` more ticks (stopping early if an outcome seals it), then flushes.
|
||||
template <typename Batch> void run_batch(Batch& b, std::uint64_t ticks, bool deaths = false) {
|
||||
for (std::uint64_t i = 0; i < ticks && !b.sealed(); ++i)
|
||||
CHECK(b.advance(evo::test::region_input(b.end_tick() + 1, deaths)));
|
||||
b.seal();
|
||||
}
|
||||
|
||||
void publish_recover_and_continue() {
|
||||
auto r = provisioned("publish", 3, 0xC0);
|
||||
const std::array<Sha256Digest, 1> retained{build};
|
||||
@@ -180,7 +191,7 @@ void publish_recover_and_continue() {
|
||||
// Genesis recovers to the exact opening state.
|
||||
auto recovered =
|
||||
recover_environmental_checkpoint(*r.runtime, *r.objects, r.key, retained, budget, 5s);
|
||||
CHECK(recovered && recovered->state->progress().tick == 0);
|
||||
CHECK(recovered && recovered->state->progress().tick == 0 && recovered->decisions.empty());
|
||||
auto head = recovered->committed;
|
||||
|
||||
// Genesis cannot be re-provisioned or published over.
|
||||
@@ -189,31 +200,55 @@ void publish_recover_and_continue() {
|
||||
filled<16>(0xC9), build, budget, 5s));
|
||||
|
||||
auto original = evo::test::region_community();
|
||||
EnvBatch batch{*original};
|
||||
// An unsealed batch is still being simulated: refused before any upload.
|
||||
CHECK(batch.advance(evo::test::region_input(1)));
|
||||
const auto before = count_objects(r.objects->directory());
|
||||
auto open = publish_environmental_batch(*r.runtime, *r.objects, *lease, head, batch,
|
||||
filled<16>(0xCA), build, budget, 5s);
|
||||
CHECK(!open && open.error().kind == PublicationError::candidate);
|
||||
CHECK(count_objects(r.objects->directory()) == before);
|
||||
std::uint8_t seed = 0xC1;
|
||||
for (int batch = 0; batch < 3; ++batch) {
|
||||
evo::test::advance_region(*original, batch == 1 ? 64 : 16);
|
||||
for (int n = 0; n < 3; ++n) {
|
||||
run_batch(batch, n == 1 ? 64 : 15);
|
||||
const auto started = std::chrono::steady_clock::now();
|
||||
auto published = publish_environmental_checkpoint(
|
||||
*r.runtime, *r.objects, *lease, head, *original, filled<16>(seed++), build, budget, 5s);
|
||||
auto published = publish_environmental_batch(*r.runtime, *r.objects, *lease, head, batch,
|
||||
filled<16>(seed++), build, budget, 5s);
|
||||
const auto us = std::chrono::duration_cast<std::chrono::microseconds>(
|
||||
std::chrono::steady_clock::now() - started)
|
||||
.count();
|
||||
CHECK(published && published->head.sequence == head.head.sequence + 1);
|
||||
CHECK(published->head.end_tick == original->progress().tick);
|
||||
std::printf(
|
||||
"checkpoint_publish_us batch=%d ticks=%llu payload_bytes=%llu us=%lld\n", batch,
|
||||
static_cast<unsigned long long>(published->head.end_tick - head.head.end_tick),
|
||||
static_cast<unsigned long long>(decode_manifest(published->manifest)->payload_bytes),
|
||||
static_cast<long long>(us));
|
||||
head = std::move(*published);
|
||||
CHECK(published && published->committed.head.sequence == head.head.sequence + 1);
|
||||
CHECK(published->committed.head.end_tick == original->progress().tick);
|
||||
CHECK(published->decisions.empty());
|
||||
const auto m = *decode_manifest(published->committed.manifest);
|
||||
CHECK(m.manifest_version == 2 &&
|
||||
m.batch.trace_bytes == trace_bytes_for(m.end_tick - m.begin_tick));
|
||||
std::printf("batch_publish_us batch=%d ticks=%llu payload_bytes=%llu us=%lld\n", n,
|
||||
static_cast<unsigned long long>(m.end_tick - m.begin_tick),
|
||||
static_cast<unsigned long long>(m.payload_bytes), static_cast<long long>(us));
|
||||
CHECK(batch.confirm(m.end_tick));
|
||||
head = std::move(published->committed);
|
||||
// The trace object is retained, bound to this link of the chain.
|
||||
auto trace = r.objects->get({m.batch.trace_sha256, m.batch.trace_bytes});
|
||||
auto decoded = trace ? decode_trace(*trace) : std::unexpected{BatchError::format};
|
||||
CHECK(decoded && decoded->header.begin_tick == m.begin_tick &&
|
||||
decoded->header.predecessor_digest == m.predecessor_digest);
|
||||
}
|
||||
// 65 ticks is more than one batch: refused before upload.
|
||||
const auto before = count_objects(r.objects->directory());
|
||||
evo::test::advance_region(*original, 65);
|
||||
auto oversized = publish_environmental_checkpoint(
|
||||
*r.runtime, *r.objects, *lease, head, *original, filled<16>(0xCA), build, budget, 5s);
|
||||
CHECK(!oversized && oversized.error().kind == PublicationError::candidate);
|
||||
CHECK(count_objects(r.objects->directory()) == before);
|
||||
// State advanced outside the recorded batch cannot be published: the replay of
|
||||
// the recorded inputs no longer reproduces it.
|
||||
auto drifted = evo::test::region_community();
|
||||
EnvBatch drift{*drifted};
|
||||
run_batch(drift, 3);
|
||||
CHECK(drifted->advance(4, evo::test::region_input(4)));
|
||||
auto diverged =
|
||||
publish_environmental_batch(*r.runtime, *r.objects, *lease, recovered->committed, drift,
|
||||
filled<16>(0xCD), build, budget, 5s);
|
||||
CHECK(!diverged && diverged.error().kind == PublicationError::candidate);
|
||||
// A batch replayed from the wrong predecessor cannot reproduce its state.
|
||||
run_batch(batch, 4);
|
||||
auto stale = publish_environmental_batch(*r.runtime, *r.objects, *lease, recovered->committed,
|
||||
batch, filled<16>(0xCB), build, budget, 5s);
|
||||
CHECK(!stale && stale.error().kind == PublicationError::candidate);
|
||||
|
||||
// A new process recovers the committed head; its trajectory matches the original's.
|
||||
auto second = FilesystemObjectStore::open(r.objects->directory().parent_path().parent_path(),
|
||||
@@ -222,11 +257,11 @@ void publish_recover_and_continue() {
|
||||
auto restored =
|
||||
recover_environmental_checkpoint(*r.runtime, **second, r.key, retained, budget, 5s);
|
||||
CHECK(restored && restored->committed.head == head.head);
|
||||
CHECK(restored->state->progress().tick == 96);
|
||||
CHECK(restored->state->progress().tick == 95);
|
||||
auto reference = evo::test::region_community();
|
||||
evo::test::advance_region(*reference, 96);
|
||||
evo::test::advance_region(*reference, 95);
|
||||
CHECK(reference->snapshot() == restored->state->snapshot());
|
||||
for (std::uint64_t tick = 97; tick <= 140; ++tick) {
|
||||
for (std::uint64_t tick = 96; tick <= 140; ++tick) {
|
||||
auto a = reference->advance(tick, evo::test::region_input(tick, true));
|
||||
auto b = restored->state->advance(tick, evo::test::region_input(tick, true));
|
||||
CHECK(a && b && *a == *b);
|
||||
@@ -238,11 +273,9 @@ void publish_recover_and_continue() {
|
||||
auto foreign =
|
||||
recover_environmental_checkpoint(*r.runtime, *r.objects, r.key, other, budget, 5s);
|
||||
CHECK(!foreign && foreign.error().kind == RecoveryError::unsupported);
|
||||
// Unretained build is refused by the database; the upload stays an orphan.
|
||||
auto next = evo::test::region_community();
|
||||
evo::test::advance_region(*next, 100);
|
||||
auto refused = publish_environmental_checkpoint(*r.runtime, *r.objects, *lease, head, *next,
|
||||
filled<16>(0xCB), filled<32>(0xB2), budget, 5s);
|
||||
// Unretained build is refused by the database; the uploads stay orphans.
|
||||
auto refused = publish_environmental_batch(*r.runtime, *r.objects, *lease, head, batch,
|
||||
filled<16>(0xCC), filled<32>(0xB2), budget, 5s);
|
||||
CHECK(!refused && refused.error().kind == PublicationError::rejected &&
|
||||
refused.error().store.kind == LeaseStoreError::unsupported);
|
||||
CHECK(release_region_lease(*r.runtime, r.key, *lease, 5s));
|
||||
@@ -256,32 +289,34 @@ void failures_leave_the_previous_head() {
|
||||
auto head = load_region_checkpoint(*r.runtime, r.key, 5s);
|
||||
CHECK(head);
|
||||
auto state = evo::test::region_community();
|
||||
evo::test::advance_region(*state, 8);
|
||||
auto first = publish_environmental_checkpoint(*r.runtime, *r.objects, *lease, *head, *state,
|
||||
filled<16>(0xD1), build, budget, 5s);
|
||||
CHECK(first);
|
||||
const auto manifest = *decode_manifest(first->manifest);
|
||||
EnvBatch batch{*state};
|
||||
run_batch(batch, 8);
|
||||
auto first = publish_environmental_batch(*r.runtime, *r.objects, *lease, *head, batch,
|
||||
filled<16>(0xD1), build, budget, 5s);
|
||||
CHECK(first && batch.confirm(8));
|
||||
const auto manifest = *decode_manifest(first->committed.manifest);
|
||||
const auto object = r.objects->directory() / hex(manifest.payload_sha256);
|
||||
|
||||
// Upload failure: no manifest is sent; the head does not move.
|
||||
evo::test::advance_region(*state, 8);
|
||||
run_batch(batch, 8);
|
||||
if (::geteuid() != 0) {
|
||||
fs::permissions(r.objects->directory(), fs::perms::owner_read | fs::perms::owner_exec,
|
||||
fs::perm_options::replace);
|
||||
auto upload = publish_environmental_checkpoint(*r.runtime, *r.objects, *lease, *first,
|
||||
*state, filled<16>(0xD2), build, budget, 5s);
|
||||
auto upload = publish_environmental_batch(*r.runtime, *r.objects, *lease, first->committed,
|
||||
batch, filled<16>(0xD2), build, budget, 5s);
|
||||
fs::permissions(r.objects->directory(), fs::perms::owner_all, fs::perm_options::replace);
|
||||
CHECK(!upload && upload.error().kind == PublicationError::upload);
|
||||
CHECK(load_region_head(*r.runtime, r.key, 5s)->sequence == 1);
|
||||
}
|
||||
|
||||
// Manifest transaction failure after a verified upload: fenced writer, orphan retained.
|
||||
// Manifest transaction failure after verified uploads: fenced writer, orphans retained.
|
||||
auto stale = *lease;
|
||||
stale.holder[0] ^= 1U;
|
||||
auto fenced = publish_environmental_checkpoint(*r.runtime, *r.objects, stale, *first, *state,
|
||||
filled<16>(0xD3), build, budget, 5s);
|
||||
auto fenced = publish_environmental_batch(*r.runtime, *r.objects, stale, first->committed,
|
||||
batch, filled<16>(0xD3), build, budget, 5s);
|
||||
CHECK(!fenced && fenced.error().kind == PublicationError::rejected &&
|
||||
fenced.error().store.kind == LeaseStoreError::fenced);
|
||||
CHECK(fenced.error().prepared.commit == filled<16>(0xD3));
|
||||
auto orphan_bytes = state->snapshot();
|
||||
auto orphan = r.objects->get({*sha256(*orphan_bytes), orphan_bytes->size()});
|
||||
CHECK(orphan && *orphan == *orphan_bytes); // Uploaded, verified, unreferenced.
|
||||
@@ -296,6 +331,11 @@ void failures_leave_the_previous_head() {
|
||||
auto missing =
|
||||
recover_environmental_checkpoint(*r.runtime, *r.objects, r.key, retained, budget, 5s);
|
||||
CHECK(!missing && missing.error().kind == RecoveryError::missing_payload);
|
||||
// Staging replays from the committed predecessor bytes; without them it stops.
|
||||
auto no_base = publish_environmental_batch(*r.runtime, *r.objects, *lease, first->committed,
|
||||
batch, filled<16>(0xD4), build, budget, 5s);
|
||||
CHECK(!no_base && no_base.error().kind == PublicationError::predecessor &&
|
||||
no_base.error().object == ObjectError::missing);
|
||||
CHECK(r.objects->put(*saved));
|
||||
fs::permissions(object, fs::perms::owner_all, fs::perm_options::replace);
|
||||
{
|
||||
@@ -310,18 +350,26 @@ void failures_leave_the_previous_head() {
|
||||
CHECK(r.objects->put(*saved));
|
||||
CHECK(recover_environmental_checkpoint(*r.runtime, *r.objects, r.key, retained, budget, 5s));
|
||||
|
||||
// The database refuses tampering with committed manifests even by the owner.
|
||||
// Committed history is immutable even to the owner (trigger), and the head
|
||||
// cannot be repointed (composite foreign key).
|
||||
auto tamper = r.owner->query("UPDATE evolution.region_commits SET manifest = "
|
||||
"overlay(manifest placing '\\x02'::bytea from 121 for 1) "
|
||||
"WHERE world_id = 7 AND region_id = 4 AND commit_sequence = 1",
|
||||
{}, 5s);
|
||||
CHECK(!tamper && std::string(tamper.error().sqlstate.data()) == "23514");
|
||||
CHECK(!tamper && std::string(tamper.error().sqlstate.data()) == "EVL09");
|
||||
auto erase = r.owner->query("DELETE FROM evolution.region_commits WHERE world_id = 7 AND "
|
||||
"region_id = 4 AND commit_sequence = 1",
|
||||
{}, 5s);
|
||||
CHECK(!erase && std::string(erase.error().sqlstate.data()) == "EVL09");
|
||||
auto repoint = r.owner->query("UPDATE evolution.region_heads SET digest = sha256('\\x00') "
|
||||
"WHERE world_id = 7 AND region_id = 4",
|
||||
{}, 5s);
|
||||
CHECK(!repoint && std::string(repoint.error().sqlstate.data()) == "23503");
|
||||
// Defense in depth: even if that constraint were lost, recovery re-digests the
|
||||
// manifest and quarantines. Restores the constraint before continuing.
|
||||
// Defense in depth: even with the trigger and digest constraint removed, recovery
|
||||
// re-digests the manifest and quarantines. Both guards are restored afterwards.
|
||||
CHECK(r.owner->query("ALTER TABLE evolution.region_commits DISABLE TRIGGER "
|
||||
"region_commits_immutable",
|
||||
{}, 5s));
|
||||
CHECK(r.owner->query("ALTER TABLE evolution.region_commits DROP CONSTRAINT "
|
||||
"region_commits_manifest_digest",
|
||||
{}, 5s));
|
||||
@@ -333,13 +381,16 @@ void failures_leave_the_previous_head() {
|
||||
auto tampered =
|
||||
recover_environmental_checkpoint(*r.runtime, *r.objects, r.key, retained, budget, 5s);
|
||||
CHECK(!tampered && tampered.error().kind == RecoveryError::manifest);
|
||||
const std::array<std::string, 1> original{"\\x" + hex(first->manifest)};
|
||||
const std::array<std::string, 1> original{"\\x" + hex(first->committed.manifest)};
|
||||
CHECK(r.owner->query("UPDATE evolution.region_commits SET manifest = $1::bytea "
|
||||
"WHERE world_id = 7 AND region_id = 4 AND commit_sequence = 1",
|
||||
original, 5s));
|
||||
CHECK(r.owner->query("ALTER TABLE evolution.region_commits ADD CONSTRAINT "
|
||||
"region_commits_manifest_digest CHECK (commit_digest = sha256(manifest))",
|
||||
{}, 5s));
|
||||
CHECK(r.owner->query("ALTER TABLE evolution.region_commits ENABLE TRIGGER "
|
||||
"region_commits_immutable",
|
||||
{}, 5s));
|
||||
CHECK(recover_environmental_checkpoint(*r.runtime, *r.objects, r.key, retained, budget, 5s));
|
||||
|
||||
// Recovery does not create genesis for an unknown region.
|
||||
@@ -347,6 +398,7 @@ void failures_leave_the_previous_head() {
|
||||
recover_environmental_checkpoint(*r.runtime, *r.objects, {7, 99}, retained, budget, 5s);
|
||||
CHECK(!none && none.error().kind == RecoveryError::not_provisioned);
|
||||
}
|
||||
|
||||
// History-capable profile: deaths commit with the state that caused them, the
|
||||
// archive survives recovery, and a region never switches profile.
|
||||
void lineage_publish_and_recover() {
|
||||
@@ -369,50 +421,260 @@ void lineage_publish_and_recover() {
|
||||
CHECK(lease);
|
||||
|
||||
auto region = evo::test::lineage_region();
|
||||
evo::test::advance_lineage(*region, 24, true); // Consumer dies at tick 20.
|
||||
auto first = publish_lineage_checkpoint(*runtime, **objects, *lease, *genesis, *region,
|
||||
filled<16>(0xE1), build, budget, 5s);
|
||||
CHECK(first && first->head.end_tick == 24);
|
||||
LineageBatch batch{*region};
|
||||
run_batch(batch, 64, true); // The consumer's death at tick 20 seals the batch there.
|
||||
CHECK(batch.end_tick() == 20);
|
||||
auto first = publish_lineage_batch(*runtime, **objects, *lease, *genesis, batch,
|
||||
filled<16>(0xE1), build, budget, 5s);
|
||||
CHECK(first && first->committed.head.end_tick == 20);
|
||||
CHECK(first->decisions.size() == 1 &&
|
||||
first->decisions[0] == BatchDecision{20, Actor::consumer, Outcome::death, 11});
|
||||
CHECK(batch.confirm(20));
|
||||
// The environmental path refuses a lineage predecessor before uploading.
|
||||
auto plain = evo::test::region_community();
|
||||
evo::test::advance_region(*plain, 30);
|
||||
auto crossed = publish_environmental_checkpoint(*runtime, **objects, *lease, *first, *plain,
|
||||
filled<16>(0xE2), build, budget, 5s);
|
||||
EnvBatch plain_batch{*plain};
|
||||
run_batch(plain_batch, 30);
|
||||
auto crossed = publish_environmental_batch(*runtime, **objects, *lease, first->committed,
|
||||
plain_batch, filled<16>(0xE2), build, budget, 5s);
|
||||
CHECK(!crossed && crossed.error().kind == PublicationError::candidate);
|
||||
// The database refuses a profile switch even when the manifest is well formed.
|
||||
auto switched = *decode_manifest(first->manifest);
|
||||
auto next = successor_manifest(switched, first->head.digest, filled<16>(0xE3), *lease, 30,
|
||||
build, 100, filled<32>(0xA2));
|
||||
auto switched = *decode_manifest(first->committed.manifest);
|
||||
auto next = successor_manifest(switched, first->committed.head.digest, filled<16>(0xE3), *lease,
|
||||
30, build, 100, filled<32>(0xA2),
|
||||
{trace_bytes_for(10), filled<32>(0x7C), 0, {}});
|
||||
CHECK(next);
|
||||
next->profile = CheckpointProfile::environmental_v1;
|
||||
next->history_profile = 0;
|
||||
auto switched_bytes = encode_manifest(*next);
|
||||
CHECK(switched_bytes);
|
||||
auto refused = publish_region_checkpoint(*runtime, *switched_bytes, 5s);
|
||||
auto refused = publish_region_checkpoint(*runtime, *switched_bytes, {}, 5s);
|
||||
CHECK(!refused && refused.error().kind == LeaseStoreError::unsupported);
|
||||
|
||||
const std::array<Sha256Digest, 1> retained{build};
|
||||
auto recovered = recover_lineage_checkpoint(*runtime, **objects, key, retained, budget, 5s);
|
||||
CHECK(recovered && recovered->committed.head == first->head);
|
||||
CHECK(recovered && recovered->committed.head == first->committed.head);
|
||||
CHECK(recovered->decisions.size() == 1 && recovered->decisions[0].commit_sequence == 1);
|
||||
const auto consumer = recovered->state->archive().organism(11);
|
||||
CHECK(consumer && !consumer->alive && consumer->died.value == 20);
|
||||
CHECK(recovered->state->archive().ancestry(2)->size() == 2); // Retained ancestry.
|
||||
CHECK(recovered->state->snapshot() == region->snapshot());
|
||||
for (std::uint64_t tick = 25; tick <= 60; ++tick) {
|
||||
LineageBatch resumed{*recovered->state};
|
||||
for (std::uint64_t tick = 21; tick <= 60; ++tick) {
|
||||
auto a = region->advance(tick, evo::test::region_input(tick));
|
||||
auto b = recovered->state->advance(tick, evo::test::region_input(tick));
|
||||
auto b = resumed.advance(evo::test::region_input(tick));
|
||||
CHECK(a && b && *a == *b);
|
||||
}
|
||||
CHECK(recovered->state->snapshot() == region->snapshot());
|
||||
auto second =
|
||||
publish_lineage_checkpoint(*runtime, **objects, *lease, recovered->committed,
|
||||
*recovered->state, filled<16>(0xE4), build, budget, 5s);
|
||||
CHECK(second && second->head.sequence == 2 && second->head.end_tick == 60);
|
||||
resumed.seal();
|
||||
auto second = publish_lineage_batch(*runtime, **objects, *lease, recovered->committed, resumed,
|
||||
filled<16>(0xE4), build, budget, 5s);
|
||||
CHECK(second && second->committed.head.sequence == 2 && second->committed.head.end_tick == 60 &&
|
||||
second->decisions.empty());
|
||||
// A host restoring the other profile refuses rather than misreading the payload.
|
||||
auto wrong = recover_environmental_checkpoint(*runtime, **objects, key, retained, budget, 5s);
|
||||
CHECK(!wrong && wrong.error().kind == RecoveryError::unsupported);
|
||||
CHECK(release_region_lease(*runtime, key, *lease, 5s));
|
||||
}
|
||||
|
||||
std::string runtime_user() {
|
||||
return env("EVO_PG_RUNTIME_USER");
|
||||
}
|
||||
// Polls resolution on a fresh connection until the prepared commit is committed.
|
||||
ResolvedCommit await_resolution(const PreparedCommit& prepared, Resolution wanted) {
|
||||
auto db = connect_as(runtime_user());
|
||||
for (int attempt = 0; attempt < 200; ++attempt) {
|
||||
auto resolved = resolve_prepared(*db, prepared, 5s);
|
||||
CHECK(resolved);
|
||||
if (resolved->status == wanted)
|
||||
return *resolved;
|
||||
std::this_thread::sleep_for(25ms);
|
||||
}
|
||||
CHECK(false);
|
||||
return {};
|
||||
}
|
||||
// Holds the lease row lock so a publication's outcome becomes ambiguous to its client.
|
||||
struct LeaseRowLock {
|
||||
std::unique_ptr<Postgres> db;
|
||||
LeaseRowLock(RegionKey key) : db(connect_as(env("PGUSER"))) {
|
||||
const std::array<std::string, 2> p{std::to_string(key.world), std::to_string(key.region)};
|
||||
CHECK(db->query("BEGIN", {}, 5s));
|
||||
CHECK(db->query("SELECT 1 FROM evolution.region_leases WHERE world_id = $1::evolution.u64 "
|
||||
"AND region_id = $2::evolution.u64 FOR UPDATE",
|
||||
p, 5s));
|
||||
}
|
||||
// Terminates backends of the runtime login queued behind this lock.
|
||||
void kill_waiters() {
|
||||
const std::array<std::string, 1> p{runtime_user()};
|
||||
CHECK(db->query("SELECT pg_terminate_backend(pid) FROM pg_stat_activity WHERE "
|
||||
"usename = $1 AND wait_event_type = 'Lock'",
|
||||
p, 5s));
|
||||
}
|
||||
void release() {
|
||||
CHECK(db->query("ROLLBACK", {}, 5s));
|
||||
}
|
||||
};
|
||||
|
||||
// M7.5 durability: permanent outcomes commit atomically with their checkpoint,
|
||||
// ambiguity is resolved by exact bytes, and nothing is ever applied twice.
|
||||
void durable_outcomes_and_ambiguity() {
|
||||
auto r = provisioned("outcomes", 7, 0xF0);
|
||||
const std::array<Sha256Digest, 1> retained{build};
|
||||
auto lease = acquire_region_lease(*r.runtime, r.key, 20000ms, 5s);
|
||||
CHECK(lease);
|
||||
auto genesis = load_region_checkpoint(*r.runtime, r.key, 5s);
|
||||
CHECK(genesis);
|
||||
auto state = evo::test::region_community();
|
||||
EnvBatch batch{*state};
|
||||
run_batch(batch, 64, true); // Sealed by the consumer's death at tick 20.
|
||||
CHECK(batch.end_tick() == 20);
|
||||
auto died = publish_environmental_batch(*r.runtime, *r.objects, *lease, *genesis, batch,
|
||||
filled<16>(0xF1), build, budget, 5s);
|
||||
CHECK(died && died->decisions.size() == 1 &&
|
||||
died->decisions[0] == BatchDecision{20, Actor::consumer, Outcome::death, 0});
|
||||
CHECK(batch.confirm(20));
|
||||
const CommittedDecision final_death{1, died->decisions[0]};
|
||||
auto decisions = load_region_decisions(*r.runtime, r.key, 5s);
|
||||
CHECK(decisions && decisions->size() == 1 && (*decisions)[0] == final_death);
|
||||
|
||||
// Outbox: at-least-once delivery, idempotent acknowledgement, no simulation effect.
|
||||
auto pending = pending_region_outbox(*r.runtime, r.key, 64, 5s);
|
||||
CHECK(pending && pending->size() == 1 && (*pending)[0] == final_death);
|
||||
CHECK(pending_region_outbox(*r.runtime, r.key, 64, 5s)->size() == 1); // Redelivered.
|
||||
auto ack = acknowledge_region_outbox(*r.runtime, r.key, died->decisions[0], 5s);
|
||||
CHECK(ack && *ack);
|
||||
ack = acknowledge_region_outbox(*r.runtime, r.key, died->decisions[0], 5s);
|
||||
CHECK(ack && !*ack);
|
||||
CHECK(pending_region_outbox(*r.runtime, r.key, 64, 5s)->empty());
|
||||
CHECK(!pending_region_outbox(*r.runtime, r.key, 65, 5s));
|
||||
const BatchDecision never{21, Actor::producer, Outcome::death, 0};
|
||||
CHECK(!acknowledge_region_outbox(*r.runtime, r.key, never, 5s));
|
||||
|
||||
// Commit succeeded but the response was lost: the exact retry returns the same
|
||||
// receipt and records nothing twice; resolution confirms it under any epoch.
|
||||
auto death_block = encode_decisions(died->decisions, 0, 20);
|
||||
CHECK(death_block);
|
||||
const PreparedCommit lost{r.key, filled<16>(0xF1), died->committed.manifest, *death_block};
|
||||
auto retried = retry_prepared(*r.runtime, lost, 5s);
|
||||
CHECK(retried && *retried == died->committed.head);
|
||||
auto resolved = resolve_prepared(*r.runtime, lost, 5s);
|
||||
CHECK(resolved && resolved->status == Resolution::committed &&
|
||||
resolved->receipt == died->committed.head);
|
||||
CHECK(load_region_decisions(*r.runtime, r.key, 5s)->size() == 1);
|
||||
// Changed bytes under the same ID conflict, whether manifest or decisions differ.
|
||||
auto changed = lost;
|
||||
changed.manifest[288] ^= 1U; // Trace digest.
|
||||
CHECK(retry_prepared(*r.runtime, changed, 5s).error().kind == LeaseStoreError::commit_conflict);
|
||||
CHECK(resolve_prepared(*r.runtime, changed, 5s)->status == Resolution::conflict);
|
||||
auto undecided = lost;
|
||||
undecided.decisions.clear(); // No longer what the manifest digests.
|
||||
CHECK(retry_prepared(*r.runtime, undecided, 5s).error().kind == LeaseStoreError::invalid);
|
||||
CHECK(resolve_prepared(*r.runtime, undecided, 5s)->status == Resolution::conflict);
|
||||
auto elsewhere = lost;
|
||||
elsewhere.key.region = 3;
|
||||
CHECK(resolve_prepared(*r.runtime, elsewhere, 5s)->status == Resolution::conflict);
|
||||
auto unknown_id = lost;
|
||||
unknown_id.commit = filled<16>(0xFE);
|
||||
CHECK(resolve_prepared(*r.runtime, unknown_id, 5s)->status == Resolution::absent);
|
||||
|
||||
// The committed dead stay dead: a later batch cannot finalize the death again.
|
||||
const auto head = died->committed;
|
||||
const std::array again{BatchDecision{25, Actor::consumer, Outcome::death, 0}};
|
||||
auto again_bytes = encode_decisions(again, 20, 30);
|
||||
CHECK(again_bytes);
|
||||
auto relapse = successor_manifest(
|
||||
*decode_manifest(head.manifest), head.head.digest, filled<16>(0xF2), *lease, 30, build, 100,
|
||||
filled<32>(0xA2), {trace_bytes_for(10), filled<32>(0x7C), 1, *sha256(*again_bytes)});
|
||||
CHECK(relapse);
|
||||
auto relapse_bytes = encode_manifest(*relapse);
|
||||
auto repeated = publish_region_checkpoint(*r.runtime, *relapse_bytes, *again_bytes, 5s);
|
||||
CHECK(!repeated && repeated.error().kind == LeaseStoreError::outcome_final);
|
||||
// A block of the right size that is not the one the manifest digests is refused.
|
||||
auto forged_block = *again_bytes;
|
||||
forged_block[0] ^= 1U; // Tick 25 -> 24: still well formed, different bytes.
|
||||
CHECK(publish_region_checkpoint(*r.runtime, *relapse_bytes, forged_block, 5s).error().kind ==
|
||||
LeaseStoreError::invalid);
|
||||
// Partial or reordered decision blocks are refused whole; nothing is recorded.
|
||||
CHECK(publish_region_checkpoint(*r.runtime, *relapse_bytes, {}, 5s).error().kind ==
|
||||
LeaseStoreError::invalid);
|
||||
const std::array two{BatchDecision{25, Actor::producer, Outcome::death, 0},
|
||||
BatchDecision{24, Actor::consumer, Outcome::death, 0}};
|
||||
std::vector<std::uint8_t> reordered(two.size() * decision_bytes);
|
||||
auto ordered = encode_decisions(std::array{two[1], two[0]}, 20, 30);
|
||||
CHECK(ordered);
|
||||
std::copy_n(ordered->begin() + 24, 24, reordered.begin());
|
||||
std::copy_n(ordered->begin(), 24, reordered.begin() + 24);
|
||||
auto reorder = *relapse;
|
||||
reorder.batch = {trace_bytes_for(10), filled<32>(0x7C), 2, *sha256(reordered)};
|
||||
CHECK(publish_region_checkpoint(*r.runtime, *encode_manifest(reorder), reordered, 5s)
|
||||
.error()
|
||||
.kind == LeaseStoreError::invalid);
|
||||
// A gap: a link from an older predecessor is not the committed head's successor.
|
||||
auto gap = successor_manifest(*decode_manifest(genesis->manifest), genesis->head.digest,
|
||||
filled<16>(0xF3), *lease, 30, build, 100, filled<32>(0xA2),
|
||||
{trace_bytes_for(30), filled<32>(0x7C), 0, {}});
|
||||
CHECK(publish_region_checkpoint(*r.runtime, *encode_manifest(*gap), {}, 5s).error().kind ==
|
||||
LeaseStoreError::head_mismatch);
|
||||
CHECK(load_region_head(*r.runtime, r.key, 5s)->sequence == 1);
|
||||
CHECK(load_region_decisions(*r.runtime, r.key, 5s)->size() == 1);
|
||||
|
||||
// Keep authority comfortably valid for the ambiguity cases (sanitizer builds are slow).
|
||||
CHECK(renew_region_lease(*r.runtime, r.key, *lease, 20000ms, 5s));
|
||||
// Ambiguous outcome that did commit: the client times out while queued, the
|
||||
// server finishes after the lock is released. Resolution adopts the receipt.
|
||||
run_batch(batch, 10);
|
||||
std::optional<LeaseRowLock> lock{std::in_place, r.key};
|
||||
auto ambiguous_db = connect_as(runtime_user());
|
||||
auto ambiguous = publish_environmental_batch(*ambiguous_db, *r.objects, *lease, head, batch,
|
||||
filled<16>(0xF4), build, budget, 300ms);
|
||||
CHECK(!ambiguous && ambiguous.error().kind == PublicationError::unknown);
|
||||
CHECK(batch.sealed()); // Still paused: never assume rollback.
|
||||
lock->release();
|
||||
lock.reset();
|
||||
const auto committed = await_resolution(ambiguous.error().prepared, Resolution::committed);
|
||||
CHECK(committed.receipt.sequence == 2 && committed.receipt.end_tick == 30);
|
||||
auto adopted = retry_prepared(*r.runtime, ambiguous.error().prepared, 5s);
|
||||
CHECK(adopted && *adopted == committed.receipt);
|
||||
CHECK(batch.confirm(30));
|
||||
|
||||
// Ambiguous outcome that did not commit: the orphaned request is terminated
|
||||
// before it reaches the row. Absent, so the same bytes are retried while authorized.
|
||||
auto after = load_region_checkpoint(*r.runtime, r.key, 5s);
|
||||
CHECK(after);
|
||||
run_batch(batch, 5);
|
||||
lock.emplace(r.key);
|
||||
auto orphan_db = connect_as(runtime_user());
|
||||
auto orphaned = publish_environmental_batch(*orphan_db, *r.objects, *lease, *after, batch,
|
||||
filled<16>(0xF5), build, budget, 300ms);
|
||||
CHECK(!orphaned && orphaned.error().kind == PublicationError::unknown);
|
||||
lock->kill_waiters();
|
||||
lock->release();
|
||||
lock.reset();
|
||||
std::this_thread::sleep_for(200ms);
|
||||
CHECK(resolve_prepared(*r.runtime, orphaned.error().prepared, 5s)->status ==
|
||||
Resolution::absent);
|
||||
auto recommitted = retry_prepared(*r.runtime, orphaned.error().prepared, 5s);
|
||||
CHECK(recommitted && recommitted->sequence == 3 && recommitted->end_tick == 35);
|
||||
CHECK(batch.confirm(35));
|
||||
|
||||
// Recovery reconciles durable outcomes with the restored lifecycle.
|
||||
auto recovered =
|
||||
recover_environmental_checkpoint(*r.runtime, *r.objects, r.key, retained, budget, 5s);
|
||||
CHECK(recovered && recovered->decisions.size() == 1 && recovered->decisions[0] == final_death &&
|
||||
recovered->state->progress().consumer_terminal == 1);
|
||||
CHECK(recovered->state->snapshot() == state->snapshot());
|
||||
// A durable death the state does not show is a history disagreement: quarantine.
|
||||
CHECK(r.owner->query("INSERT INTO evolution.region_decisions(world_id, region_id, outcome, "
|
||||
"actor, tick, organism, commit_sequence) VALUES (7, 7, 1, 0, 30, 0, 2)",
|
||||
{}, 5s));
|
||||
auto disagreed =
|
||||
recover_environmental_checkpoint(*r.runtime, *r.objects, r.key, retained, budget, 5s);
|
||||
CHECK(!disagreed && disagreed.error().kind == RecoveryError::history);
|
||||
auto undo = r.owner->query("DELETE FROM evolution.region_decisions WHERE world_id = 7 AND "
|
||||
"region_id = 7 AND actor = 0",
|
||||
{}, 5s);
|
||||
CHECK(!undo && std::string(undo.error().sqlstate.data()) == "EVL09"); // Stays quarantined.
|
||||
CHECK(release_region_lease(*r.runtime, r.key, *lease, 5s));
|
||||
}
|
||||
} // namespace
|
||||
|
||||
void run_region_checkpoint_store_tests() {
|
||||
@@ -421,4 +683,5 @@ void run_region_checkpoint_store_tests() {
|
||||
publish_recover_and_continue();
|
||||
failures_leave_the_previous_head();
|
||||
lineage_publish_and_recover();
|
||||
durable_outcomes_and_ambiguity();
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user