merge(chv-hybrid): 合入 B 轨——租约 + 排队 + 升级提示 + 网关门

合入 feat-chv-hybrid-b (667a427)。executor.py 两处内容冲突按 hunk
解决,全部取 B 侧;合入时发现 merge 复制出第二个 suspend_after_task
定义(A 版截断体),已删除。

集成统一(A/B 语义冲突三处,设计 §4.1 的 state 模型为准):
- state 模型:A 的 flag 模型(resident: True + legacy state)与 B
  的 state 模型(state ∈ resident-active/paused/hibernated/demoted)
  统一为 state 模型;_mark_resident_record 标记常驻时同时置
  resident-active,_ensure_resident_capacity 同时认 flag 与 state。
- 容量等待:A 的 bool 版 _wait_for_capacity 改名 _poll_capacity,
  B 的 raise 版保留原名;_wait_for_resident_slot 豁免已有常驻登记
  的租户(它已占一个槽位,再等等于把自己的槽位算成满——A 轨
  test_resident_cap_blocks_new_resident 第二腿在合入时因此失败,
  已修并通过)。
- 档位大小写:A 用小写 paid/heavy,B 用 Paid/Heavy;比较一律走
  _norm_tier + HYBRID_RESIDENT_TIERS(resident_tier_eligible 与
  promote_to_resident 两处),fail-closed 语义不变。
- 严格快照校验翻为默认开启(设计 §2.2):_snapshot_strict_tenant
  默认 True,显式 CHV_SNAPSHOT_STRICT_TENANT=0 才关闭;旧恢复
  测试的 fixture 已登记带匹配 tenant 的快照,A 轨 legacy 测试改
  为显式 opt-out 覆盖。

生产接线:worker 只调 dispatch_task,dispatch_task_hybrid 在生产
路径上无调用方;dispatch_task 顶部加一道门,warm_pool_size>=1 时
走 dispatch_task_hybrid,==0 时保持旧冷启语义(不成环:hybrid 的
<=0 分支退化为 dispatch_task_persistent,后者 <=0 调回 dispatch_task)。

旧网关测试更新(B 的语义变更):test_worker_gateway_guard_allows_chv
前加常驻登记;新增 test_worker_gateway_guard_denies_pool_vm。

测试:test_chv_hybrid_a/b 64 项 + 受影响旧测试 16 项(Gateway、
RestoreNoFallback、RestoreSnapshotTest)全部通过;全量待跑。
This commit is contained in:
host committed 2026-11-08 22:00:12 +00:00
commit 99d9e5771b
5 files changed
+1639 -100

No files matched your search

+837 -88
View File
File diff suppressed because it is too large. Load diff
+20
View File
@@ -26,6 +26,7 @@ from pathlib import Path
from unittest import mock from unittest import mock
try: try:
from cloud.chv.disk import record_snapshot
from cloud.chv.executor import CloudHypervisorExecutor from cloud.chv.executor import CloudHypervisorExecutor
from cloud.chv.hostkeys import dirty_marker_path from cloud.chv.hostkeys import dirty_marker_path
from cloud.chv.worker_adapter import ( from cloud.chv.worker_adapter import (
@@ -38,6 +39,7 @@ try:
from cloud.sandbox import ChvFailure from cloud.sandbox import ChvFailure
except ImportError: # discover -s cloud/tests 时把项目根目录补进 sys.path except ImportError: # discover -s cloud/tests 时把项目根目录补进 sys.path
sys.path.insert(0, str(Path(__file__).resolve().parents[2])) sys.path.insert(0, str(Path(__file__).resolve().parents[2]))
from cloud.chv.disk import record_snapshot
from cloud.chv.executor import CloudHypervisorExecutor from cloud.chv.executor import CloudHypervisorExecutor
from cloud.chv.hostkeys import dirty_marker_path from cloud.chv.hostkeys import dirty_marker_path
from cloud.chv.worker_adapter import ( from cloud.chv.worker_adapter import (
@@ -288,8 +290,20 @@ class RestoreSnapshotTest(CloseupCase):
fh.write(b"S" * 60) fh.write(b"S" * 60)
return "file://" + path, path return "file://" + path, path
def _register_snapshot(self, url, tenant="proj1", vm="chv-abc123"):
# 严格快照校验(设计 §2.2,默认开启):恢复须登记带
# tenant_id 且与请求方一致;_task() 的 project_id 是
# proj1,即请求租户。登记用 record_snapshot(与
# snapshot_vm 同款登记口径)。vm 参数只定登记落在哪
# 个 chv-* 目录(find_snapshot_entry 全盘扫描),默认
# chv-abc123;net_register 失败测试须用别的 vm 名,
# 否则登记建的目录会撞它同名 blocker 文件。
record_snapshot(self.state_dir, vm, url, 20.0,
tenant_id=tenant)
def test_restore_failure_closes_vm(self): def test_restore_failure_closes_vm(self):
url, _ = self._snap_url() url, _ = self._snap_url()
self._register_snapshot(url)
ex = self.make_exec() ex = self.make_exec()
def runner(cmd, timeout=None): def runner(cmd, timeout=None):
@@ -305,6 +319,7 @@ class RestoreSnapshotTest(CloseupCase):
def test_restore_success_and_wake_from_hibernate(self): def test_restore_success_and_wake_from_hibernate(self):
url, _ = self._snap_url() url, _ = self._snap_url()
self._register_snapshot(url)
ex = self.make_exec() ex = self.make_exec()
out = ex.restore_vm(self._task(), url) out = ex.restore_vm(self._task(), url)
self.assertEqual(out["mode"], "restored") self.assertEqual(out["mode"], "restored")
@@ -327,6 +342,7 @@ class RestoreSnapshotTest(CloseupCase):
# 已登记 VM 静默复用 tap/mac/ip,且自身不登记、后 # 已登记 VM 静默复用 tap/mac/ip,且自身不登记、后
# 来的 wake 也看不到它,两边都漏判。 # 来的 wake 也看不到它,两边都漏判。
url, _ = self._snap_url() url, _ = self._snap_url()
self._register_snapshot(url)
self._register_other_vm_net() self._register_other_vm_net()
ex = self.make_exec(env_extra={ ex = self.make_exec(env_extra={
"CHV_NET_JSON": '[{"tap": "tap0", ' "CHV_NET_JSON": '[{"tap": "tap0", '
@@ -343,6 +359,7 @@ class RestoreSnapshotTest(CloseupCase):
# feat-chv-cover100:恢复成功后登记 net.json,后来 # feat-chv-cover100:恢复成功后登记 net.json,后来
# 的 wake 才能看到恢复出的 VM 占用的 tap/mac/ip。 # 的 wake 才能看到恢复出的 VM 占用的 tap/mac/ip。
url, _ = self._snap_url() url, _ = self._snap_url()
self._register_snapshot(url)
ex = self.make_exec(env_extra={ ex = self.make_exec(env_extra={
"CHV_NET_JSON": '[{"tap": "tap0", ' "CHV_NET_JSON": '[{"tap": "tap0", '
'"mac": "aa:bb:cc:dd:ee:01", ' '"mac": "aa:bb:cc:dd:ee:01", '
@@ -361,6 +378,7 @@ class RestoreSnapshotTest(CloseupCase):
#(P2-19 语义),restore_vm 的 except 会 stop_vm #(P2-19 语义),restore_vm 的 except 会 stop_vm
# 清掉半恢复 VM(shutdown 已发出)。 # 清掉半恢复 VM(shutdown 已发出)。
url, _ = self._snap_url() url, _ = self._snap_url()
self._register_snapshot(url, vm="chv-zzz")
blocker = os.path.join(self.state_dir, "chv-abc123") blocker = os.path.join(self.state_dir, "chv-abc123")
os.makedirs(self.state_dir, exist_ok=True) os.makedirs(self.state_dir, exist_ok=True)
with open(blocker, "w") as fh: with open(blocker, "w") as fh:
@@ -377,6 +395,7 @@ class RestoreSnapshotTest(CloseupCase):
def test_restore_without_net_json_writes_no_registry(self): def test_restore_without_net_json_writes_no_registry(self):
url, _ = self._snap_url() url, _ = self._snap_url()
self._register_snapshot(url)
ex = self.make_exec() ex = self.make_exec()
out = ex.restore_vm(self._task(), url) out = ex.restore_vm(self._task(), url)
self.assertEqual(out["mode"], "restored") self.assertEqual(out["mode"], "restored")
@@ -385,6 +404,7 @@ class RestoreSnapshotTest(CloseupCase):
def test_restore_provisioned_missing_known_hosts(self): def test_restore_provisioned_missing_known_hosts(self):
url, _ = self._snap_url() url, _ = self._snap_url()
self._register_snapshot(url)
ex = self.make_exec(env_extra={"CHV_HOST_KEY_MODE": "provisioned"}) ex = self.make_exec(env_extra={"CHV_HOST_KEY_MODE": "provisioned"})
with self.assertRaises(ChvFailure) as ctx: with self.assertRaises(ChvFailure) as ctx:
ex.restore_vm(self._task(), url) ex.restore_vm(self._task(), url)
+33 -3
View File
@@ -21,7 +21,9 @@ from types import SimpleNamespace
try: try:
from cloud.chv.api_client import _unix_socket_api from cloud.chv.api_client import _unix_socket_api
from cloud.chv.disk import check_disk_quota, provision_task_disk from cloud.chv.disk import (check_disk_quota, provision_task_disk,
record_snapshot)
from cloud.chv.executor import write_persistent_vm
from cloud.chv.hostkeys import build_seed_disk, seed_disk_path from cloud.chv.hostkeys import build_seed_disk, seed_disk_path
from cloud.chv.experimental_pool import WarmPool from cloud.chv.experimental_pool import WarmPool
from cloud.sandbox import ( from cloud.sandbox import (
@@ -32,7 +34,9 @@ try:
except ImportError: except ImportError:
sys.path.insert(0, str(Path(__file__).resolve().parents[2])) sys.path.insert(0, str(Path(__file__).resolve().parents[2]))
from cloud.chv.api_client import _unix_socket_api from cloud.chv.api_client import _unix_socket_api
from cloud.chv.disk import check_disk_quota, provision_task_disk from cloud.chv.disk import (check_disk_quota, provision_task_disk,
record_snapshot)
from cloud.chv.executor import write_persistent_vm
from cloud.chv.hostkeys import build_seed_disk, seed_disk_path from cloud.chv.hostkeys import build_seed_disk, seed_disk_path
from cloud.chv.experimental_pool import WarmPool from cloud.chv.experimental_pool import WarmPool
from cloud.sandbox import ( from cloud.sandbox import (
@@ -227,8 +231,13 @@ class RestoreNoFallbackTest(ChvFixCase):
# feat-chv-secfix(L2):恢复源须在白名单目录内, # feat-chv-secfix(L2):恢复源须在白名单目录内,
# 本测试的意图是“起 VMM 前失败”,用 state_dir 下路径。 # 本测试的意图是“起 VMM 前失败”,用 state_dir 下路径。
snap_url = "file://" + os.path.join(self.state_dir, "snap", "x") snap_url = "file://" + os.path.join(self.state_dir, "snap", "x")
# 严格快照校验(设计 §2.2,默认开启):恢复须登记带
# tenant_id 且与请求方一致;本测试意图是 known-hosts
# 门,登记与请求方同租户 p1 后流程才走到该门。
record_snapshot(self.state_dir, "chv-r1", snap_url, 20.0,
tenant_id="p1")
with self.assertRaises(ChvFailure) as ctx: with self.assertRaises(ChvFailure) as ctx:
ex.restore_vm({"id": "r1"}, snap_url) ex.restore_vm({"id": "r1", "project_id": "p1"}, snap_url)
self.assertEqual(ctx.exception.kind, "host_key_not_pinned") self.assertEqual(ctx.exception.kind, "host_key_not_pinned")
self.assertEqual(launches, []) self.assertEqual(launches, [])
@@ -304,6 +313,14 @@ class GatewayExtraEnvTest(ChvFixCase):
db_path = os.path.join(self.dir, "control.db") db_path = os.path.join(self.dir, "control.db")
init_db(db_path) init_db(db_path)
ex = self.make_exec() ex = self.make_exec()
# 混合模式网关分流(设计 §6):网关任务只接常驻 VM。
# 本测试的租户 p1(project_id → tenant)须有常驻有效态
# 登记 + Paid 档位,否则 _check_gateway_vm_class 以
# gateway_pool_denied 拒绝;池 VM 拒绝腿见下一测试。
write_persistent_vm(self.state_dir, "p1", {
"vm": "chv-persist-p1", "tenant_id": "p1",
"state": "resident-active", "tier": "Paid",
"origin": "task", "gitea_synced": False})
worker = Worker(db_path, "/fake/host.py", executor=ex, worker = Worker(db_path, "/fake/host.py", executor=ex,
env={"XCM_API_KEY": "test-placeholder"}) env={"XCM_API_KEY": "test-placeholder"})
task = {"id": "gw1", "project_id": "p1", "prompt": "做目标", task = {"id": "gw1", "project_id": "p1", "prompt": "做目标",
@@ -330,6 +347,19 @@ class GatewayExtraEnvTest(ChvFixCase):
os.path.join(self.state_dir, "chv-gw1", os.path.join(self.state_dir, "chv-gw1",
"ssh_host_ed25519_key"))) "ssh_host_ed25519_key")))
def test_worker_gateway_guard_denies_pool_vm(self):
# 混合模式网关分流(设计 §6):无常驻租约的网关任务
# 走池即以 gateway_pool_denied 明确拒绝,不偷偷放行。
ex = self.make_exec()
task = {"id": "gw2", "project_id": "p1", "prompt": "做目标",
"goal_mode": 1,
"_gateway_provider": "sub2api",
"_gateway_env": {"SUB2API_API_KEY": "gwkey",
"SUB2API_BASE_URL": "http://gw/"}}
with self.assertRaises(ChvFailure) as ctx:
ex.dispatch_task(task)
self.assertEqual(ctx.exception.kind, "gateway_pool_denied")
# -- P1-2 取消/清理/看门狗 ---------------------------------------------- # -- P1-2 取消/清理/看门狗 ----------------------------------------------
class CancelCleanupTest(ChvFixCase): class CancelCleanupTest(ChvFixCase):
+39 -9
View File
@@ -283,7 +283,7 @@ class HybridRouteTest(HybridACase):
self.assertEqual(out["allocation"], "resident-cold") self.assertEqual(out["allocation"], "resident-cold")
self.assertEqual(out["vm"], "chv-persist-paid1") self.assertEqual(out["vm"], "chv-persist-paid1")
rec = read_persistent_vm(self.state_dir, "paid1") rec = read_persistent_vm(self.state_dir, "paid1")
self.assertEqual(rec["state"], "paused") self.assertEqual(rec["state"], "resident-active")
self.assertEqual(rec["tier"], "paid") self.assertEqual(rec["tier"], "paid")
self.assertTrue(rec["resident"]) self.assertTrue(rec["resident"])
self.assertEqual(ex._resident_count(), 1) self.assertEqual(ex._resident_count(), 1)
@@ -301,7 +301,7 @@ class HybridRouteTest(HybridACase):
self.assertEqual(out["vm"], "chv-persist-paid1") self.assertEqual(out["vm"], "chv-persist-paid1")
self.assertIn("resume_seconds", out) self.assertIn("resume_seconds", out)
rec = read_persistent_vm(self.state_dir, "paid1") rec = read_persistent_vm(self.state_dir, "paid1")
self.assertEqual(rec["state"], "paused") self.assertEqual(rec["state"], "resident-active")
self.assertEqual(rec["tier"], "paid") self.assertEqual(rec["tier"], "paid")
self.assertTrue(rec["resident"]) self.assertTrue(rec["resident"])
@@ -311,7 +311,7 @@ class HybridRouteTest(HybridACase):
for tenant in ("r1", "r2"): for tenant in ("r1", "r2"):
write_persistent_vm(self.state_dir, tenant, { write_persistent_vm(self.state_dir, tenant, {
"vm": "chv-persist-%s" % tenant, "vm": "chv-persist-%s" % tenant,
"tenant_id": tenant, "state": "paused", "tenant_id": tenant, "state": "resident-active",
"origin": "task", "gitea_synced": False, "origin": "task", "gitea_synced": False,
"tier": "paid", "resident": True}) "tier": "paid", "resident": True})
self.assertEqual(ex._resident_count(), 2) self.assertEqual(ex._resident_count(), 2)
@@ -324,11 +324,11 @@ class HybridRouteTest(HybridACase):
ex2 = self.make_exec(env_extra={"CHV_QUEUE_TIMEOUT": "0"}) ex2 = self.make_exec(env_extra={"CHV_QUEUE_TIMEOUT": "0"})
write_persistent_vm(self.state_dir, "r1", { write_persistent_vm(self.state_dir, "r1", {
"vm": "chv-persist-r1", "tenant_id": "r1", "vm": "chv-persist-r1", "tenant_id": "r1",
"state": "paused", "origin": "task", "state": "resident-active", "origin": "task",
"gitea_synced": False, "tier": "paid", "resident": True}) "gitea_synced": False, "tier": "paid", "resident": True})
write_persistent_vm(self.state_dir, "r2", { write_persistent_vm(self.state_dir, "r2", {
"vm": "chv-persist-r2", "tenant_id": "r2", "vm": "chv-persist-r2", "tenant_id": "r2",
"state": "paused", "origin": "task", "state": "resident-active", "origin": "task",
"gitea_synced": False, "tier": "paid", "resident": True}) "gitea_synced": False, "tier": "paid", "resident": True})
self._socket("chv-persist-r1") self._socket("chv-persist-r1")
self.vm_states["chv-persist-r1"] = "Paused" self.vm_states["chv-persist-r1"] = "Paused"
@@ -670,12 +670,15 @@ class SnapshotTenantStrictTest(HybridACase):
"snapshot_tenant_mismatch") "snapshot_tenant_mismatch")
self.assertIn("vm_restore_denied", self._event_names()) self.assertIn("vm_restore_denied", self._event_names())
def test_legacy_warn_pass_preserved_when_not_strict(self): def test_legacy_warn_pass_when_strict_opted_out(self):
# 混合模式未启用时,_validate_restore 沿用阶段 1 的 # 严格校验默认开启(设计 §2.2,集成时翻为默认);
# legacy 警告放行(旧测试依赖该语义,本轨不改)。 # legacy 警告放行只在显式 CHV_SNAPSHOT_STRICT_TENANT=0
# 时保留(有未标记旧快照的环境的逃生口)。
url = self._snap_url("chv-a") url = self._snap_url("chv-a")
self._record("chv-a", url, None) self._record("chv-a", url, None)
ex = self.make_exec(env_extra={"CHV_WARM_POOL_SIZE": "0"}) ex = self.make_exec(env_extra={
"CHV_WARM_POOL_SIZE": "0",
"CHV_SNAPSHOT_STRICT_TENANT": "0"})
snapshot_url, _net = ex._validate_restore( snapshot_url, _net = ex._validate_restore(
{"id": "task-a", "project_id": "tenant-a"}, url) {"id": "task-a", "project_id": "tenant-a"}, url)
self.assertEqual(snapshot_url, url) self.assertEqual(snapshot_url, url)
@@ -684,5 +687,32 @@ class SnapshotTenantStrictTest(HybridACase):
self.assertNotIn("vm_restore_denied", self._event_names()) self.assertNotIn("vm_restore_denied", self._event_names())
# -- dispatch 入口接线(集成时补)--------------------------------------
class DispatchWiringTest(HybridACase):
def test_dispatch_task_routes_to_hybrid_when_pool_on(self):
# warm_pool_size>=1 时 dispatch_task 走混合路由。
ex = self.make_exec()
out = ex.dispatch_task(self._task(task_id="t1"))
self.assertEqual(out["status"], "completed")
self.assertEqual(out["route"], "pool")
self.assertEqual(out["allocation"], "pool-cold")
def test_dispatch_task_routes_resident_when_paid(self):
ex = self.make_exec()
out = ex.dispatch_task(
self._task(task_id="t1", project="paid1", tier="paid"))
self.assertEqual(out["status"], "completed")
self.assertEqual(out["route"], "resident")
self.assertEqual(out["allocation"], "resident-cold")
def test_dispatch_task_stays_cold_when_pool_off(self):
# warm_pool_size==0 时 dispatch_task 保持旧冷启语义,
# 不进混合路由(结果无 route 字段)。
ex = self.make_exec(env_extra={"CHV_WARM_POOL_SIZE": "0"})
out = ex.dispatch_task(self._task(task_id="t1"))
self.assertEqual(out["status"], "completed")
self.assertNotIn("route", out)
if __name__ == "__main__": if __name__ == "__main__":
unittest.main() unittest.main()
+710
View File
@@ -0,0 +1,710 @@
"""CHV 混合模式 B 轨(feat-chv-hybrid-b)测试。
只用标准库与 mock runner/launcher/api_client,绝不启动真实
VM(与 test_chv_opt_stage1.py 同款边界)。覆盖(设计
DESIGN-HYBRID.md §9.1 第 5–9 项):
- 登记字段扩展 + tier 校验(含无 tier 行拒绝常驻)
- 租约挂起/降级计时(环境变量注入短阈值测状态机转换,
含降级同步失败中止)
- 排队超时拒绝(host_capacity_exceeded,不插队、不抢占、
不升降档)
- 升级提示点去重(一周内同一租户最多一次)
- 网关分流(常驻可接、池不接)
"""
import json
import os
import sys
import tempfile
import unittest
from pathlib import Path
from unittest import mock
try:
from cloud.chv import ChvFailure
from cloud.chv import CloudHypervisorExecutor
from cloud.chv.disk import (
read_persistent_vm,
write_persistent_vm,
write_vm_tenant,
)
except ImportError: # discover -s cloud/tests 时把项目根目录补进 sys.path
sys.path.insert(0, str(Path(__file__).resolve().parents[2]))
from cloud.chv import ChvFailure
from cloud.chv import CloudHypervisorExecutor
from cloud.chv.disk import (
read_persistent_vm,
write_persistent_vm,
write_vm_tenant,
)
class FakeResult:
def __init__(self, returncode=0, stdout="", stderr=""):
self.returncode = returncode
self.stdout = stdout
self.stderr = stderr
class FakeProc:
def __init__(self, alive=True):
self._alive = alive
def poll(self):
return None if self._alive else 1
def terminate(self):
self._alive = False
def wait(self, timeout=None):
return 0
def kill(self):
self._alive = False
def _write(path, content=b"x"):
with open(path, "wb") as fh:
fh.write(content)
os.chmod(path, 0o755)
return path
class HybridBCase(unittest.TestCase):
def setUp(self):
self.tmp = tempfile.TemporaryDirectory()
self.dir = self.tmp.name
self.binary = _write(os.path.join(self.dir, "cloud-hypervisor"))
self.ch_remote = _write(os.path.join(self.dir, "ch-remote"))
self.kernel = _write(os.path.join(self.dir, "vmlinux"))
self.rootfs = _write(os.path.join(self.dir, "base.qcow2"),
b"R" * 100)
self.ssh_key = _write(os.path.join(self.dir, "id_chv"))
self.kvm = _write(os.path.join(self.dir, "kvm"))
self.cpuinfo = os.path.join(self.dir, "cpuinfo")
with open(self.cpuinfo, "w") as fh:
fh.write("flags : fpu vmx sse\n")
self.state_dir = os.path.join(self.dir, "state")
self.calls = []
self.api_calls = []
self.launches = []
self.vm_states = {}
self.sleeps = []
self.events = []
def tearDown(self):
self.tmp.cleanup()
def make_exec(self, runner=None, api_client=None, launcher=None,
env_extra=None, sleeper=None, **overrides):
env = {"CHV_EGRESS_PROXY": "http://10.0.0.1:8888",
"CHV_HOST_KEY_MODE": "keyscan",
"CHV_SSH_FORWARD": "1",
"CHV_HOST_ID": "host-test-1"}
if env_extra:
env.update(env_extra)
if runner is None:
runner = self._default_runner
if api_client is None:
api_client = self._default_api
if launcher is None:
launcher = self._default_launcher
kwargs = dict(
binary=self.binary, ch_remote=self.ch_remote,
kernel=self.kernel, rootfs=self.rootfs,
ssh_key=self.ssh_key, kvm_path=self.kvm,
cpuinfo_path=self.cpuinfo, state_dir=self.state_dir,
runner=runner, launcher=launcher, api_client=api_client,
sleeper=sleeper if sleeper is not None
else (lambda s: self.sleeps.append(s)),
env=env)
kwargs.update(overrides)
ex = CloudHypervisorExecutor(**kwargs)
ex.audit_callback = lambda event, task, detail, kind=None: \
self.events.append((event, detail, kind))
return ex
def _default_runner(self, cmd, timeout=None):
self.calls.append(list(cmd))
if cmd[0] == "ssh-keyscan":
return FakeResult(
0, "[127.0.0.1]:2222 ssh-ed25519 AAAAB3test\n", "")
if cmd[0] == "ssh":
if cmd[-1] == "true":
return FakeResult(0, "", "")
if "dispatch" in cmd:
return FakeResult(0, json.dumps(
{"status": "completed", "text": "ok-dispatch"}), "")
return FakeResult(0, "", "")
if cmd and os.path.basename(str(cmd[0])) == "ch-remote":
if "snapshot" in cmd:
url = cmd[cmd.index("snapshot") + 1]
if url.startswith("file://"):
snap_dir = url[len("file://"):]
os.makedirs(snap_dir, exist_ok=True)
with open(os.path.join(snap_dir, "state.json"),
"w") as fh:
fh.write("{}")
return FakeResult(0, "", "")
return FakeResult(0, "", "")
return FakeResult(0, "", "")
def _default_api(self, socket_path, method, path, payload=None):
self.api_calls.append((method, path, payload))
if path == "/vm.info":
state = self.vm_states.get(
os.path.basename(socket_path)[:-len(".sock")])
if state:
return (200, {"state": state})
return (200, {})
return (200, {})
def _default_launcher(self, cmd):
self.launches.append(list(cmd))
return FakeProc(alive=True)
def _task(self, task_id="abc123", project="proj1"):
return {"id": task_id, "prompt": "做一个加法函数",
"project_id": project}
def _event_names(self):
return [e[0] for e in self.events]
def _write_resident(self, ex, tenant, state, tier, last_active,
vm=None):
vm = vm or ex.persistent_vm_name(tenant)
write_persistent_vm(ex.state_dir, tenant, {
"vm": vm, "tenant_id": tenant, "tier": tier,
"state": state, "last_active_at": last_active,
"lease_expires_at": last_active + 1800 + 86400,
"origin": "task", "gitea_synced": True,
"updated_at": last_active})
return vm
def _live_vm(self, ex, vm):
"""给一台 VM 造“活着”的外形:socket 文件 + vm.info。"""
sock = os.path.join(ex.state_dir, vm + ".sock")
os.makedirs(ex.state_dir, exist_ok=True)
with open(sock, "w") as fh:
fh.write("")
return sock
# -- 登记字段扩展 + tier 校验(第 5 项)-------------------------------
class ResidentRegistryTest(HybridBCase):
def test_promote_writes_extended_fields(self):
ex = self.make_exec()
out = ex.promote_to_resident("t1", "Paid")
self.assertTrue(out["promoted"])
self.assertFalse(out["queued"])
record = read_persistent_vm(ex.state_dir, "t1")
self.assertEqual(record["tier"], "Paid")
self.assertEqual(record["state"], "resident-active")
self.assertIn("last_active_at", record)
self.assertIn("lease_expires_at", record)
self.assertGreater(record["lease_expires_at"],
record["last_active_at"])
self.assertIn("vm_resident_promoted", self._event_names())
detail = [d for e, d, _k in self.events
if e == "vm_resident_promoted"][0]
self.assertIn("host_id=host-test-1", detail)
self.assertIn("tenant=t1", detail)
def test_promote_free_tier_denied(self):
ex = self.make_exec()
with self.assertRaises(ChvFailure) as ctx:
ex.promote_to_resident("tfree", "Free")
self.assertEqual(ctx.exception.kind, "tier_denied")
self.assertIsNone(read_persistent_vm(ex.state_dir, "tfree"))
def test_promote_no_tier_row_denied(self):
# 无 tier 行 = 无常驻资格(fail-closed,走池):登记里
# 没有 tier 字段、也不传 tier 参数时同样拒绝。
ex = self.make_exec()
with self.assertRaises(ChvFailure) as ctx:
ex.promote_to_resident("tnope")
self.assertEqual(ctx.exception.kind, "tier_denied")
def test_tier_eligible_matrix(self):
ex = self.make_exec()
self.assertTrue(ex.resident_tier_eligible("x", "Paid"))
self.assertTrue(ex.resident_tier_eligible("x", "Heavy"))
self.assertFalse(ex.resident_tier_eligible("x", "Free"))
self.assertFalse(ex.resident_tier_eligible("x", None))
self.assertFalse(ex.resident_tier_eligible("nobody"))
def test_promote_over_cap_queued_not_preempt(self):
# max_vms 默认 4 → 常驻上限 2;第 3 个排队,不抢占前两个。
ex = self.make_exec()
self.assertTrue(ex.promote_to_resident("t1", "Paid")["promoted"])
self.assertTrue(ex.promote_to_resident("t2", "Heavy")["promoted"])
out = ex.promote_to_resident("t3", "Paid")
self.assertFalse(out["promoted"])
self.assertTrue(out["queued"])
self.assertEqual(out["reason"], "resident_capacity_exceeded")
self.assertEqual(ex._resident_count(), 2)
self.assertEqual(
read_persistent_vm(ex.state_dir, "t1")["state"],
"resident-active")
self.assertEqual(
read_persistent_vm(ex.state_dir, "t2")["state"],
"resident-active")
self.assertIsNone(read_persistent_vm(ex.state_dir, "t3"))
def test_promote_idempotent_no_double_slot(self):
ex = self.make_exec()
ex.promote_to_resident("t1", "Paid")
out = ex.promote_to_resident("t1", "Paid")
self.assertTrue(out["promoted"])
self.assertEqual(ex._resident_count(), 1)
def test_touch_renews_only_resident(self):
ex = self.make_exec(env_extra={
"CHV_RESIDENT_IDLE_SUSPEND_SECONDS": "60",
"CHV_RESIDENT_IDLE_DEMOTE_SECONDS": "120"})
now = 1800000000.0
# legacy paused 登记:touch 不写、不造租约字段。
write_persistent_vm(ex.state_dir, "tleg", {
"vm": ex.persistent_vm_name("tleg"), "tenant_id": "tleg",
"state": "paused", "origin": "task",
"gitea_synced": True, "updated_at": now - 100})
self.assertFalse(ex.touch_resident_activity("tleg", now=now))
self.assertNotIn("last_active_at",
read_persistent_vm(ex.state_dir, "tleg"))
# 常驻登记:touch 续期并重算 lease_expires_at。
vm = self._write_resident(ex, "t2", "resident-active",
"Paid", now - 100)
self.assertTrue(ex.touch_resident_activity("t2", now=now))
record = read_persistent_vm(ex.state_dir, "t2")
self.assertEqual(record["last_active_at"], now)
self.assertEqual(record["lease_expires_at"], now + 60 + 120)
self.assertEqual(record["state"], "resident-active")
self.assertEqual(record["tier"], "Paid")
def test_suspend_after_task_keeps_resident_state_and_tier(self):
ex = self.make_exec()
ex.promote_to_resident("t1", "Paid")
vm = ex.persistent_vm_name("t1")
self.assertTrue(ex.suspend_after_task({"id": vm},
tenant_id="t1"))
record = read_persistent_vm(ex.state_dir, "t1")
self.assertEqual(record["state"], "resident-active")
self.assertEqual(record["tier"], "Paid")
self.assertIn("last_active_at", record)
def test_acquire_resident_paused_paid_tier(self):
ex = self.make_exec(env_extra={"CHV_WARM_POOL_SIZE": "1"})
now = 1800000000.0
vm = self._write_resident(ex, "t1", "resident-paused",
"Paid", now - 10)
write_vm_tenant(ex.state_dir, vm, "t1")
self._live_vm(ex, vm)
self.vm_states[vm] = "Paused"
out = ex.acquire_for_task(self._task(project="t1"))
self.assertEqual(out["mode"], "resumed")
self.assertEqual(out["vm"], vm)
record = read_persistent_vm(ex.state_dir, "t1")
# resume 后仍是常驻态(不写回 legacy running),tier 保留。
self.assertEqual(record["state"], "resident-paused")
self.assertEqual(record["tier"], "Paid")
def test_acquire_resident_paused_free_tier_denied(self):
ex = self.make_exec(env_extra={"CHV_WARM_POOL_SIZE": "1"})
now = 1800000000.0
vm = self._write_resident(ex, "t1", "resident-paused",
"Free", now - 10)
write_vm_tenant(ex.state_dir, vm, "t1")
self._live_vm(ex, vm)
self.vm_states[vm] = "Paused"
out = ex.acquire_for_task(self._task(project="t1"))
self.assertEqual(out["mode"], "cold")
self.assertIn("resident_tier_denied", self._event_names())
record = read_persistent_vm(ex.state_dir, "t1")
self.assertEqual(record["state"], "resident-paused")
self.assertEqual(record["tier"], "Free")
def test_acquire_legacy_paused_without_tier_still_works(self):
# legacy paused 登记无 tier 字段:走旧语义,仍可认领。
ex = self.make_exec(env_extra={"CHV_WARM_POOL_SIZE": "1"})
vm = ex.persistent_vm_name("tleg")
write_persistent_vm(ex.state_dir, "tleg", {
"vm": vm, "tenant_id": "tleg", "state": "paused",
"origin": "prewarm", "gitea_synced": True,
"updated_at": 1800000000.0})
write_vm_tenant(ex.state_dir, vm, "tleg")
self._live_vm(ex, vm)
self.vm_states[vm] = "Paused"
out = ex.acquire_for_task(self._task(project="tleg"))
self.assertEqual(out["mode"], "resumed")
# -- 租约挂起/降级计时(第 6 项)--------------------------------------
class LeaseTimerTest(HybridBCase):
def _lease_exec(self):
return self.make_exec(env_extra={
"CHV_RESIDENT_IDLE_SUSPEND_SECONDS": "60",
"CHV_RESIDENT_IDLE_DEMOTE_SECONDS": "120"})
def test_sweep_suspends_idle_active_already_paused(self):
ex = self._lease_exec()
now = 1800000000.0
vm = self._write_resident(ex, "t1", "resident-active",
"Paid", now - 61)
self._live_vm(ex, vm)
self.vm_states[vm] = "Paused"
out = ex.sweep_resident_leases(now=now)
self.assertEqual(out["suspended"], ["t1"])
self.assertEqual(out["demoted"], [])
record = read_persistent_vm(ex.state_dir, "t1")
self.assertEqual(record["state"], "resident-paused")
self.assertEqual(record["tier"], "Paid")
self.assertIn("vm_resident_suspended", self._event_names())
detail = [d for e, d, _k in self.events
if e == "vm_resident_suspended"][0]
self.assertIn("host_id=host-test-1", detail)
self.assertIn("tenant=t1", detail)
# VM 本已 Paused:sweep 不应再发 pause。
self.assertFalse(any(c[-1] == "pause" for c in self.calls
if c and "ch-remote" in c[0]))
def test_sweep_suspends_running_vm_calls_pause(self):
ex = self._lease_exec()
now = 1800000000.0
vm = self._write_resident(ex, "t1", "resident-active",
"Paid", now - 61)
self._live_vm(ex, vm)
self.vm_states[vm] = "Running"
out = ex.sweep_resident_leases(now=now)
self.assertEqual(out["suspended"], ["t1"])
self.assertTrue(any(c[-1] == "pause" for c in self.calls
if c and "ch-remote" in c[0]))
self.assertEqual(
read_persistent_vm(ex.state_dir, "t1")["state"],
"resident-paused")
def test_sweep_no_suspend_before_threshold(self):
ex = self._lease_exec()
now = 1800000000.0
self._write_resident(ex, "t1", "resident-active",
"Paid", now - 30)
out = ex.sweep_resident_leases(now=now)
self.assertEqual(out, {"suspended": [], "demoted": [],
"aborted": []})
self.assertEqual(
read_persistent_vm(ex.state_dir, "t1")["state"],
"resident-active")
def test_sweep_demotes_after_24h(self):
ex = self._lease_exec()
now = 1800000000.0
vm = self._write_resident(ex, "t1", "resident-paused",
"Paid", now - 61 - 121)
self._live_vm(ex, vm)
self.vm_states[vm] = "Paused"
# 用户数据盘与任务盘:降级后须保留(快照输入 + 用户数据)。
vm_dir = os.path.join(ex.state_dir, vm)
os.makedirs(vm_dir, exist_ok=True)
with open(os.path.join(vm_dir, "data.raw"), "wb") as fh:
fh.write(b"D" * 10)
with open(os.path.join(vm_dir, "disk.raw"), "wb") as fh:
fh.write(b"K" * 10)
out = ex.sweep_resident_leases(now=now)
self.assertEqual(out["demoted"], ["t1"])
record = read_persistent_vm(ex.state_dir, "t1")
self.assertEqual(record["state"], "demoted")
self.assertEqual(record["last_active_at"], now - 61 - 121)
self.assertIn("snapshot_url", record)
self.assertIn("vm_resident_demoted", self._event_names())
detail = [d for e, d, _k in self.events
if e == "vm_resident_demoted"][0]
self.assertIn("host_id=host-test-1", detail)
self.assertIn("tenant=t1", detail)
self.assertTrue(os.path.exists(
os.path.join(vm_dir, "data.raw")))
self.assertTrue(os.path.exists(
os.path.join(vm_dir, "disk.raw")))
# 槽位已释放:可再 promote 别的租户。
self.assertEqual(ex._resident_count(), 0)
def test_demote_sync_failure_aborts(self):
ex = self._lease_exec()
now = 1800000000.0
vm = self._write_resident(ex, "t1", "resident-paused",
"Paid", now - 61 - 121)
self._live_vm(ex, vm)
self.vm_states[vm] = "Paused"
out = ex.demote_resident(
"t1", sync_fn=lambda tenant, v: False, now=now)
self.assertFalse(out["demoted"])
self.assertEqual(out["reason"], "sync_failed")
record = read_persistent_vm(ex.state_dir, "t1")
self.assertEqual(record["state"], "resident-paused")
self.assertNotIn("snapshot_url", record)
# VM 保留:socket 还在,未销毁。
self.assertTrue(os.path.exists(
os.path.join(ex.state_dir, vm + ".sock")))
self.assertIn("vm_resident_demote_aborted",
self._event_names())
def test_demote_snapshot_failure_aborts(self):
ex = self._lease_exec()
now = 1800000000.0
vm = self._write_resident(ex, "t1", "resident-paused",
"Paid", now - 61 - 121)
self._live_vm(ex, vm)
self.vm_states[vm] = "Paused"
with mock.patch.object(
ex, "snapshot_vm",
side_effect=ChvFailure("snapshot_failed", "boom")):
out = ex.demote_resident("t1", now=now)
self.assertFalse(out["demoted"])
self.assertEqual(out["reason"], "snapshot_failed")
self.assertEqual(
read_persistent_vm(ex.state_dir, "t1")["state"],
"resident-paused")
def test_demote_not_demotable_states(self):
ex = self._lease_exec()
now = 1800000000.0
self._write_resident(ex, "t1", "resident-active",
"Paid", now - 100000)
out = ex.demote_resident("t1", now=now)
self.assertFalse(out["demoted"])
self.assertEqual(out["reason"], "not_demotable")
out = ex.demote_resident("ghost", now=now)
self.assertFalse(out["demoted"])
self.assertEqual(out["reason"], "not_demotable")
def test_sweep_demote_aborted_on_sync_hook_failure(self):
ex = self._lease_exec()
ex.resident_sync_hook = lambda tenant, vm: False
now = 1800000000.0
vm = self._write_resident(ex, "t1", "resident-paused",
"Paid", now - 61 - 121)
self._live_vm(ex, vm)
self.vm_states[vm] = "Paused"
out = ex.sweep_resident_leases(now=now)
self.assertEqual(out["demoted"], [])
self.assertEqual(out["aborted"], ["t1"])
self.assertEqual(
read_persistent_vm(ex.state_dir, "t1")["state"],
"resident-paused")
# -- 排队超时拒绝(第 7 项)--------------------------------------------
class QueueTest(HybridBCase):
def _queue_exec(self):
return self.make_exec(env_extra={
"CHV_MAX_VMS": "1",
"CHV_QUEUE_TIMEOUT_SECONDS": "10",
"CHV_QUEUE_POLL_SECONDS": "2"})
def test_wait_for_capacity_timeout_rejects(self):
ex = self._queue_exec()
occupant = FakeProc(alive=True)
ex._vms["chv-other"] = occupant
with self.assertRaises(ChvFailure) as ctx:
ex._wait_for_capacity(self._task(), exclude="chv-abc123")
self.assertEqual(ctx.exception.kind, "host_capacity_exceeded")
# 轮询 10/2 = 5 次 sleeper 调用后才拒绝。
self.assertEqual(len(self.sleeps), 5)
# 不抢占:占位 VM 仍活着;队列计数已归零。
self.assertIsNone(occupant.poll())
self.assertEqual(ex.queue_depth(), 0)
def test_wait_for_capacity_free_passes_immediately(self):
ex = self._queue_exec()
ex._wait_for_capacity(self._task(), exclude="chv-abc123")
self.assertEqual(self.sleeps, [])
def test_run_with_vm_queues_before_wake(self):
ex = self.make_exec(env_extra={
"CHV_MAX_VMS": "2",
"CHV_QUEUE_TIMEOUT_SECONDS": "10",
"CHV_QUEUE_POLL_SECONDS": "2"})
ex._vms["chv-other1"] = FakeProc(alive=True)
ex._vms["chv-other2"] = FakeProc(alive=True)
# 常驻租户排队:档位与租约原样保留,不升降档。
# (max_vms=2 → 常驻上限 1,promote 恰好占满 1 槽。)
ex.promote_to_resident("tq", "Paid")
wake_calls = []
ex.wake_vm = lambda *a, **k: wake_calls.append(a)
with self.assertRaises(ChvFailure) as ctx:
ex._run_with_vm(self._task(project="tq"),
["dispatch", "--name", "tq",
"--task", "hi"])
self.assertEqual(ctx.exception.kind, "host_capacity_exceeded")
self.assertEqual(wake_calls, [])
record = read_persistent_vm(ex.state_dir, "tq")
self.assertEqual(record["tier"], "Paid")
self.assertEqual(record["state"], "resident-active")
def test_wait_for_resident_slot_timeout(self):
ex = self.make_exec(env_extra={
"CHV_MAX_VMS": "4",
"CHV_QUEUE_TIMEOUT_SECONDS": "10",
"CHV_QUEUE_POLL_SECONDS": "2"})
ex.promote_to_resident("t1", "Paid")
ex.promote_to_resident("t2", "Heavy")
with self.assertRaises(ChvFailure) as ctx:
ex._wait_for_resident_slot("t3")
self.assertEqual(ctx.exception.kind, "host_capacity_exceeded")
# 不抢占、不挪池:前两个常驻原样。
self.assertEqual(ex._resident_count(), 2)
self.assertEqual(
read_persistent_vm(ex.state_dir, "t1")["state"],
"resident-active")
def test_wait_for_resident_slot_free_passes(self):
ex = self.make_exec(env_extra={"CHV_MAX_VMS": "4"})
ex.promote_to_resident("t1", "Paid")
ex._wait_for_resident_slot("t2")
self.assertEqual(self.sleeps, [])
# -- 升级提示点去重(第 8 项)------------------------------------------
class UpgradeHintTest(HybridBCase):
def test_hint_at_100_tasks_in_30d_and_dedup(self):
ex = self.make_exec()
now = 1800000000.0
hinted_at = None
for i in range(100):
out = ex.record_pool_task("tfree", tier="Free",
now=now - 1000 + i * 10)
if out["hinted"]:
hinted_at = i
self.assertEqual(hinted_at, 99)
self.assertEqual(
self._event_names().count("resident_upgrade_hint"), 1)
detail = [d for e, d, _k in self.events
if e == "resident_upgrade_hint"][0]
self.assertIn("host_id=host-test-1", detail)
self.assertIn("tenant=tfree", detail)
self.assertIn("tier=Free", detail)
self.assertIn("30d_tasks=100", detail)
# 第 101 次:一周内同一租户不再记第二次。
out = ex.record_pool_task("tfree", tier="Free", now=now + 10)
self.assertFalse(out["hinted"])
self.assertEqual(
self._event_names().count("resident_upgrade_hint"), 1)
# 一周后:可再记一次。
out = ex.record_pool_task("tfree", tier="Free",
now=now + 8 * 86400)
self.assertTrue(out["hinted"])
self.assertEqual(
self._event_names().count("resident_upgrade_hint"), 2)
def test_hint_on_7day_streak(self):
ex = self.make_exec()
now = 1800000000.0
today = int(now // 86400)
hinted = False
for day_offset in range(6, -1, -1):
day_start = (today - day_offset) * 86400 + 3600
for j in range(3):
out = ex.record_pool_task(
"tstreak", tier="Free",
now=day_start + j * 60)
hinted = hinted or out["hinted"]
self.assertTrue(hinted)
self.assertEqual(
self._event_names().count("resident_upgrade_hint"), 1)
detail = [d for e, d, _k in self.events
if e == "resident_upgrade_hint"][0]
self.assertIn("7d_streak_3_per_day", detail)
def test_no_hint_below_streak(self):
ex = self.make_exec()
now = 1800000000.0
today = int(now // 86400)
for day_offset in range(5, -1, -1):
day_start = (today - day_offset) * 86400 + 3600
for j in range(3):
out = ex.record_pool_task(
"tshort", tier="Free", now=day_start + j * 60)
self.assertFalse(out["hinted"])
self.assertEqual(self._event_names(), [])
def test_paid_tier_not_recorded(self):
ex = self.make_exec()
out = ex.record_pool_task("tpaid", tier="Paid",
now=1800000000.0)
self.assertFalse(out["recorded"])
self.assertFalse(out["hinted"])
self.assertFalse(os.path.exists(
ex._pool_task_file("tpaid")))
# -- 网关分流(第 9 项)-------------------------------------------------
class GatewaySplitTest(HybridBCase):
def test_gateway_allowed_resident_paid(self):
ex = self.make_exec()
ex.promote_to_resident("t1", "Paid")
task = self._task(project="t1")
self.assertTrue(ex.gateway_allowed_for_task(
task, model_provider="sub2api"))
self.assertTrue(ex.gateway_allowed_for_task(
dict(task, _gateway_provider="sub2api")))
def test_gateway_allowed_hibernated_resident(self):
ex = self.make_exec()
now = 1800000000.0
self._write_resident(ex, "t1", "hibernated", "Paid", now - 10)
self.assertTrue(ex.gateway_allowed_for_task(
self._task(project="t1"), model_provider="sub2api"))
def test_gateway_denied_pool_tenants(self):
ex = self.make_exec()
# 无登记:池。
self.assertFalse(ex.gateway_allowed_for_task(
self._task(project="tpool"), model_provider="sub2api"))
# 常驻态但 tier 是 Free:无常驻资格,按池算。
now = 1800000000.0
self._write_resident(ex, "tfree", "resident-active",
"Free", now - 10)
self.assertFalse(ex.gateway_allowed_for_task(
self._task(project="tfree"), model_provider="sub2api"))
# demoted:租约已收回,按池算。
self._write_resident(ex, "tdem", "demoted", "Paid", now - 10)
self.assertFalse(ex.gateway_allowed_for_task(
self._task(project="tdem"), model_provider="sub2api"))
# legacy paused 登记:不是常驻态,按池算。
vm = ex.persistent_vm_name("tleg")
write_persistent_vm(ex.state_dir, "tleg", {
"vm": vm, "tenant_id": "tleg", "state": "paused",
"origin": "task", "gitea_synced": True,
"updated_at": now})
self.assertFalse(ex.gateway_allowed_for_task(
self._task(project="tleg"), model_provider="sub2api"))
def test_gateway_check_non_gateway_not_blocked(self):
ex = self.make_exec()
# 非网关任务不由本门拦(model_provider 不是网关)。
self.assertTrue(ex.gateway_allowed_for_task(
self._task(project="tpool"), model_provider=None))
self.assertTrue(ex.gateway_allowed_for_task(
self._task(project="tpool"), model_provider="xcmapi"))
def test_run_with_vm_gateway_pool_denied_before_wake(self):
ex = self.make_exec()
wake_calls = []
ex.wake_vm = lambda *a, **k: wake_calls.append(a)
with self.assertRaises(ChvFailure) as ctx:
ex._run_with_vm(self._task(project="tpool"),
["dispatch", "--name", "tpool",
"--task", "hi"],
model_provider="sub2api")
self.assertEqual(ctx.exception.kind, "gateway_pool_denied")
self.assertEqual(wake_calls, [])
def test_run_with_vm_gateway_resident_passes_gate(self):
ex = self.make_exec()
ex.promote_to_resident("t1", "Paid")
# 常驻租户的网关任务过门不抛错(全程 dispatch 由既有
# 测试覆盖,这里只断言门本身放行)。
ex._check_gateway_vm_class(
self._task(project="t1"), model_provider="sub2api")
if __name__ == "__main__":
unittest.main()