feat: live update log in UI and suppress lifecycle Telegram during SAC update

Poll remote update_script.log while SSH update runs; skip lifecycle notify
when host agent_update_state is running (SAC v0.5.1).
This commit is contained in:
PTah
2026-07-08 11:46:53 +10:00
parent 1639261cde
commit d47131cd9f
9 changed files with 157 additions and 5 deletions
+1
View File
@@ -567,6 +567,7 @@ def update_host_agent_via_ssh(
host, host,
title=title, title=title,
runner=run_ssh_monitor_update_action, runner=run_ssh_monitor_update_action,
poll_remote_log=True,
) )
except RemoteActionAlreadyRunningError as exc: except RemoteActionAlreadyRunningError as exc:
raise HTTPException(status_code=409, detail=str(exc)) from exc raise HTTPException(status_code=409, detail=str(exc)) from exc
+64 -1
View File
@@ -16,7 +16,7 @@ from app.services.agent_update import execute_agent_update_fallback
from app.services.agent_update_types import AgentUpdateFallbackResult from app.services.agent_update_types import AgentUpdateFallbackResult
from app.services.linux_admin_settings import get_effective_linux_admin_config from app.services.linux_admin_settings import get_effective_linux_admin_config
from app.services.agent_update_settings import get_effective_agent_update_config from app.services.agent_update_settings import get_effective_agent_update_config
from app.services.ssh_connect import iter_ssh_targets, run_ssh_monitor_update, SshCommandResult from app.services.ssh_connect import iter_ssh_targets, run_ssh_monitor_update, SshCommandResult, tail_ssh_monitor_update_log
logger = logging.getLogger(__name__) logger = logging.getLogger(__name__)
@@ -113,12 +113,61 @@ def _apply_ssh_update_result(db: Session, host: Host, result: SshCommandResult)
host.agent_update_last_error = (result.message or "SSH update failed")[:2000] host.agent_update_last_error = (result.message or "SSH update failed")[:2000]
_REMOTE_LOG_POLL_SEC = 2.0
def _poll_ssh_update_log_loop(
host_id: int,
*,
stop: threading.Event,
) -> None:
"""Периодически подтягивает tail update_script.log в hosts.remote_action для UI."""
while not stop.wait(_REMOTE_LOG_POLL_SEC):
session = SessionLocal()
try:
row = session.get(Host, host_id)
if row is None or row.agent_update_state != "running":
continue
cfg = get_effective_linux_admin_config(session)
if not cfg.configured:
continue
try:
targets = iter_ssh_targets(row)
except Exception:
continue
tail = ""
for target in targets:
try:
tail = tail_ssh_monitor_update_log(
target=target,
user=cfg.user,
password=cfg.password,
)
except Exception:
logger.debug("update log tail failed host_id=%s target=%s", host_id, target, exc_info=True)
continue
if tail:
break
payload = dict(row.remote_action or {})
if tail:
payload["output"] = tail
payload["message"] = "Выполняется обновление… (лог с хоста)"
row.remote_action = payload
session.commit()
except Exception:
logger.debug("remote log poll failed host_id=%s", host_id, exc_info=True)
session.rollback()
finally:
session.close()
def start_host_remote_action( def start_host_remote_action(
db: Session, db: Session,
host: Host, host: Host,
*, *,
title: str, title: str,
runner: ActionRunner, runner: ActionRunner,
poll_remote_log: bool = False,
) -> dict[str, Any]: ) -> dict[str, Any]:
host_id = int(host.id) host_id = int(host.id)
with _lock: with _lock:
@@ -133,6 +182,17 @@ def start_host_remote_action(
def _worker() -> None: def _worker() -> None:
session = SessionLocal() session = SessionLocal()
stop_poll = threading.Event()
poll_thread: threading.Thread | None = None
if poll_remote_log and runner is run_ssh_monitor_update_action:
poll_thread = threading.Thread(
target=_poll_ssh_update_log_loop,
args=(host_id,),
kwargs={"stop": stop_poll},
name=f"sac-remote-log-{host_id}",
daemon=True,
)
poll_thread.start()
try: try:
row = session.get(Host, host_id) row = session.get(Host, host_id)
if row is None: if row is None:
@@ -167,6 +227,9 @@ def start_host_remote_action(
} }
session.commit() session.commit()
finally: finally:
stop_poll.set()
if poll_thread is not None:
poll_thread.join(timeout=5.0)
session.close() session.close()
with _lock: with _lock:
_running.discard(host_id) _running.discard(host_id)
+36 -1
View File
@@ -4,7 +4,7 @@ import logging
from sqlalchemy.orm import Session from sqlalchemy.orm import Session
from app.models import Event, Problem from app.models import Event, Host, Problem
from app.services import email_notify, mobile_notify, telegram_notify, webhook_notify from app.services import email_notify, mobile_notify, telegram_notify, webhook_notify
from app.services.notification_cooldown import should_notify_event, should_notify_problem from app.services.notification_cooldown import should_notify_event, should_notify_problem
from app.services.notification_policy import get_effective_notification_policy from app.services.notification_policy import get_effective_notification_policy
@@ -111,8 +111,43 @@ def notify_daily_report(event: Event, *, db: Session | None = None) -> None:
_dispatch_lifecycle_channels(event, db=db, policy=policy) _dispatch_lifecycle_channels(event, db=db, policy=policy)
def _host_sac_update_running(event: Event, db: Session | None) -> bool:
"""Пока SAC выполняет agent-update на хосте — не слать lifecycle в Telegram."""
if db is None or not event.host_id:
return False
host = db.get(Host, event.host_id)
if host is None:
return False
if (host.agent_update_state or "").strip().lower() == "running":
logger.info(
"notify lifecycle skipped host update running host_id=%s event_id=%s",
host.id,
event.event_id,
)
return True
return False
def _lifecycle_suppress_notifications(event: Event, *, db: Session | None = None) -> bool:
"""Не слать TG при штатном SAC/cron update (lifecycle с trigger deploy_recycle)."""
if _host_sac_update_running(event, db):
return True
details = event.details if isinstance(event.details, dict) else {}
trigger = str(details.get("trigger") or "").strip().lower()
if trigger in ("deploy_recycle", "sac_update", "agent_update"):
logger.info(
"notify lifecycle skipped trigger=%s event_id=%s",
trigger,
event.event_id,
)
return True
return False
def notify_lifecycle(event: Event, *, db: Session | None = None) -> None: def notify_lifecycle(event: Event, *, db: Session | None = None) -> None:
"""Старт/стоп/reload агента — всегда в каналы SAC (кроме TG, если telegram_via=agent).""" """Старт/стоп/reload агента — всегда в каналы SAC (кроме TG, если telegram_via=agent)."""
if _lifecycle_suppress_notifications(event, db=db):
return
if _skip_notifications_for_hidden_event(event, db): if _skip_notifications_for_hidden_event(event, db):
return return
policy = get_effective_notification_policy(db) policy = get_effective_notification_policy(db)
+27
View File
@@ -9,6 +9,7 @@ from dataclasses import dataclass
from app.config import get_settings from app.config import get_settings
from app.models import Host from app.models import Host
SSH_MONITOR_UPDATE_LOG_PATH = "/var/log/update_script.log"
SSH_MONITOR_UPDATE_SCRIPT = "/opt/scripts/update_ssh_monitor.sh" SSH_MONITOR_UPDATE_SCRIPT = "/opt/scripts/update_ssh_monitor.sh"
SSH_MONITOR_UPDATE_DIR = "/opt/scripts/update" SSH_MONITOR_UPDATE_DIR = "/opt/scripts/update"
SSH_MONITOR_REPO_NAME = "ssh-monitor" SSH_MONITOR_REPO_NAME = "ssh-monitor"
@@ -549,3 +550,29 @@ def run_ssh_monitor_update(
exit_code=updated.exit_code, exit_code=updated.exit_code,
) )
return updated return updated
def tail_ssh_monitor_update_log(
*,
target: str,
user: str,
password: str,
lines: int = 120,
) -> str:
"""Хвост /var/log/update_script.log на удалённом хосте (для live-лога в SAC UI)."""
safe_lines = max(20, min(int(lines), 400))
remote_cmd = (
f"test -r {SSH_MONITOR_UPDATE_LOG_PATH} && "
f"tail -n {safe_lines} {SSH_MONITOR_UPDATE_LOG_PATH} 2>/dev/null || true"
)
result = run_ssh_command(
target=target,
user=user,
password=password,
remote_cmd=remote_cmd,
command_timeout_sec=25,
need_root=True,
login_shell=False,
)
text = (result.stdout or "").strip()
return _truncate_output(text)
+1 -1
View File
@@ -1,5 +1,5 @@
"""Единый источник версии SAC (API, health, логи, OpenAPI).""" """Единый источник версии SAC (API, health, логи, OpenAPI)."""
APP_NAME = "Security Alert Center" APP_NAME = "Security Alert Center"
APP_VERSION = "0.5.0" APP_VERSION = "0.5.1"
APP_VERSION_LABEL = f"{APP_NAME} v.{APP_VERSION}" APP_VERSION_LABEL = f"{APP_NAME} v.{APP_VERSION}"
+2 -1
View File
@@ -32,7 +32,8 @@
- [ ] При **shutdown** / `SIGTERM` не увеличивать `sac-fail.count` (или отдельный флаг «update in progress»). - [ ] При **shutdown** / `SIGTERM` не увеличивать `sac-fail.count` (или отдельный флаг «update in progress»).
- [ ] После успешного heartbeat сбрасывать fail counter агрессивнее (уже частично есть). - [ ] После успешного heartbeat сбрасывать fail counter агрессивнее (уже частично есть).
- [ ] Watchdog: не слать Telegram при штатном restart во время SAC-update (флаг/state file от updater) — **отдельно от** подписи сервера (2.1.9). - [x] Watchdog: не слать Telegram при штатном restart во время SAC-update (state file `/var/lib/ssh-monitor/agent-update-in-progress` от updater) — **2.2.1-SAC**
- [x] Lifecycle stop/start при SAC-update: подавление на агенте + state file — **2.2.1-SAC**
- [ ] Документировать рекомендуемый `SAC_TIMEOUT_SEC` при тяжёлом ingest (сейчас 45s). - [ ] Документировать рекомендуемый `SAC_TIMEOUT_SEC` при тяжёлом ingest (сейчас 45s).
--- ---
+13 -1
View File
@@ -19,8 +19,18 @@
</p> </p>
<p v-if="message && !loading" class="host-action-log-message" :class="messageClass">{{ message }}</p> <p v-if="message && !loading" class="host-action-log-message" :class="messageClass">{{ message }}</p>
<p v-else-if="message && loading" class="muted host-action-log-sub host-action-log-message">{{ message }}</p> <p v-else-if="message && loading" class="muted host-action-log-sub host-action-log-message">{{ message }}</p>
<pre v-if="output" class="host-action-log-output">{{ output }}</pre> <pre v-if="output || (loading && logVisible)" class="host-action-log-output">{{
output || "Ожидание лога с хоста\n(обновление /var/log/update_script.log)"
}}</pre>
<div class="host-action-log-actions"> <div class="host-action-log-actions">
<button
v-if="loading"
type="button"
class="secondary"
@click="emit('toggle-log')"
>
{{ logVisible ? "Скрыть лог" : "Показать лог" }}
</button>
<button v-if="loading" type="button" class="secondary" @click="emit('close')"> <button v-if="loading" type="button" class="secondary" @click="emit('close')">
Свернуть Свернуть
</button> </button>
@@ -42,10 +52,12 @@ const props = defineProps<{
ok: boolean | null; ok: boolean | null;
message: string; message: string;
output: string; output: string;
logVisible: boolean;
}>(); }>();
const emit = defineEmits<{ const emit = defineEmits<{
close: []; close: [];
"toggle-log": [];
}>(); }>();
const messageClass = computed(() => { const messageClass = computed(() => {
@@ -9,7 +9,9 @@
:ok="entry.ok" :ok="entry.ok"
:message="entry.message" :message="entry.message"
:output="entry.output" :output="entry.output"
:log-visible="entry.logVisible"
@close="closeHostRemoteActionLog(entry.id)" @close="closeHostRemoteActionLog(entry.id)"
@toggle-log="toggleHostRemoteActionLog(entry.id)"
/> />
</div> </div>
</template> </template>
@@ -20,6 +22,7 @@ import HostActionLogModal from "./HostActionLogModal.vue";
import { import {
closeHostRemoteActionLog, closeHostRemoteActionLog,
hostRemoteActionLogs, hostRemoteActionLogs,
toggleHostRemoteActionLog,
} from "../composables/useHostRemoteAction"; } from "../composables/useHostRemoteAction";
const openLogs = computed(() => hostRemoteActionLogs.filter((e) => e.open)); const openLogs = computed(() => hostRemoteActionLogs.filter((e) => e.open));
@@ -16,6 +16,7 @@ export type HostRemoteActionLogEntry = {
id: string; id: string;
hostId: number; hostId: number;
open: boolean; open: boolean;
logVisible: boolean;
loading: boolean; loading: boolean;
ok: boolean | null; ok: boolean | null;
title: string; title: string;
@@ -180,6 +181,7 @@ export async function pollHostRemoteAction(
id: newLogId(), id: newLogId(),
hostId, hostId,
open: true, open: true,
logVisible: false,
loading: true, loading: true,
ok: null, ok: null,
title, title,
@@ -229,6 +231,7 @@ export async function runHostRemoteAction(
id: newLogId(), id: newLogId(),
hostId, hostId,
open: true, open: true,
logVisible: false,
loading: true, loading: true,
ok: null, ok: null,
title, title,
@@ -246,6 +249,13 @@ export async function runHostRemoteAction(
} }
} }
export function toggleHostRemoteActionLog(entryId: string) {
const entry = findLogById(entryId);
if (!entry) return;
entry.logVisible = !entry.logVisible;
entry.open = true;
}
export function closeHostRemoteActionLog(entryId: string) { export function closeHostRemoteActionLog(entryId: string) {
clearAutoCloseTimer(entryId); clearAutoCloseTimer(entryId);
const entry = findLogById(entryId); const entry = findLogById(entryId);