mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-11-16 02:46:38 +03:00
8215a3eab2
The rejoin rebuilt the warm history from every stored row, including the ones compaction archived. Undo and retry compare that view with the gateway's live rows and refused after restoring a compressed session. Skip archived rows, as the rollback and resume paths do.
173 lines
8.5 KiB
Python
173 lines
8.5 KiB
Python
"""Classic CLI snapshot restore: hand off the profile offline, then rejoin its authority."""
|
|
from __future__ import annotations
|
|
|
|
from contextlib import suppress
|
|
import logging
|
|
import os
|
|
|
|
from hermes_cli.cli_commands_session_tools import _pr, _t
|
|
from hermes_cli.gateway_client import GatewayClientError
|
|
from hermes_cli.gateway_profile_lifecycle import ack_confirmed, ack_failure
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
def restore_snapshot_command(cli, parts):
|
|
from hermes_cli.backup import list_quick_snapshots, quick_snapshot_dir, restore_quick_snapshot
|
|
from hermes_cli.gateway_tools_notify import gateway_store
|
|
if len(parts) < 3:
|
|
print(f" {_t('snapshot.usage_restore')}")
|
|
snaps = list_quick_snapshots(limit=1)
|
|
if snaps:
|
|
print(f" {_t('snapshot.most_recent', snapshot_id=snaps[0]['id'])}")
|
|
return
|
|
snap_id = parts[2]
|
|
try:
|
|
idx = int(snap_id)
|
|
except ValueError:
|
|
idx = None
|
|
if idx is not None:
|
|
snaps = list_quick_snapshots()
|
|
if not 1 <= idx <= len(snaps):
|
|
return print(f" {_t('snapshot.invalid_number', max=len(snaps))}")
|
|
snap_id = snaps[idx - 1]["id"]
|
|
store = gateway_store(cli)
|
|
chat = store is not None and getattr(cli, "_gateway_session_open", False)
|
|
# A remote gateway's state is not on this host: its chat restores this host's own snapshot offline,
|
|
# as before, and keeps the remote connection open.
|
|
gateway = chat and not os.environ.get("HERMES_TUI_GATEWAY_URL", "").strip()
|
|
try:
|
|
# An invalid/missing target must not take a working gateway offline.
|
|
if gateway and quick_snapshot_dir(snap_id) is not None:
|
|
restored = _restore_gateway_snapshot(cli, store, snap_id)
|
|
else:
|
|
local_db = getattr(cli, "_session_db", None)
|
|
if local_db is not None and not chat:
|
|
with suppress(Exception):
|
|
local_db.close()
|
|
cli._session_db = None
|
|
restored = restore_quick_snapshot(snap_id)
|
|
except (GatewayClientError, OSError, ValueError) as exc: # ValueError: a corrupt manifest.json
|
|
logger.warning("Snapshot restore failed: %s", exc, exc_info=True)
|
|
return _pr(f" {_t('snapshot.restore_incomplete', snapshot_id=snap_id)}", f" {exc}")
|
|
if restored:
|
|
print(f" {_t('snapshot.restored', snapshot_id=snap_id)}")
|
|
if not gateway:
|
|
print(f" {_t('snapshot.restart_recommended')}")
|
|
elif snap_id in {s.get("id") for s in list_quick_snapshots(limit=10**6)}:
|
|
print(f" {_t('snapshot.restore_incomplete', snapshot_id=snap_id)}")
|
|
else:
|
|
print(f" {_t('snapshot.not_found', snapshot_id=snap_id)}")
|
|
|
|
|
|
def _pause_owner(home, control_home, identity):
|
|
"""Hand the profile off; True when this restore parked a satellite and must serve it again."""
|
|
from gateway.control_socket import pause_gateway_for_update, request_unserve_profile
|
|
from hermes_cli.data_cleanup_holders import pause_acked
|
|
from hermes_cli.profiles import parked_marker_path
|
|
satellite = control_home != home
|
|
ours = False
|
|
if satellite:
|
|
marker = parked_marker_path(home)
|
|
# Persist intent BEFORE teardown: the periodic rescan must not re-serve it mid-restore.
|
|
try:
|
|
marker.touch(exist_ok=False)
|
|
ours = True
|
|
except FileExistsError:
|
|
pass # A profile the user parked stays parked and unserved.
|
|
accepted = False
|
|
try:
|
|
ack = request_unserve_profile(control_home, home.name)
|
|
accepted = ack_confirmed(ack, "unserved", home.name) or (
|
|
isinstance(ack, dict) and ack.get("pending") and not ack.get("error"))
|
|
finally:
|
|
if ours and not accepted:
|
|
marker.unlink(missing_ok=True)
|
|
else:
|
|
accepted = pause_acked(pause_gateway_for_update(control_home, timeout=8), identity)
|
|
if not accepted:
|
|
# Stopping the host gateway would take down every profile it multiplexes.
|
|
retry = (f"Park the profile with `hermes -p {home.name} gateway stop` and retry."
|
|
if satellite else "Stop the gateway and retry.")
|
|
raise GatewayClientError(f"Gateway refused the snapshot maintenance handoff; nothing was restored. {retry}")
|
|
return ours
|
|
|
|
|
|
def _restore_gateway_snapshot(cli, store, snap_id):
|
|
from gateway.control_socket import request_serve_profile_hot
|
|
from hermes_cli.backup import restore_quick_snapshot
|
|
from hermes_cli.data_cleanup_holders import manual_gateway_identity
|
|
from hermes_cli.gateway_connection import GatewayConnection
|
|
from hermes_cli.gateway_runtime import control_home_for, discover_gateway_endpoint
|
|
from hermes_cli.profiles import parked_marker_path
|
|
from hermes_cli.profiles_remove import wait_profile_unowned
|
|
from hermes_constants import get_hermes_home
|
|
from hermes_state_registry import close_all_under
|
|
home = get_hermes_home().resolve()
|
|
observed = discover_gateway_endpoint(home)
|
|
if observed.state != "ready" or observed.endpoint is None:
|
|
raise GatewayClientError("Local gateway is not ready for snapshot maintenance. Reconnect and retry.")
|
|
control_home = control_home_for(home, observed.endpoint).resolve()
|
|
identity = None
|
|
if control_home == home:
|
|
# A supervised owner is relaunched right after the pause and takes the profile back.
|
|
try:
|
|
identity = manual_gateway_identity(home)
|
|
except RuntimeError as exc:
|
|
raise GatewayClientError(f"Nothing was restored: {exc}.") from exc
|
|
store.close_reader()
|
|
close_all_under(home)
|
|
# A final viewer cannot redial from an idle poll while maintenance is replacing its database.
|
|
store.connection.close()
|
|
unserved = False
|
|
try:
|
|
unserved = _pause_owner(home, control_home, identity)
|
|
if not wait_profile_unowned(home):
|
|
raise GatewayClientError("Gateway still owns the profile; nothing was restored. "
|
|
"Wait for its running work to stop and retry.")
|
|
return restore_quick_snapshot(snap_id)
|
|
finally:
|
|
if unserved:
|
|
parked_marker_path(home).unlink(missing_ok=True) # serve-profile refuses a parked profile
|
|
ack = request_serve_profile_hot(control_home, home.name)
|
|
if not ack_confirmed(ack, "served", home.name):
|
|
logger.warning("Profile re-serve after snapshot restore: %s", ack)
|
|
print(f" Profile '{home.name}' is not served again yet: {ack_failure(ack)}. "
|
|
"The host retries on its next rescan (within 30s).")
|
|
store.snapshots.clear()
|
|
store.current = None
|
|
# The restore's own result or error decides the reply; a failed rejoin must not replace it.
|
|
try:
|
|
drained, store.connection = store.connection, GatewayConnection.shared()
|
|
# The relaunched owner revoked this CLI's launch-only key: every resume re-supplies it.
|
|
store.connection.launch_api_key = drained.launch_api_key
|
|
_rejoin_cli(cli, store)
|
|
except Exception as exc:
|
|
logger.warning("Rejoin after snapshot restore failed: %s", exc, exc_info=True)
|
|
print(f" Reconnecting to the gateway failed: {exc}. Run /new or restart the CLI.")
|
|
|
|
|
|
def _rejoin_cli(cli, store):
|
|
from hermes_cli.cli_init_mixin import _ROUTE_FLAGS, _gateway_launch, lease_handed_over
|
|
from hermes_cli.gateway_tools import bind_session
|
|
# Same-id restores still replace the connection and invalidate the cached proxies.
|
|
cli.agent = None
|
|
for attr in ("_goal_manager", "_loop_manager", "_heartbeat_manager"):
|
|
setattr(cli, attr, None)
|
|
# A snapshot older than the gateway's import of a legacy row makes this attach import it again.
|
|
with lease_handed_over(cli, getattr(cli, "_active_session_lease", None) is not None):
|
|
try:
|
|
snapshot = store.attach(cli.session_id)
|
|
except GatewayClientError as exc:
|
|
if str(exc) != "not_found":
|
|
raise
|
|
# The snapshot predates this session: only the owner may mint its replacement.
|
|
launch = {key: value for key, value in _gateway_launch(cli).items() if key not in _ROUTE_FLAGS}
|
|
store.create(launch, record_cwd=False)
|
|
snapshot = store.current
|
|
cli.session_id = snapshot["stored_session_id"]
|
|
cli.conversation_history = [{key: value for key, value in message.items() if key != "row_id"}
|
|
for message in snapshot.get("messages") or [] if not message.get("archived")]
|
|
bind_session(cli, store.connection)
|
|
getattr(cli, "_write_terminal_breadcrumb", lambda: None)()
|