Files
cx-ui/backend/triagelib/cxbridge.py
Parham Monfared 8892144e0a
Some checks failed
build-and-deploy / test (push) Has been cancelled
build-and-deploy / image (push) Has been cancelled
build-and-deploy / deploy (push) Has been cancelled
Add shift handover and RunPod, and make CX-Tools work in a container
Handover
- The Confluence shift doc becomes the landing page: shift metadata, the
  top-of-page checks, key updates with their Zendesk/Jira refs and status, and
  the free-text comments. "Hand over shift" closes the shift, opens the next
  one and carries the live items across, dropping anything done or marked
  "remove at end of shift" - the retyping this replaces.
- The RunPod table on that page is read from live host state instead of being
  copied in by hand, with the six-colour key preserved.

RunPod
- GraphQL client keyed on CX_RUNPOD_API_KEY. The old console login is kept as a
  fallback but cannot run unattended: the account has 2FA, so Clerk verifies the
  password and then asks for an emailed code and never issues a session. That is
  the real cause of the "No active session found" failure, and the client now
  says so instead of failing opaquely. TOTP is supported if the account moves to
  an authenticator app.
- Hosts and their listing history are persisted, so "most problematic hosts" can
  be ranked and each machine has a timeline of who listed or unlisted it, with
  the Zendesk comment and the error hint.
- The unlisting emails are parsed for the error block (they arrive
  quoted-printable) and classified into a likely cause and a next step.

Zendesk and Jira
- Unlisting raises a Zendesk ticket that follows the format of RunPod's own
  email, keyed on the machine so one machine keeps one thread, posted as an
  internal note.
- Jira is split in two: the Infrahub/OIE instance and the RunPod/RMA one, which
  may be a different Atlassian site. Blank RunPod values fall back to the
  defaults rather than failing.

Running in a container
- CX-Tools reads its keys from 1Password, which needs a desktop app. Config is a
  dataclass whose lookups live in per-field default factories, so passing
  CX_INFRAHUB_TOKEN/CX_INFRAINSIGHT_TOKEN in means those factories never run and
  CX-Tools itself stays unmodified.
- CX-Tools reaches OpenStack with `docker exec <region>-osc`, so the image now
  carries the Docker client (the static binary, not the docker.io package) and
  compose mounts the host socket with group_add for it. Verified from inside the
  container: live OpenStack and Infrahub calls both succeed.

Also fixes Settings, which read the environment at class-definition time and so
ignored anything set afterwards; a fresh Settings() silently returned stale
values. Caught by the Jira scoping tests.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-06 08:21:59 +01:00

325 lines
12 KiB
Python

"""Read-only adapter over the CX-Tools (vmc) collectors.
CX-Tools is imported as an unmodified library: this module never writes to the
CX-Tools tree and only calls collectors and query helpers that read. Every
OpenStack subcommand this module can reach is checked against READ_ONLY_VERBS
before it runs, so a bug here cannot mutate a live instance.
"""
from __future__ import annotations
import os
import sys
import threading
from typing import Any, Optional
DEFAULT_CX_TOOLS_PATHS = (
os.environ.get("CX_TOOLS_PATH", ""),
os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), "CX-Tools"),
os.path.expanduser("~/scripts/CX-Tools"),
os.path.expanduser("~/scripts/cx-tooling"),
)
# OpenStack verbs the diagnosis path is allowed to reach. Anything that could
# change state (create/delete/set/unset/shelve/reboot/...) is absent on purpose.
READ_ONLY_VERBS = frozenset({"show", "list"})
class BridgeError(RuntimeError):
"""Raised when CX-Tools cannot be located, imported, or authenticated."""
def locate_cx_tools() -> str:
for candidate in DEFAULT_CX_TOOLS_PATHS:
if not candidate:
continue
path = os.path.abspath(os.path.expanduser(candidate))
if os.path.isfile(os.path.join(path, "cxlib", "__init__.py")):
return path
raise BridgeError(
"Could not find the CX-Tools checkout. Set CX_TOOLS_PATH to the directory "
"that contains cxlib/ and the vmc entry point."
)
_lock = threading.Lock()
_state: dict[str, Any] = {"path": "", "cx": None, "config": None}
def _import_cxlib(path: str):
if path not in sys.path:
sys.path.insert(0, path)
try:
import cxlib # noqa: PLC0415 (import is intentionally deferred)
except Exception as exc: # pragma: no cover - depends on the local checkout
raise BridgeError(f"Failed to import cxlib from {path}: {exc}") from exc
return cxlib
def _tokens_from_env() -> dict[str, str]:
"""API keys supplied directly, bypassing 1Password.
CX-Tools reads its keys from 1Password, which needs an interactive session
and a desktop app - neither exists in a container. `Config` is a dataclass
whose 1Password lookups live in per-field default factories, so passing the
values in means those factories never run. CX-Tools itself is unmodified.
"""
return {
"api_key": os.environ.get("CX_INFRAHUB_TOKEN", "").strip(),
"insight_api_key": os.environ.get("CX_INFRAINSIGHT_TOKEN", "").strip(),
}
def _silence_1password(path: str) -> None:
"""Stop the CX-Tools secret loader from reaching for `op`.
`Config.os_cmd` re-reads the merged config on every OpenStack call, which
would otherwise retry a sign-in that cannot succeed here. Marking the module
as already loaded makes those calls no-ops.
"""
if path not in sys.path:
sys.path.insert(0, path)
try:
import secrets_1password # noqa: PLC0415
except Exception:
return
secrets_1password._loaded = True
def bootstrap() -> tuple[Any, Any]:
"""Import cxlib and build the shared Config, loading secrets exactly once.
With CX_INFRAHUB_TOKEN / CX_INFRAINSIGHT_TOKEN set, credentials come from
the environment. Without them, CX-Tools falls back to 1Password, which is
what a laptop run does.
"""
with _lock:
if _state["config"] is not None:
return _state["cx"], _state["config"]
path = locate_cx_tools()
cx = _import_cxlib(path)
tokens = _tokens_from_env()
if tokens["api_key"]:
_silence_1password(path)
config = cx.Config(no_color=True, debug=bool(os.environ.get("CX_DEBUG")),
**{k: v for k, v in tokens.items() if v})
else:
config = cx.Config(no_color=True, debug=bool(os.environ.get("CX_DEBUG")))
if not config.api_key or config.api_key in {"REDACT", "REPLACE_WITH_API_KEY"}:
raise BridgeError(
"No Infrahub API key. Set CX_INFRAHUB_TOKEN (and CX_INFRAINSIGHT_TOKEN), "
"or run `op signin` in this shell for the 1Password path."
)
_state.update({"path": path, "cx": cx, "config": config})
return cx, config
def cx() -> Any:
return bootstrap()[0]
def config() -> Any:
return bootstrap()[1]
def cx_tools_path() -> str:
bootstrap()
return str(_state["path"])
def quiet_progress() -> Any:
"""A Progress object that renders nothing, for use off the terminal."""
c = cx()
return c.Progress(c.C(False), 1, enabled=False)
def _guard_openstack(args: list[str]) -> None:
verbs = [a for a in args if not str(a).startswith("-")]
if not any(v in READ_ONLY_VERBS for v in verbs):
raise BridgeError(f"Refusing to run a non-read-only OpenStack command: {' '.join(args)}")
def os_json(region: str, args: list[str], timeout: int = 90) -> tuple[bool, Any, str]:
"""Run a read-only `openstack ... -f json` command through CX-Tools."""
_guard_openstack(args)
return cx().os_json(config(), region, args, timeout=timeout)
# --- collectors -------------------------------------------------------------
def collect_vm(target: str, *, region: str = "", org_id: Optional[str] = None, ssh_timeout: int = 3) -> dict[str, Any]:
"""Full VM reconciliation: the same payload `vmc --json <target>` emits."""
return cx().collect_vm(
config(),
target,
region_arg=region or "",
org_id=org_id,
ssh_timeout=ssh_timeout,
include_ih_events=True,
include_volumes=True,
include_all_ih_events=True,
progress=quiet_progress(),
)
def collect_host(host: str, *, ssh_timeout: int = 3) -> dict[str, Any]:
"""Host reconciliation: the same payload `vmc --json --host <host>` emits."""
return cx().collect_host(
config(),
host,
ssh_timeout=ssh_timeout,
include_ih_events=True,
include_volumes=True,
progress=quiet_progress(),
)
def collect_vm_contacts(target: str, *, region: str = "", org_id: Optional[str] = None) -> dict[str, Any]:
return cx().collect_vm_contacts(
config(),
target,
region_arg=region or "",
org_id=org_id,
progress=quiet_progress(),
)
# --- targeted queries used by individual runbooks --------------------------
def openstack_events(region: str, openstack_id: str, limit: Optional[int] = 5) -> list[dict[str, Any]]:
ok, events, _raw = cx().server_event_list(config(), region, openstack_id)
if not ok:
return []
return events[:limit] if limit else events
def openstack_event_detail(region: str, openstack_id: str, request_id: str) -> dict[str, Any]:
ok, detail, _raw = cx().server_event_show(config(), region, openstack_id, request_id)
return detail if ok else {}
def failed_openstack_event(region: str, openstack_id: str, scan: int = 5) -> dict[str, Any]:
"""Return the most recent OpenStack event whose detail reports a failure.
The state runbooks all say "the most recent failed event is the thing to
escalate", so this walks recent events newest-first and returns the first
one whose result is not Success, together with its detail rows.
"""
c = cx()
for event in openstack_events(region, openstack_id, limit=scan):
request_id = c.event_request_id(event)
if not request_id:
continue
detail = openstack_event_detail(region, openstack_id, request_id)
if not detail:
continue
rows = dict((str(k), str(v)) for k, v in c.event_detail_rows(detail))
result = rows.get("Result", "")
if result and result.lower() != "success":
return {"request_id": request_id, "action": rows.get("Action", ""), "rows": rows}
return {}
def infrahub_events(infrahub_id: str, limit: Optional[int] = None) -> list[list[str]]:
c = cx()
ok, data, _raw = c.query_vm_events(config(), str(infrahub_id))
if not ok:
return []
return c.infrahub_event_rows(data, limit)
def host_health(region: str, host: str) -> dict[str, Any]:
"""Hypervisor, Nova service and OVS agent signals for one host.
This is the cheap subset of `vmc --host` - the runbooks' "Host Health
Checks" entry point - without collecting every instance on the host.
"""
c = cx()
cfg = config()
ok_hv, hv, raw_hv, hv_name = c.hypervisor_show_host(cfg, region, host)
if not ok_hv:
return {"ok": False, "error": raw_hv, "host": host, "region": region}
state = c.normalize_empty(c.first_present(hv, "state", "State", default=""))
status = c.normalize_empty(c.first_present(hv, "status", "Status", default=""))
disabled_reason = ""
if status.lower() == "disabled":
for candidate in dict.fromkeys([x for x in (c.normalize_empty(hv_name), host) if x]):
ok_svc, services, _raw = c.compute_service_list_host(cfg, region, candidate)
if ok_svc:
disabled_reason = c.disabled_reason_from_services(services)
if disabled_reason:
break
ovs: dict[str, Any] = {}
ok_agents, agents, _raw_agents = c.network_agent_list_host(cfg, region, host)
if ok_agents:
ovs = c.ovs_agent_summary(agents)
if ovs.get("agent_id"):
ok_show, detail, _raw_show = c.network_agent_show(cfg, region, str(ovs["agent_id"]))
if ok_show:
ovs["last_heartbeat_at"] = c.normalize_empty(
detail.get("last_heartbeat_at") or detail.get("Last Heartbeat At") or ovs.get("last_heartbeat_at")
)
return {
"ok": True,
"error": "",
"host": host,
"hypervisor_name": hv_name,
"region": region,
"nova_state": state or "N/A",
"nova_status": status or "N/A",
"disabled_reason": disabled_reason,
"uptime": c.host_uptime_summary(hv),
"aggregates": c.host_aggregates_summary(hv),
"ovs_alive": ovs.get("alive"),
"ovs_state": ovs.get("state"),
"ovs_last_heartbeat": ovs.get("last_heartbeat_at") or "",
"running_vms": c.normalize_empty(c.first_present(hv, "running_vms", "Running VMs", default="")) or "N/A",
"free_disk_gb": c.normalize_empty(c.first_present(hv, "free_disk_gb", "Free Disk GB", default="")) or "N/A",
"local_disk_free": c.normalize_empty(c.first_present(hv, "disk_available_least", "Disk Available Least", default="")) or "N/A",
}
def host_gpu_census(region: str, host: str) -> dict[str, Any]:
"""Sum GPU counts of every instance on a host.
Implements the ERROR-runbook check for the NUMA/PCI fault: "check if the
host is full prior to escalation - add the values after the x, if it = 8
then it is FULL".
"""
c = cx()
ok, rows, raw = c.server_list_on_host(config(), region, host)
if not ok:
return {"ok": False, "error": raw, "total_gpus": None, "instances": []}
total = 0
unknown = 0
instances: list[dict[str, str]] = []
for row in rows:
flavor = c.get_row_field(row, "Flavor", "flavor") or c.flavor_name_from_any(row)
count = c.gpu_count_from_flavor_name(flavor)
if count.isdigit():
total += int(count)
else:
unknown += 1
instances.append({
"name": c.get_row_field(row, "Name", "name") or "N/A",
"openstack_id": c.openstack_id_from_row(row) or "N/A",
"status": c.get_row_field(row, "Status", "status") or "N/A",
"flavor": flavor or "N/A",
"gpus": count,
})
return {
"ok": True,
"error": "",
"total_gpus": total,
"unknown_flavors": unknown,
"instances": instances,
}
def json_safe(obj: Any) -> Any:
return cx().json_safe(obj)