Extract managed-containerd proxy client

This commit is contained in:
Lyncoln
2026-09-17 11:43:00 -04:00
parent 9df5af71c0
commit e3099eb2b2
3 changed files with 156 additions and 199 deletions
+8 -198
View File
@@ -1,99 +1,13 @@
# Generic managed-containerd client helpers.
# Shadow calls are observational only and must never affect OCF action results.
MC_PROXY_SOCKET_DEFAULT="/run/user/1000/managed-containerd/managed-containerd.sock"
mc_proxy_shadow_monitor()
{
local authoritative_rc="$1"
local socket_path="${MC_PROXY_SOCKET:-${MC_PROXY_SOCKET_DEFAULT}}"
local resource="${OCF_RESOURCE_INSTANCE:-}"
local response
MC_PROXY_MODULE_DIR="$(
cd "$(dirname "${BASH_SOURCE[0]}")" &&
pwd
)"
resource="${resource%%:*}"
[ -n "${resource}" ] || return 0
[ -S "${socket_path}" ] || return 0
command -v python3 >/dev/null 2>&1 || return 0
response="$(
timeout --kill-after=1s 3s python3 - "${socket_path}" "${resource}" "${authoritative_rc}" <<'PY'
import json
import socket
import sys
path, resource, authoritative_raw = sys.argv[1], sys.argv[2], sys.argv[3]
authoritative_rc = int(authoritative_raw)
request = {"action": "monitor", "resource": resource}
s = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
s.settimeout(2.0)
s.connect(path)
s.sendall(json.dumps(request).encode("utf-8") + b"\n")
raw = b""
while b"\n" not in raw:
chunk = s.recv(65536)
if not chunk:
break
raw += chunk
s.close()
obj = json.loads(raw.decode("utf-8").strip())
if not obj.get("ok"):
print("ERROR|" + str(obj.get("error", "unknown proxy error")))
raise SystemExit(0)
result = obj.get("result") or {}
ha = result.get("ha") or {}
proxy_rc = ha.get("ocf_rc")
if proxy_rc is None:
comparison = "UNCLASSIFIED"
else:
comparison = "MATCH" if int(proxy_rc) == authoritative_rc else "MISMATCH"
receiver = result.get("wal_receiver") or {}
print("OK|{}|{}|{}|{}|{}|{}|{}|{}|{}|{}".format(
comparison,
"null" if proxy_rc is None else proxy_rc,
ha.get("health", "unknown"),
ha.get("ha_role", "unknown"),
ha.get("state", "unknown"),
str(ha.get("promotion_eligible", "unknown")).lower(),
result.get("role", "unknown"),
str(result.get("streaming", "unknown")).lower(),
receiver.get("slot_name") or "",
receiver.get("sender_host") or "",
))
PY
)" || response="TRANSPORT_ERROR"
case "${response}" in
OK\|MATCH\|*)
ocf_log info \
"managed-containerd shadow monitor: resource=${resource} authoritative_rc=${authoritative_rc} proxy=${response}"
;;
OK\|UNCLASSIFIED\|*)
ocf_log warn \
"managed-containerd shadow monitor unclassified: resource=${resource} authoritative_rc=${authoritative_rc} proxy=${response}"
;;
OK\|MISMATCH\|*)
ocf_log err \
"managed-containerd shadow monitor MISMATCH: resource=${resource} authoritative_rc=${authoritative_rc} proxy=${response}"
;;
ERROR\|*)
ocf_log warn \
"managed-containerd shadow monitor failed: resource=${resource} authoritative_rc=${authoritative_rc} proxy=${response}"
;;
*)
ocf_log warn \
"managed-containerd shadow monitor transport failure: resource=${resource} authoritative_rc=${authoritative_rc}"
;;
esac
return 0
}
MC_PROXY_CLIENT_DEFAULT="${MC_PROXY_MODULE_DIR}/proxy_client.py"
# Query managed-containerd for the authoritative monitor result.
@@ -133,116 +47,12 @@ mc_proxy_monitor()
response="$(
timeout --kill-after=1s 3s \
python3 - \
python3 \
"${MC_PROXY_CLIENT:-${MC_PROXY_CLIENT_DEFAULT}}" \
"${socket_path}" \
"${resource}" \
"${MC_PROXY_RC_TRANSPORT}" \
"${MC_PROXY_RC_HARD_FAILURE}" <<'PY'
import json
import socket
import sys
path, resource = sys.argv[1], sys.argv[2]
TRANSPORT_FAILURE = int(sys.argv[3])
PROTOCOL_FAILURE = int(sys.argv[4])
PROTOCOL_VERSION = 1
request = {
"protocol_version": PROTOCOL_VERSION,
"action": "monitor",
"resource": resource,
}
try:
s = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
s.settimeout(2.0)
try:
s.connect(path)
s.sendall(json.dumps(request).encode("utf-8") + b"\n")
raw = b""
while b"\n" not in raw:
chunk = s.recv(65536)
if not chunk:
break
raw += chunk
finally:
s.close()
except (socket.timeout, ConnectionError, OSError):
# Local daemon/control-plane reachability failure.
# Recent authoritative state may be used for bounded grace.
raise SystemExit(TRANSPORT_FAILURE)
# Transport succeeded. Invalid daemon output is therefore a
# protocol/result failure and must never be hidden by cached health.
if not raw:
raise SystemExit(PROTOCOL_FAILURE)
try:
obj = json.loads(raw.decode("utf-8").strip())
except (UnicodeDecodeError, json.JSONDecodeError):
raise SystemExit(PROTOCOL_FAILURE)
if not isinstance(obj, dict):
raise SystemExit(PROTOCOL_FAILURE)
# The daemon is reachable at this point, so any protocol mismatch is a hard
# control-plane failure. It must never be hidden by cached application health.
if obj.get("protocol_version") != PROTOCOL_VERSION:
raise SystemExit(PROTOCOL_FAILURE)
if not obj.get("ok"):
raise SystemExit(PROTOCOL_FAILURE)
if obj.get("resource") != resource:
raise SystemExit(PROTOCOL_FAILURE)
if obj.get("action") != "monitor":
raise SystemExit(PROTOCOL_FAILURE)
result = obj.get("result")
if not isinstance(result, dict):
raise SystemExit(PROTOCOL_FAILURE)
ha = result.get("ha")
if not isinstance(ha, dict):
raise SystemExit(PROTOCOL_FAILURE)
rc = ha.get("ocf_rc")
try:
rc = int(rc)
except (TypeError, ValueError):
raise SystemExit(PROTOCOL_FAILURE)
if rc not in (0, 1, 7, 8):
raise SystemExit(PROTOCOL_FAILURE)
receiver = result.get("wal_receiver") or {}
if not isinstance(receiver, dict):
raise SystemExit(PROTOCOL_FAILURE)
payload = "|".join([
str(rc),
str(ha.get("health", "unknown")),
str(ha.get("ha_role", "unknown")),
str(ha.get("state", "unknown")),
str(ha.get("promotion_eligible", "unknown")).lower(),
str(result.get("role", "unknown")),
str(result.get("streaming", "unknown")).lower(),
str(receiver.get("slot_name") or ""),
str(receiver.get("sender_host") or ""),
])
print(payload)
PY
"${MC_PROXY_RC_HARD_FAILURE}"
)"
rc=$?