fix: skip unhealthy direct ollama probes
All checks were successful
CD Pipeline / deploy (push) Successful in 1m10s

This commit is contained in:
OoO
2026-06-18 13:46:30 +08:00
parent ee50e440ce
commit bd942e9427
6 changed files with 132 additions and 5 deletions

View File

@@ -211,6 +211,84 @@ def _host_label_for_embedding_health(host: str) -> str:
return ''
def _host_label_for_direct_health(host: str) -> str:
"""Map only direct GCP Ollama URLs to host_health_probes labels."""
if not host:
return ''
if '34.143.170.20:11434' in host:
return 'Primary (GCP)'
if '34.21.145.224:11434' in host:
return 'Secondary (GCP)'
return ''
def _recent_direct_host_unhealthy(host: str) -> bool:
"""Skip recent unhealthy direct GCP endpoints before trying proxy rescue.
This is deliberately direct-host only. Proxy rescue URLs must still be
probed even when the direct GCP endpoint is unhealthy.
"""
if not _env_flag('OLLAMA_RESOLVE_HOST_HEALTH_SKIP_ENABLED', True):
return False
host_label = _host_label_for_direct_health(host)
if not host_label:
return False
try:
window_minutes = int(os.getenv('OLLAMA_RESOLVE_HOST_HEALTH_SKIP_WINDOW_MINUTES', '20'))
except (TypeError, ValueError):
window_minutes = 20
window_minutes = max(1, window_minutes)
try:
from sqlalchemy import text as sa_text
from database.manager import get_session
session = get_session()
try:
row = session.execute(
sa_text("""
SELECT healthy, error_msg, probed_at
FROM host_health_probes
WHERE host_label = :host_label
ORDER BY probed_at DESC
LIMIT 1
"""),
{'host_label': host_label},
).fetchone()
finally:
session.close()
except Exception:
logger.debug("[OllamaHost] direct host health skip fail-open for host=%s", host, exc_info=True)
return False
if not row:
return False
healthy, error_msg, probed_at = row[0], row[1], row[2]
if probed_at:
try:
now = datetime.now(probed_at.tzinfo) if getattr(probed_at, 'tzinfo', None) else datetime.now()
if now - probed_at > timedelta(minutes=window_minutes):
return False
except Exception:
logger.debug("[OllamaHost] could not evaluate host health probe age for host=%s", host, exc_info=True)
return False
if bool(healthy):
return False
logger.warning(
"[OllamaHost] skip recent unhealthy direct host=%s label=%s window=%sm error=%s",
host,
host_label,
window_minutes,
(error_msg or '')[:180],
)
return True
def _recent_embedding_host_unhealthy(host: str) -> bool:
"""Skip known-bad GCP embedding runtimes using recent host_health_probes rows.
@@ -507,9 +585,11 @@ def resolve_ollama_host(primary: str = OLLAMA_HOST_PRIMARY,
primary_proxy = _proxy_rescue_for_primary(primary)
secondary_proxy = _proxy_rescue_for_secondary(secondary)
primary_recent_unhealthy = _recent_direct_host_unhealthy(primary)
secondary_recent_unhealthy = _recent_direct_host_unhealthy(secondary)
# B4: primary 若被標 unhealthy先嘗試同順位 110 proxy再嘗試 secondary
if not _is_unhealthy(primary) and _is_reachable(primary):
if not _is_unhealthy(primary) and not primary_recent_unhealthy and _is_reachable(primary):
selected = primary
logger.info(f"[OllamaHost] Primary 主機可用: {primary}")
elif primary_proxy and not _is_unhealthy(primary_proxy) and _is_reachable(primary_proxy):
@@ -518,7 +598,7 @@ def resolve_ollama_host(primary: str = OLLAMA_HOST_PRIMARY,
"[OllamaHost] Primary direct 不可用,使用 110 primary proxy: %s",
primary_proxy,
)
elif not _is_unhealthy(secondary) and _is_reachable(secondary):
elif not _is_unhealthy(secondary) and not secondary_recent_unhealthy and _is_reachable(secondary):
selected = secondary
logger.info(f"[OllamaHost] Primary 不可用,使用 Secondary: {secondary}")
elif secondary_proxy and not _is_unhealthy(secondary_proxy) and _is_reachable(secondary_proxy):