.lychee.toml

# Ignore transient failures on gnu.org (it sometimes refuses connections)
exclude = [
  "^https://www.gnu.org/software/make/?$",
  "^https://www.envoyproxy.io/",
  "^https://gateway-api-inference-extension.sigs.k8s.io/guides/serving-multiple-inference-pools-latest/",
  "^https://github.com/kubernetes-sigs/gateway-api-inference-extension/blob/main/site-src/guides/metrics-and-observability.md"
]

# Per-request timeout in seconds. Sized for slow upstream docs sites
# (e.g. sigs.k8s.io subdomains under load); 20s was tight enough to fire
# spurious timeouts on otherwise-reachable links.
timeout = 60

# Retry settings. Lychee retries on connection errors and timeouts.
# Worst-case per dead link with these values:
#   timeout * (max_retries + 1) + max_retries * retry_wait_time
#   = 60 * 3 + 2 * 2 = 184s.
# Lychee runs link checks concurrently (default 128 parallel), so total
# wall-clock is bounded by the slowest single link, not the sum across
# all links. Two retries is enough to ride out brief connection resets
# or 503s without piling extra waits on a truly-dead host.
max_retries = 2
retry_wait_time = 2

# Accept non-200 status codes (429: rate limits)
accept = [200, 429]