This is a lookup table: see an error on the Logs page, find it here. Formats vary slightly across core versions - match on the keywords.

Connection errors

KeywordMeaning and fix
dial tcp ... i/o timeoutDestination (or node) timed out. Sporadic: ignore. Clustered: identify the node, switch, or run the all-timeout flow.
connect: connection refusedThe target port refused. Node port changed (update the subscription) or the destination service is down.
EOF / connection reset by peerThe far end cut the connection. One site: it blocks that node's IP - switch nodes. All sites: node or line interference.
no such hostName resolution failed. Check DNS config; or the domain is simply wrong/defunct.
use of closed network connectionNormal noise during switches (nodes changed, connections closed en masse). Ignore.
Sponsored Where does the subscription link come from? Our partner provider gives you 1 GB of high-speed Hong Kong data at signup — import it in one click. Get high-speed nodes

Startup and config errors

KeywordMeaning and fix
bind: address already in usePort conflict - see the port guide.
yaml: line N: mapping values are not allowedYAML syntax error at line N (indentation or a missing space after a colon, usually).
proxy N: unsupport proxy type / unknown fieldThe config contains a protocol or field this core doesn't know (subscription added newer node types). That node is skipped, the rest work; mass omissions → see the core-version guide.
rule ... error: proxy [X] not foundA rule references a nonexistent group name (renamed the group, not the rules) - the whole profile fails to load.
start tun interface errorTUN failed to start - see TUN issues.

Subscription and provider errors

KeywordMeaning and fix
provider ... initial failed / update failedA rule-set or node provider failed to download. Its URL usually needs the proxy: connect through any node, then reload the profile.
404 / 403 / 429 (during subscription update)Link dead, banned or rate-limited - see subscription failures.

Three habits for reading logs

  1. Read timestamps: only the seconds around your reproduction matter; stale errors mislead.
  2. Look for repetition: the same error against the same target, over and over, is diagnostic; a single line rarely is.
  3. Cross-check with Connections: log says timeout → the Connections row shows which rule and node carried it. When the two line up, the cause is in hand.
Tip: for anything not listed, search the English error text (minus IPs and ports) - all Clash-family clients share the core's messages, and community coverage is extensive.