Connection resilience: what LangChain retries automatically and what it does not

TL;DR: Chat models retry failed requests with exponential backoff automatically: up to 6 times for network errors, 429 rate limits, and 5xx server errors. 401 and 404 are never retried.

The fix

1. Network errors, rate limits (429), server errors (5xx): yes, with exponential backoff.
2. Client errors like 401 unauthorized or 404: no. Retrying those is pointless; fix the credentials or the request.
3. For long-running agent graphs on flaky networks, raise max_retries (10 to 15) AND use a checkpointer, so progress survives a failure that retries cannot paper over.
4. The exceptions carry an is_retryable attribute. Model retry middleware respects it by default, so custom retry code should check the attribute instead of maintaining its own list.
5. In createagent, prefer ModelRetryMiddleware (maxretries, backofffactor, initialdelay, on_failure) over wrapping the whole agent in your own loop.

When to use this

• This covers exactly what the title says: Connection resilience.
• You are setting this up for the first time, or auditing an existing setup.
• You want the key gotchas in one place before you start.

When not to use this

• You are doing a different workflow with Connection; these steps are specific to the title above.
• You need the full reference docs; this is the short path, not the manual.

Compatibility

• Mentions 3.6 in the original; behavior may differ on much older or newer releases.