Python — Errors and Recovery#

Failure is a typed outcome, not a silence. The rule matches every binding: only backpressure is safe to retry blindly.

Classed exceptions on the mesh#

BackpressureError and NotConnectedError are raised by the reliable mesh-stream send path on MeshNode — send_on_stream (and send_blocking) — not by the bus emit (which drops under load rather than raising). Catch them by type:

from net_sdk import MeshNode, BackpressureError, NotConnectedError
 
try:
    node.send_on_stream(stream, [payload])   # `stream` is an open MeshStream
except BackpressureError:
    # window full — the only blindly-retry-safe case
    ...
except NotConnectedError:
    # connection lost — a state change, not a retry
    ...

MeshNode.send_with_retry(...) retries BackpressureError for you (5 ms → 200 ms backoff), so prefer it over hand-rolling the loop above.

• BackpressureError — the ring buffer/window was full. Retry with backoff, or slow the producer. The only error a blind retry can fix.
• NotConnectedError — the session is gone. A state change; retrying won't undo it.
• Serialization / config failures are bugs — retrying reruns the bug.

The full taxonomy — the core ingestion/consumer/adapter errors plus subsystem errors (token, scaling, stream) — is the Error Codes reference.

Recover an nRPC call#

Retry, hedge, and circuit-breaker helpers wrap the raw call — the same three strategies as Rust and TypeScript. Calling a tool or service by name (rather than a pinned node id) lets the mesh pick a provider, so a substitutable capability fails over to a standby when the primary dies. The end-to-end patterns and the exact helper surface are in Recover a Failed Workflow and Typed RPC with nRPC.

The one rule#

Retry BackpressureError. Treat serialization/config errors as bugs, auth errors as "get a new credential," and NotConnectedError / closed streams as state changes retrying won't fix.