Omission failure
An omission failure occurs when a process or communication channel fails to perform an expected message send, delivery or receive under the stated protocol model.
An omission model describes missing communication actions, not necessarily a stopped process. A process can omit one message and continue other work. Send omission, receive omission and channel loss name different boundaries: failure to send, failure to receive and loss after sending are not interchangeable diagnoses. Pure omission does not mean fabricating or altering message contents. [1]
The abstraction matters. Graversen, Montesi and Peressotti model receive omission broadly enough to cover a timeout or a receive attempted before its message is available. This is not proof that a physical network discarded a packet. A late message is not necessarily a permanently lost message. State the protocol's timing and communication assumptions before assigning a subtype. [1]
A caller's timeout alone cannot identify which boundary failed. RFC 5531 explains that receiving no RPC reply leaves execution uncertain, including with TCP. A request omitted before execution and a reply omitted after execution can leave the same caller-visible silence. A delayed request may also execute after the caller stops waiting: a local timeout does not by itself cancel remote work. Execution, an observed reply and durable storage are separate claims; persistence depends on the operation's storage contract. [2]
Crash-stop excludes all later steps by the crashed process; omission can be selective. In Chandra and Toueg's asynchronous crash model, channels are reliable but delays have no fixed bound; silence cannot distinguish a crashed process from one that is slow. Their failure detector is a separate abstraction, not proof of a particular communication fault. [3]
Relations
Limitations
- An observed missing reply does not establish non-execution, safe retry or durable completion. [2]
Citations
- Graversen et al. A Promising Future: Omission Failures in Choreographic Programming. arXiv:1712.05465v4 (2025), introduction and sections 2.2-2.3.
- RFC 5531: RPC, Remote Procedure Call Protocol Specification Version 2, section 5, Transports and Semantics.
- Chandra and Toueg. Unreliable Failure Detectors for Reliable Distributed Systems. JACM 43(2), 225-267, 1996; introduction and section 2.1.
Sources & caveats
- Primary-source synthesis of a failure-model concept; omission subtypes depend on the stated abstraction, not a packet-level diagnosis from a timeout.
- No network experiment, loss probability or device failure rate was measured for this entry.