{
  "slug": "perfect-failure-detector",
  "term": "Perfect failure detector",
  "aliases": [
    "class P failure detector"
  ],
  "definition": "A perfect failure detector eventually and permanently suspects every crashed process at every correct process, without suspecting any process before it crashes.",
  "note": "Class P combines strong completeness with strong accuracy in Chandra and Toueg's crash-stop model. A correct process never crashes during the modeled run; a crashed process does not recover. Strong completeness concerns every crashed process and every correct observer, not just one observer. [ct96]\n\nStrong accuracy forbids premature suspicion even of a process that will crash later. Completeness is eventual: the class alone specifies neither immediate detection nor a numerical detection deadline. An eventually perfect detector relaxes accuracy by allowing initial mistakes. [ct96]\n\nThese are properties of detector histories in a stated system model, not a certification obtained from a heartbeat sample. Arbitrarily delayed messages prevent a timeout alone from distinguishing a crash from a slow process in a fully asynchronous system. Timing assumptions must be supplied separately. [ct96]",
  "related": [
    "failure-detector",
    "eventually-perfect-failure-detector",
    "partial-synchrony",
    "flp-impossibility",
    "fail-stop",
    "crash-stop"
  ],
  "citations": [
    {
      "id": "ct96",
      "title": "Chandra and Toueg. Unreliable Failure Detectors for Reliable Distributed Systems. JACM 43(2), 225-267, 1996; sections 2.1-2.4 and 9.",
      "url": "https://www.cs.utexas.edu/~lorenzo/corsi/cs380d/papers/p225-chandra.pdf",
      "accessed": "2026-11-16"
    }
  ],
  "provenance": [
    "CT1996 sections 2.1-2.4 define the crash-stop model, completeness, accuracy and class P. This entry is a source synthesis, not a measured implementation or a recovery-protocol guarantee."
  ],
  "status": "unknown",
  "advantages": [],
  "limitations": [
    {
      "text": "These crash-stop process properties do not certify storage health, establish Byzantine fault tolerance, cover recovery after restart or enforce fencing against a stale writer.",
      "citationIds": [
        "ct96"
      ]
    }
  ],
  "supersedes": [],
  "supersededBy": [],
  "firstSeen": null,
  "lastSeen": null,
  "updated": "2026-11-20"
}
