AGENT SKILL

0.2.1-understand-failure_detector

暂无简介

License 许可
MIT
最近更新
2026-10-09
Stars
★ 0
Forks
0

03 SKILL.md 正文

understand-failure_detector

CALL NUMBER: deep_consensus.failure_detector : distributed_consensus_and_replication_protocols(20), deep_raft(6) DEFINITION: Mechanism to distinguish crashed nodes from slow ones based on timeouts, heartbeats, or phi_accrual.

Invoke this skill to understand failure_detector down to its primitives. The RELATIVE ROOT below is the least-fixed-point closure of everything it bundles from — the full import cone, grouped by the lib each prim comes from. Projected from a prover-typed KB (MAP/SWI-Prolog consistency gate): every reference below resolves.

THE RELATIVE ROOT (the import cone, by lib)

from ?

  • agreement (d7): The property that a set of distributed processes reach the same decision outcome on a given value. Uniform agreement requires all processes including those that later crash to decide identically; asymmetric agreement permits different decision classes for different process roles such as primary versus backup.

from deep_consensus

  • failure_mode (d1): Classification of ways components can fail: crash_stop, crash_recovery, omission, timing, response, arbitrary_byzantine.
  • byzantine_failure (d2): Component exhibits arbitrary behavior including lies, omissions, or timing violations.
  • crash_failure (d2): Component ceases operation entirely without warning or recovery attempt.
  • network_partition (d2): Connectivity loss between node subgroups preventing inter-group communication.
  • shadow_operations (d3): Redundant execution paths validated against primary to detect silent corruption.
  • replica_synchronization (d4): Propagation of updates to ensure all replicas converge toward identical state.
  • availability (d4): Probability system remains operational and accessible despite failures.
  • safety (d4): Property requiring system never reaches inconsistent or invalid state.
  • gossip_dissemination (d5): Probabilistic epidemic protocol spreading updates through pairwise exchanges.
  • liveness (d5): Property requiring system to eventually respond or make progress under fair conditions.
  • quorum_read (d5): Reading from sufficient replicas to guarantee seeing latest write.
  • quorum_write (d5): Writing to sufficient replicas to guarantee subsequent readers see the value.
  • replication_factor (d5): Count of replica copies maintained for durability and availability.
  • strong_quorum (d5): Quorum where read and write quorums overlap (r + w > n).
  • view_change (d10): Transition of consensus membership to new configuration after membership modification.

from deep_raft

  • a005 (d5): Uniform agreement: all processes including those that later fail must decide on the same value, stricter than ordinary agreement which only binds correct processes.
  • a007 (d7): Quorum in agreement: a subset of processes sufficient to guarantee agreement can be reached, typically intersecting any valid quorum to ensure overlap.
  • a008 (d7): Threshold in agreement: the minimum number of correct processes required to guarantee safety and liveness under given failure assumptions.
  • a009 (d8): Asymmetric agreement: one class of processes (e.g., primary) must agree while another class (e.g., backups) may reach different decisions, used in primary-backup protocols.
  • a002 (d8): Validity in agreement: the decided value must have been proposed by some process in the system, guaranteeing agreements are grounded in actual proposals.
  • a004 (d9): Termination in agreement: every correct process eventually decides some value, ensuring the protocol makes progress and does not stall indefinitely.

from distributed_consensus_and_replication_protocols

  • state_machine_replication (d3): Technique where deterministic state machines are replicated across nodes via a common log.
  • recovery (d3): Process of restoring a failed node or the system to a consistent state.
  • split_brain (d3): Scenario where network partition causes multiple nodes to believe they are leader; resolved by quorum.
  • replicated_log (d4): Ordered log of commands maintained across replicas; foundation of state machine replication.
  • roll_back (d4): Reverting uncommitted changes to maintain consistency after transaction abort.
  • roll_forward (d4): Reapplying committed log entries to reconstruct state after failure.
  • quorum (d4): Subset of nodes sufficient for reads or writes; typically majority (n/2+1) for strong consistency.
  • anti_entropy (d5): Repair of divergent replicas by comparing Merkle trees and exchanging differences.
  • read_repair (d5): Background correction of stale replicas when read reveals outdated data.
  • linearizability (d5): Strongest consistency model: every operation appears atomic and ordered in real time across all nodes.
  • semi_synchronous_replication (d5): Write waits for quorum of replicas to acknowledge; balances latency and durability.
  • synchronous_replication (d5): Write waits for all replicas to acknowledge before returning to client.
  • state_transfer (d6): Sending full current state to a lagging replica for rapid catch-up.
  • eventual_consistency (d6): Given no new updates, all replicas will eventually converge to the same value.
  • consensus (d6): Agreement on a single value among a set of distributed processes; the core problem consensus protocols solve.
  • catch_up_recovery (d7): Process of syncing a lagging or new replica with current leader state.
  • fault_tolerance (d7): System's ability to continue operating correctly despite component failures.
  • reconfiguration (d8): Process of changing cluster membership while maintaining consistency guarantees.
  • membership_change (d9): Adding or removing nodes from the cluster without service interruption.
  • joint_consensus (d10): Raft's two-phase configuration change using joint configuration of old and new members.

CONSUMERS (what needs this)

failure_detection_timeout, heartbeat, phi_accrual_detector, preemptive_recovery


Projected from the distributed consensus and replication protocols KB (196 concepts / 187 relations) — consistency-typed by MAP; the facet list after the colon IS the cross-lib dependency web.

(leaf — this is an actual skill.)

04 同仓库更多 Skills