Instance

The settings currently in use by this instance.

The SWIM.Member representing this instance, also referred to as “myself”.

Sequence numbers are used to identify messages and pair them up into request/replies.

Lifeguard IV.A. Local Health Multiplier (LHM)

These different sources of feedback are combined in a Local Health Multiplier (LHM). LHM is a saturating counter, with a max value S and min value zero, meaning it will not increase above S or decrease below zero.

The local health multiplier (LHM for short) is designed to relax the probeInterval and pingTimeout.

The value MUST be >= 0.

Creates a new SWIM algorithm instance.

Adjust the Local Health-aware Multiplier based on the event causing it.

Current SWIM protocol period (i.e. which round of gossip the instance is in).

Local Health Aware Suspicion timeout calculation, as defined Lifeguard IV.B.

Suspicion timeout is logarithmically decaying from suspicionTimeoutPeriodsMax to suspicionTimeoutPeriodsMin depending on a number of suspicion confirmations.

Suspicion timeout adjusted according to number of known independent suspicions of given member.

See: Lifeguard IV-B: Local Health Aware Suspicion

The timeout for a given suspicion is calculated as follows:

                                            log(C + 1) 􏰁
SuspicionTimeout =􏰀 max(Min, Max − (Max−Min) ----------)
                                            log(K + 1)

where:

• Min and Max are the minimum and maximum Suspicion timeout. See Section V-C for discussion of their configuration.
• K is the number of independent suspicions required to be received before setting the suspicion timeout to Min. We default K to 3.
• C is the number of independent suspicions about that member received since the local suspicion was raised.

Checks if a deadline is expired (relating to current time).

Create a gossip payload (i.e. a set of SWIM.Gossip messages) that should be gossiped with failure detector messages, or using some other medium.

Returns status of the passed in peer’s member of the cluster, if known.

Checks if the passed in peer is already a known member of the swim cluster.

Note: .dead members are eventually removed from the swim instance and as such peers are not remembered forever!

Returns specific SWIM.Member instance for the passed in peer.

Returns specific SWIM.Member instance for the passed in node.

Count of only non-dead members.

Count of all “other” members known to this instance (meaning members other than myself).

This is equal to n-1 where n is the number of nodes in the cluster.

Count of all members, including the myself node as well as any unreachable and dead nodes which are still kept in the membership.

Lists all members known to this SWIM instance currently, potentially including even .dead nodes.

Describes how a periodic tick should be handled.

Directs a shell implementation about how to handle an incoming .ping.

Directs a shell implementation about how to handle an incoming .pingRequest.

Describes how a pingRequest should be performed.

Only a single target peer is used, however it may be pinged “through” a few other members. The amount of fan-out in pingRequests is configurable by swim.indirectProbeCount.

Directs a shell implementation about how to handle an incoming .pingRequest.

This should be called on first successful (non-nack) pingRequestResponse

Directs a shell implementation about how to handle an incoming ping request response.

Indicates the gossip payload was processed and changes to the membership were made.

Directs how to handle the result of a confirmDead call.

Events which cause the modification of the Local health aware Multiplier to be adjusted.

The LHM is increased (in increments of 1) whenever an event occurs that indicates that the instance is not processing incoming messages in timely order.

It is decreased and decreased (by 1), whenever it processes a successful ping/ack cycle, meaning that is is healthy and properly processing incoming messages on time.

Allows for convenient adding of additional metadata to the SWIM.Instance.metadata.

While the SWIM.Instance is not meant to be logging by itself, it does offer metadata for loggers to use.