Database Reference
In-Depth Information
tion, first advanced by the Advanced Institute of Science and Technology in Japan in 2004.
Accrual failure detection is based on two primary ideas: that failure detection should be flex-
ible by being decoupled from the application being monitored, and outputting a continuous
level of “suspicion” regarding how confident the monitor is that a node has failed. This is
desirable because it can take into account fluctuations in the network environment. Suspicion
offers a more fluid and proactive indication of the weaker or stronger possibility of failure
based on interpretation (the sampling of heartbeats), as opposed to a simple binary assess-
ment.
Failure
detection
is
implemented
in
Cassandra
by
the
org.apache.cassandra.gms.FailureDetector
class.
You can read the original Phi Accrual Failure Detection paper at
http://ddg.jaist.ac.jp/pub/
Faullt T
t Tollerance
Fault tolerance is the system's ability to continue operating in the event of a failure of one or
more of its components. Fault tolerance is also referred to as graceful degradation, meaning
that if the system operation degrades following a failure, the degraded performance is relat-
ive only to the failed component(s).
Gossiip
The gossiper is responsible for ensuring that all of the nodes in a cluster are aware of the im-
portant state information in the other nodes. The gossiper runs every second to ensure that
even nodes that have failed or are not yet online are able to receive node states. It is designed
to perform predictably, even at sharply increased loads. The gossip protocol supports rebal-
ancing of keys across the nodes and supports
Failure Detection
. Gossip is an important part
of the anti-entropy strategy.
The state information that the gossiper shares is structured as key/value pairs. In Cassandra,
the gossip protocol continues to gossip state information to other nodes until it is made ob-
solete by newer data.
When a server node is started, it registers itself with the gossiper. For more information,
check out the
org.apache.cassandra.service.StorageService
class.
Also see the Amazon paper on gossip at
http://www.cs.cornell.edu/home/rvr/papers/low-
