How deeply does ScyllaDB in Action cover fault tolerance?

Explains replication, peer routing, eventual consistency, and acknowledgment thresholds, then verifies quorum behavior by stopping a node.

Fault tolerance

As explained in ScyllaDB in Action

Fault tolerance is the ability of a system to continue providing correct or available service despite component failures.

Fault tolerance is a database’s ability to keep serving requests when components fail. In ScyllaDB, it comes from running a cluster of peer nodes, distributing data among them, and storing multiple copies of each row. A request can succeed even when some replicas are unavailable, as long as its consistency requirement is met. section 1.0 section 1.2.1

Replication factor determines how many locations hold a row. Consistency level determines how many of those replicas must acknowledge a read or write. These are separate choices: replication supplies redundant copies, while consistency sets the number of responsive copies required for a successful query. section 1.2.1 section 1.4

Table 1.1 Sample of consistency level options, assuming a cluster with three replicas

Consistency level	Description	Number required to succeed	Failures tolerated
`ALL`	Requires all replicas to succeed	3	0
`QUORUM`	Requires a majority of replicas to succeed	2	1
`ONE`	Requires a single replica to succeed	1	2

With three replicas, ALL requires every replica and tolerates no failures; QUORUM requires two and tolerates one; ONE requires one and tolerates two. Lowering the required acknowledgments preserves availability during failures but raises the chance of inconsistent results. section 1.2.1

This availability has a trade-off. ScyllaDB favors eventual consistency: replicas may not agree at every instant, but the system converges toward correctness over time. Requiring ALL maximizes agreement for a query but makes a single unavailable replica prevent success; weaker levels improve availability by accepting less immediate consistency. section 1.1.3 section 1.2.2

Unlike a leader-follower design, ScyllaDB treats nodes as peers. A client may contact any node; the cluster maps rows to their owners and forwards requests as needed. This removes dependence on one central leader while replication protects data when an individual node is lost. section 1.2.1 section 1.1.3

A three-node failure test

Use a three-node cluster with replication factor 3 and QUORUM consistency. Stop one node, then run quorum queries: two replicas can still respond, so the cluster remains operational for those queries. The same setup cannot satisfy ALL while that node is down. section 2.4 section 1.2.1

Fault tolerance also shapes recovery. After a node crashes, ScyllaDB self-heals by streaming data from other nodes to restore the cluster. The design goal is availability during node loss, while consistency levels let an application decide how much immediate agreement each query needs. section 1.4 section 8.1

How ScyllaDB in Action covers it

liveBook manning.com

Read ScyllaDB in Action on liveBook Get the book at manning.com

Also explained in

liveBook manning.com

Akka in Action, Second EditionFault tolerance is the ability of a system to continue operating and preserve important services when components or operations fail.
liveBook manning.com

Bootstrapping Microservices, Second Edition: With Docker, Kubernetes, GitHub Actions, and TerraformFault tolerance is the ability of a system to continue providing acceptable service when some components fail.
liveBook manning.com

Apache Kafka in ActionFault tolerance is the ability of a system to continue operating correctly despite failures in components or infrastructure.
liveBook manning.com

Designing Deep Learning Systems: A guide for software engineersThe ability of a system to continue operating correctly or recover when components or operations fail.

Compare how 5 books explain fault tolerance →

Related concepts

transaction logaddressed by · A transaction log is an ordered record of transactions, transaction-related events, or resulting data changes, maintained as a history and often used for tracking or recovery.
resiliencecontrasts with · Resilience is the capacity of a system or organization to withstand disruption, adapt, recover, or continue functioning acceptably.
stretched clusterproduced by · A stretched cluster is a distributed cluster whose nodes are deployed across multiple physical locations or failure domains.
failureaddresses · A failure is an unsuccessful outcome or condition in which an operation, process, or system does not achieve an expected result.
human-in-the-loopimplemented by · Human-in-the-loop is a system pattern in which people provide input, judgment, review, or intervention within an automated process.

This explainer was generated by Manning's knowledge graph from the text of ScyllaDB in Action. The section links point to the parts of the book it draws on.