Transitional Architecture
Temporary integrations can enable earlier delivery. Their purpose, consumers, operating cost, recovery obligations, and removal conditions need management.
-
-
Strangler Fig
Moving capabilities is only part of replacement. Plan intermediate contracts, data authority, recovery, and the dependencies that must end before retirement.
-
Architecture Principle
A usable architecture principle explains its direction, scope, and authority, with clear criteria for compliance or tradeoffs.
-
Target Architecture
The target describes the intended arrangement. Gap analysis identifies changes; the transition plan orders them into feasible intermediate states.
-
Durable Execution
If a process dies halfway through a five-step task, what remembers that steps one to three are done? Durable execution records it, and resumes from there.
-
Technical Debt
Technical debt needs a concrete limitation and evidence of extra work. Normal operating cost and the number of technologies alone do not measure it.
-
Reference Architecture
A runnable example can test a reference design, but does not guarantee its suitability. Maintain assumptions, validation, support, and consumer updates.
-
Two-Phase Commit
A prepared transaction keeps its locks until it is told the answer. A recovered coordinator replays its decision; until then the rows stay held.
-
Compensating Transaction
Cancel the reservation, do not put the number back. Between the booking and the cancellation, other people booked too.
-
Saga
A saga exposes intermediate results. Compensation and forward recovery need explicit policies, and affected observers may need separate corrections.
-
API Gateway
A gateway checks that the token is valid and the quota is not spent. Which invoices belong to whom is a question its backend should answer.
-
Idempotency Key
Reuse a durable operation key within the provider’s retention window. A recovered response may still leave the business outcome unresolved.
-
Problem Details
A client matching on your error message has built a dependency on your copywriting. The type URI exists so it does not have to.
-
Breaking Change
A change can break the supported contract or expose an unsupported assumption. Check both, including changes that silently alter results.
-
Modular Monolith
A modular monolith makes interfaces and data ownership explicit within one deployment. These boundaries reduce the work of service extraction.
-
Cloud Native
The definition names loose coupling and five operational qualities. Containers and microservices appear in it as examples, and explicitly as a partial list.
-
Stateless Service
Cache in the instance, keep truth outside it. The difference decides whether a restart is routine or an incident.
-
Data Fabric
Ask which capability is actually being offered — cataloguing, virtualization, lineage, policy enforcement — rather than whether to adopt the concept.
-
Two-Phase Commit Sink
A completed checkpoint records progress; a destination commit exposes output. The sink must recover safely when failure separates those two actions.
-
Unclean Leader Election
When no safe leader candidate remains, unclean election can restore availability from a lagging replica while losing acknowledged records. Check ISR and eligible leader replicas.
-
Narrow and Wide Dependencies
Some operations keep each partition’s work local; others need rows from everywhere. That difference is where stage boundaries and expensive recovery come from.
-
COST Metric
A system that scales beautifully may still be slower than one thread. COST names the hardware a platform needs before it beats a good single-threaded program.
-
Shared Database Integration
Letting several systems use the same tables removes copies to sync and seems to need no interface — until one team changes a column others depended on.
-
Bounded Context
“Customer” means one thing to billing and another to support. A bounded context is where a term has one meaning; at its edges, meaning must be translated.
-
Sensitivity and Tradeoff Points
Some design decisions quietly control a quality like latency; others affect several qualities at once, sometimes in opposite directions. ATAM names both so they get measured.
-
ATAM (Architecture Tradeoff Analysis Method)
ATAM evaluates an architecture by turning quality goals into concrete scenarios and finding the decisions they hinge on: risks, sensitivity and tradeoff points.
-
Multi-Tenancy
Multi-tenancy lets several teams or customers share one system. The design question is what they share, what stays separate, and what that separation costs.
-
Optimistic Concurrency Control
Optimistic concurrency control lets writers prepare changes without holding locks throughout the work, then validates and commits atomically. Conflicts can require a retry.
-
Raft
Raft gets several servers to agree on a shared state and keeps working while a majority survives — designed above all to be understandable.
-
Merkle Tree
A hierarchy of hashes lets two replicas locate exactly which ranges differ — without sending or comparing every row they hold.
-
Commit Wait
Spanner waits until TrueTime establishes that the commit timestamp is in the past. Clock uncertainty and concurrent commit work determine the extra delay.
-
TrueTime
TrueTime returns an interval, not an instant — turning clock uncertainty into a bounded number a system can reason about rather than assume away.
-
Lamport Clock
A counter passed along with messages orders events by what could have influenced what — no physical clock, and no synchronization required.
-
Quorum
A quorum sets how many replicas must respond. Overlap gives read and write sets a shared replica, while retention and version-selection rules determine what a read returns.
-
CAP Theorem
CAP forbids perfect consistency with perfect availability during a partition — a narrower claim than the “pick two of three” framing suggests.
-
PACELC
PACELC adds what CAP omits: when there is no partition, you are still trading latency against consistency — and that is the trade you live with daily.
-
Linearizability
Linearizability preserves real-time order between non-overlapping operations. Concurrent reads and writes can be ordered within their execution intervals.
-
Causal Consistency
Operations that could have influenced one another are seen in order; genuinely concurrent ones may be seen differently by different readers.
-
Last Writer Wins
Last writer wins can make replicas converge under an agreed ordering rule. Preserving the discarded write and conflict history requires a separate mechanism.
-
Eventual Consistency
Replicas converge on a record once updates to it stop — with no bound on how long, and no statement about what a reader sees meanwhile. Both gaps matter.
-
Partial Failure
Some parts fail while others keep working, and a caller often cannot tell a lost request from a lost reply. That ambiguity is what makes it hard.
-
Quality Attribute
A quality attribute says how well a system performs, not what it performs. It becomes useful once the target, the boundary, and the criterion for judging it are stated.
-
C4 Model
The C4 model describes software at four zoom levels and keeps deployment separate, so structure and where it runs stay two distinct questions.
-
4+1 Architectural View Model
Kruchten’s 4+1 model describes a system through four views plus scenarios, keeping structure, concurrency, modules, and deployment as separate concerns.
-
Solution Architecture
Solution architecture focuses on delivering a particular business capability. It may concern one system or changes across several, with explicit constraints and ownership.
-
Enterprise Architecture
Enterprise architecture links organizational strategy to architectural decisions at several scopes, setting principles and standards that lower-level designs can trace to outcomes.
-
Amdahl’s Law
Amdahl’s law shows how serial work limits parallel speedup. Added coordination costs can make more workers slower rather than merely less useful.
-
Lease
Lease expiry lets another worker recover ownership after a failure. Fencing tokens make the protected resource reject writes from a stale holder.
-
Token Bucket
A token bucket admits requests while it has tokens and refills at a constant rate, so capacity sets the largest burst and the refill rate sets the long-run average.
-
Deadline Propagation
Deadline propagation gives downstream calls the time remaining from the original request budget, replacing unrelated per-layer timeouts.