Field notes
Shorter learning logs, reading notes, and architectural sparks before they become essays.
Browse Notes
55 entries in the Cloud Sandbox.
Year
16 entries
25 Jul 2026
about 18 minutes min readSimple routing publishes the ordinary destination for a private DNS name. Failover routing adds an active-passive decision between prepared primary and secondary destinations. The right choice depends on the failure boundary, trustworthy branch signals, secondary readiness and client caching behaviour.
21 Jul 2026
about 14 minutes min readUsing a self-hosted observability platform does not require ignoring CloudWatch. For many AWS-managed services, CloudWatch is the native telemetry source. bfstore can export those signals while retaining control over correlation, retention, querying and operational experience.
6 Jul 2026
about 19 minutes min readRoute 53 private hosted zones already resolve through AmazonProvidedDNS inside VPCs that receive the zone. VPC Resolver endpoints and Global Resolver extend private DNS beyond that native VPC resolver view; neither is required for ordinary resolution inside an associated VPC.
27 Jun 2026
about 14 minutes min readEvery Amazon VPC already contains access to Route 53 VPC Resolver through AmazonProvidedDNS. Understanding its addresses, DHCP configuration, VPC attributes, private hosted zones, forwarding boundaries and query evidence makes AWS DNS architecture much easier to reason about.
23 Jun 2026
about 19 minutes min readKubernetes CPU limits are enforced as time-based cgroup quotas. A Go service can exhaust its allowance during a short burst and then be throttled, producing tail-latency spikes even when average CPU usage looks reasonable.
22 Jun 2026
about 16 minutes min readA golden path begins as a recommended route through software delivery. It becomes a platform when that route is discoverable, self-service, supported, observable and continuously improved through feedback from the teams using it.
8 Jun 2026
about 18 minutes min readAWS manages the NAT gateway service, but architects still choose its availability mode, routing scope, cost boundary and operational blast radius. Zonal, regional, distributed and centralised egress designs each move failure and expense rather than removing them.
12 May 2026
about 19 minutes min readIPv6 addresses are easier to read once their text-compression rules become mechanical. Remove leading zeros within each group, compress one longest run of zero groups, prefer the leftmost run when tied and expand to exactly eight groups when checking an address.
29 Apr 2026
about 15 minutes min readAn AWS route table is easier to understand when read as a destination lookup rather than a sequence of rows. Start with the packet destination, find the longest matching prefix, follow the selected target and then verify the onward and return paths.
8 Apr 2026
about 16 minutes min readMost day-to-day IPv4 subnetting does not require writing all 32 bits. A small set of calculations covering host bits, address counts, block increments, alignment and AWS reservations is enough to reason confidently about CIDR ranges.
26 Mar 2026
about 20 minutes min readEmergency AWS access should not be the easiest route to privileged authority. A strong break-glass process introduces deliberate activation friction, independent recovery paths, narrow roles, immediate alerts and a tested return to normal operation.
10 Mar 2026
about 16 minutes min readHumans and software may both receive temporary AWS credentials, but their identity journeys should remain visibly different. Workforce federation, interactive sessions and organisational groups solve different problems from workload trust, runtime roles and automatic credential delivery.
24 Feb 2026
about 21 minutes min readA dedicated AWS policy-test account provides a controlled place to exercise Service Control Policies with representative identities and workflows before those policies reach development, production or organisation-wide boundaries.
9 Feb 2026
about 16 minutes min readA service control policy defines the maximum permissions available inside an AWS member account. An action passing the SCP layer still needs a valid grant path and must survive every other applicable permission limit and explicit deny.
27 Jan 2026
about 14 minutes min readAWS organisational units should group accounts that require the same inherited controls and operational treatment. Their architecture lies in the governance contract accounts receive, not in how tidy the hierarchy looks.
11 Jan 2026
about 14 minutes min readThe AWS Organizations management account owns the organisation and several organisation-level operations. Keeping it quiet, tightly accessed and free of ordinary workloads reduces risk around an account that service control policies cannot constrain.
Year
7 entries
28 Dec 2025
about 16 minutes min readChange management does not need to mean queues, weekly approval meetings and ceremonial forms. Routine changes can follow tested, observable and pre-authorised paths, leaving human judgement for exceptional risk.
23 Nov 2025
about 15 minutes min readProvider-native architecture can preserve provider-specific identity, networking, data and security capabilities without exposing every cloud primitive to application teams. A platform can translate developer intent into precise, versioned and well-operated native implementations.
19 Oct 2025
about 17 minutes min readHigh-cardinality attributes can make telemetry precise and searchable, but uncontrolled dimensions create enormous metric series, storage costs and slow queries. Good observability preserves useful context without turning every unique value into a permanent index.
17 Aug 2025
about 17 minutes min readIdentity federation allows one system to rely on identity claims issued by another. The difficult work is deciding which issuer is trusted, which claims are accepted, how they map to local authority and how that trust is revoked.
15 Jun 2025
about 17 minutes min readRecovery point and recovery time objectives describe acceptable data loss and service interruption. A restore drill reveals whether the backups, infrastructure, access, procedures and people can actually meet them.
13 Apr 2025
about 12 minutes min readGuardrails preserve approved operating conditions, while gates decide whether work should cross important boundaries. Effective governance uses each where its control model fits rather than turning every safeguard into an approval queue.
9 Feb 2025
about 14 minutes min readCloud accounts, Kubernetes clusters and network segments contain different kinds of failure, compromise and operational mistakes. Resilient architecture depends on understanding what each boundary can stop, how it can be crossed and where shared dependencies reconnect otherwise isolated systems.
Year
6 entries
29 Dec 2024
about 12 minutes min readA developer platform should automate repetitive work and contain unnecessary infrastructure complexity without hiding the behaviour, risks and ownership engineers need to operate their services safely.
17 Nov 2024
about 27 minutes min readSelf-hosting platform tools can improve control, portability and learning, but it also transfers availability, security, upgrades, recovery and support into the organisation. The software may be free; operational ownership is not.
20 Oct 2024
about 17 minutes min readAn error budget translates a service-level objective into an explicit allowance for failure. It gives product, platform and service teams a shared basis for deciding when to continue delivery, reduce risk or prioritise reliability.
18 Aug 2024
about 18 minutes min readThe OpenTelemetry Collector creates a platform boundary between telemetry producers and storage systems. It can receive, process, route and export signals without making every application understand every backend.
14 Apr 2024
about 11 minutes min readA golden path makes the common, safe route through an internal platform easy to discover and use. It guides teams towards supported outcomes without pretending that every workload has identical needs.
11 Feb 2024
about 13 minutes min readA reconciliation loop repeatedly observes a system, compares reality with an expectation and acts on the difference. The same pattern appears in Kubernetes, GitOps, autoscaling, health management and operational recovery.
Year
7 entries
29 Dec 2023
about 14 minutes min readA reproducible cloud environment needs more than infrastructure code. Its inputs, state, identities, artefacts, policies, verification and recovery procedures must all be controlled well enough for the environment to be rebuilt and trusted.
26 Nov 2023
about 17 minutes min readA distributed trace reconstructs one operation from spans recorded by several processes. The story survives service boundaries only when trace context travels with the work.
22 Oct 2023
about 11 minutes min readBuilding an artefact once and promoting that exact object through each environment preserves the connection between source code, test evidence and the software that reaches production.
20 Aug 2023
about 17 minutes min readA role's trust policy controls who may obtain a session for it. Its permissions policies control what that session may request. Secure role design requires both questions to be answered narrowly.
18 Jun 2023
about 15 minutes min readTerraform state records which real infrastructure objects correspond to configuration addresses. Losing that state does not destroy the infrastructure, but it can leave Terraform unable to manage it safely.
16 Apr 2023
about 21 minutes min readImperative automation performs a sequence of steps. Desired-state automation records an outcome, observes reality and keeps reconciling drift after the original command has finished.
12 Feb 2023
about 25 minutes min readLinux namespaces give container processes isolated views of the system. Control groups group and account for tasks, allowing configured controllers to prioritise, protect or limit their use of shared resources.
Year
6 entries
27 Nov 2022
about 27 minutes min readThe transactional outbox closes the failure gap between committing local business state and recording the message that must be published because of it. Delivery, deduplication and workflow completion remain separate responsibilities.
30 Oct 2022
about 22 minutes min readA dead-letter queue protects the healthy processing path by isolating messages that cannot currently be handled. It only becomes a recovery mechanism when failures have owners, evidence and a safe route back into processing.
9 Oct 2022
about 30 minutes min readA traditional Kafka consumer group is one logical subscriber made from cooperating members. Kafka assigns each subscribed partition to one member at a time, while separate groups consume independently.
14 Aug 2022
about 38 minutes min readKafka preserves record order within a partition, not across an entire topic. Choosing a partition key therefore defines which events share an ordered history and which may be processed concurrently.
19 Jun 2022
about 33 minutes min readCommands ask an owner to attempt an action. Events report facts that an owner says have already happened. They may travel through the same broker, but they create different contracts.
13 Feb 2022
about 35 minutes min readGenerated Protobuf code makes messages convenient to use in one programming language, but the schema carries the enduring contract, wire identity and compatibility decisions.
Year
7 entries
19 Dec 2021
about 28 minutes min readRetrying repeats an operation after failure. Recovery is the larger job of restoring useful service, resolving uncertain state and repairing any consequences left behind.
21 Nov 2021
about 27 minutes min readA network failure can tell a caller that an interaction did not complete from its perspective without revealing whether the remote operation never started, is still running or already succeeded.
19 Sep 2021
about 24 minutes min readBreaking atomicity, consistency, isolation and durability into the concrete database promises they make around transactions, concurrency and failure.
18 Jul 2021
about 26 minutes min readSynchronous and asynchronous describe how work is coordinated, not whether the work itself is fast or slow.
23 May 2021
about 20 minutes min readScheduling a Bash script was easy. Making repeated, overlapping and partially failed runs safe was the real automation problem.
21 Mar 2021
about 21 minutes min readLooking beneath Bash's vertical bar to see the file descriptors, processes, buffering and exit statuses that make a pipeline work.
14 Feb 2021
about 20 minutes min readA small Bash experiment showing how quotes and backslashes decide whether characters become syntax, expansion, patterns or literal data.
Year
6 entries
20 Dec 2020
about 18 minutes min readA small exploration of standard output, standard error and the file descriptors that let the shell send command output to terminals, files and other processes.
11 Oct 2020
about 15 minutes min readA first look at DNS as a distributed system for answering questions about names, rather than simply a phone book for IP addresses.
19 Jul 2020
about 15 minutes min readA first attempt to understand TCP and UDP as different transport contracts rather than simply 'reliable' versus 'fast'.
17 May 2020
about 9 minutes min readA small Bash experiment that finally separated what a command prints from whether the command actually succeeded.
12 Apr 2020
about 10 minutes min readPoking around Linux's /proc filesystem to see what a running process looks like from the outside.
16 Feb 2020
about 6 minutes min readA small Linux experiment to pin down the difference between a program stored on disk and a process executing it.