arrca
arrca · RCA FOR KUBERNETES

Reduce your time to recover from an outage.

A continuously-updated topology of your Kubernetes estate built from the k8s API and OTel metrics and traces.

THE PROBLEM

Microservices at scale fail in ways dashboards can't see.

A REST API can be a simple HTTP client or a battle-tested system with timeouts, circuit breaking, stateful retries, idempotency and other fault-tolerant mechanisms.

01
Aggregated p99 is too coarse

A single endpoint usually spans many execution paths (three sequential DB calls give 2³ = 8 distinct branches depending on whether they opened a new connection or borrowed from a pool). Collapsing them into one number averages away exactly the variation that matters.

02
Retry storms

Poor retry policies and error handling can cause cascading failures, keeping services from recovering long after the original fault is gone.

03
The Distributed Monolith

A common microservices trap is services calling each other in cyclic or ad-hoc ways. This reintroduces all the tight coupling of a monolith, but now with the network latency and failure modes of a distributed system.

THE APPROACH

arrca is an observability platform designed to be cloud-native and microservices-first.

arrca-flows

Derive insights from your traces.

arrca collapses each assembled trace into a canonical variant of a root span.

That's what turns thousands of traces into a handful of recurring “interesting” structures you can triage instead of trying to understand the flow during an outage.

142 traces → 1 flow
checkout 1.31s
├ cart-svc 71ms
├ orders-api 96ms
├ payments-api 920ms
└ postgres 812ms
└ inventory 54ms
p99 1.24s ↑3×
arrca-graphs

Derive insights from your K8s topology.

arrca maps your pods, deployments, nodes, HPAs, services and custom CRDs and fuses in the call and dependency structure from live traces and metrics.

One navigable graph gives you a clear picture of single points of failure, cyclical calls, and poor upstream / downstream choices.

AI ROOT-CAUSE ANALYSIS

Debugging production is often like finding a needle-in-a-haystack problem.

arrca's AI agents look at real production data and distilled signals from your graphs and flows, then surface the most likely candidates for root-cause analysis.

arrca-agents · RCA candidates ranked by likelihood
01 payment → postgres 87%

connection pool exhausted

02 payment → pg-gateway 61%

error rate increased

03 cart-svc 28%

possible missing index on table 'cart_items'

NATIVE STACK COVERAGE

Kafka, MySQL, HTTP, gRPC, application runtimes (the JVM, goroutines), containers, nodes, pods, autoscalers and many more are all first-class entities in graphs and flows.

KafkaMySQLHTTPgRPCJVMgoroutinescontainersnodespodsautoscalers