Scaling Reliability Observability

single point of failure

component whose failure breaks system

replication

copies data or services across nodes

failover

switching to backup after failure

health check

test if service is alive

autoscaling

adjusting server count based on load

rate limiter

limits request volume per user or key

token bucket

rate limiting with refillable tokens

fixed window

rate limit over fixed time interval

sliding window

rate limit over moving interval

backpressure

slowing intake when system is overloaded

queue depth

number of waiting tasks

retry storm

many retries worsening outage

exponential backoff

wait longer between retries

idempotent operation

safe to retry without duplicate effect

SLA

service level agreement

SLO

service level objective

error budget

allowed amount of unreliability

logs

event records

metrics

numeric system measurements

traces

request path through services

alert

notification when system crosses threshold

dashboard

visual display of system health