From building a bus to running a fleet
Rekha has just become the first site-reliability engineer at YatraBus, an app for booking intercity bus tickets. The product works: people book seats. But it has grown from one server to 40 services, and the founders are tired of hearing about outages from angry customers on social media. (Rekha, YatraBus and every number here are made up for teaching.)
Her father runs a fleet of real buses, and he puts the difference this way: "Building a bus that starts is one job. Running a fleet on real roads is another." A fleet needs:
- Dashboards to see which bus is failing.
- Locks, so strangers can't drive off.
- A safe way to swap engines without stopping traffic.
- A depot that grows with demand.
- Someone watching the fuel bill.
Interviewers ask about this layer because it's where senior judgement shows. Rekha's first year covers all of it:
| Incident |
Idea it forces |
| "Customers told us we were down before our dashboards did" |
Observability: metrics, logs, traces |
| "Booking is slow, but which of ten services?" |
Tracing and correlation IDs |
| "The pager woke me at 3 am because CPU hit 80%" |
SLIs, SLOs, error budgets, symptom alerts |
| "A contractor's cloud key was found on GitHub" |
Zero-trust security and secrets |
| "Friday's deploy broke booking for 40 minutes" |
Deployment strategies |
| "Renaming one column took the site down" |
Expand, migrate, contract |
| "Should we move to Kubernetes?" |
Kubernetes essentials, and when not to |
| "A cloud region failed during Diwali travel week" |
Regions, RPO and RTO |
| "The cloud bill doubled in a quarter" |
Cost control |
This guide builds on Reliability, security and operations and Capacity estimation.
1. "Customers told us before our dashboards did": observability
YatraBus had dashboards, but they showed CPU and memory, which looked fine while payments were silently failing.
- Monitoring answers known questions: "Is latency above 300 ms?"
- Observability lets you ask new questions about unknown failures, using the data the system emits.
There are three signal types, plus events:
| Signal |
What it is |
Best for |
Watch out for |
| Metrics |
Numbers over time (counters, gauges, histograms) |
Alerting, trends, capacity |
High-cardinality labels explode storage |
| Logs |
Timestamped event records |
Detail about one event |
Volume and cost; unstructured text is hard to query |
| Traces |
The path of one request across services, as spans |
Finding which hop is slow |
Sampling decisions; propagation must be everywhere |
You've reached the free preview of this post.
Subscribe to the Blogs plan (₹99/mo) — or use any Algomi Suite, Career, Investor, or Trader plan — to keep reading.
See plans
