Skip to content
EncryptTechBD
Home Industries Our Work Blog Pricing About Contact Discuss Your Project
বাংলা (BN)
Cloud, DevOps & FinOps 6 min read

Observability costs: logs, metrics and traces in Bangladesh

Why the observability bill grows faster than traffic, how the three signals are priced differently, and a calculator you fill with your own volumes.

E
EncryptTechBD · 0 comments
Exterior of the Swiss National Supercomputing Centre
Photo: CSCS / Wikimedia Commons (BY-SA)

Observability spending has a habit of growing without a decision. A team adds a log line to debug a problem, keeps it because it might be useful, turns on debug logging in production during an incident and forgets to turn it off, and eighteen months later the monitoring bill is larger than the database bill.

The three signals behave differently. Logs are priced mostly by volume, metrics by time series and cardinality, and traces by spans and retention. A model that treats them as one line item will mislead you, so this guide separates them and gives you a worksheet rather than a headline figure.

What you are actually buying

  • Logs: text or structured events, usually billed per gigabyte ingested and again per gigabyte stored. Search and long retention add cost.
  • Metrics: numeric time series, usually billed by series count, retention and query volume. High-cardinality labels such as user ID or order ID multiply the series count quickly.
  • Traces: request paths made of spans. Cost rises with request volume, span count per request and sampling policy.
  • People and process: dashboards, alert rules, on-call time and the engineering hours to maintain instrumentation. This line is often larger than the licence and rarely written down.

The cost lines to budget

  1. Ingestion. Volume per day multiplied by the vendor rate. Debug logs, health checks and access logs are the usual hidden bulk.
  2. Retention. Hot storage for recent search, cold storage for compliance. Most teams need far less hot retention than they buy.
  3. Indexing and query. Some platforms charge by data scanned rather than data stored, which makes a wide dashboard expensive to open.
  4. Metric cardinality. Every unique combination of labels is a series. A label containing a customer identifier can create a series per customer.
  5. APM hosts or agents. Priced per monitored host, container or service instance, which rises with autoscaling.
  6. Trace volume. Spans per request times requests per month. Sampling is the main control, and tail-based sampling keeps the interesting traces.
  7. Data egress. Sending telemetry across regions or providers can attract transfer charges.
  8. Implementation and maintenance. Instrumentation, dashboards, alert tuning and upgrades.
  9. Noise. The cost of alerts nobody trusts, which produces slower response and eventually a second tool bought to fix the first.

The cost model

Build the estimate in four blocks and keep them separate.

Logs equal daily gigabytes multiplied by thirty, multiplied by the ingestion rate, plus retained gigabytes multiplied by the storage rate. Split production, staging and security logs, because they have different retention needs.

Metrics equal active series multiplied by the per-series rate, plus query cost. Estimate series by counting the label combinations on your busiest metric, not by counting metrics.

Traces equal requests per month multiplied by average spans per request, multiplied by the sampled percentage, multiplied by the per-span rate.

People equal the hours per month spent maintaining instrumentation and tuning alerts, multiplied by a loaded hourly rate, plus the on-call allowance.

Add a contingency for the first three months after a new service launches, when logging is at its most verbose and nobody has tuned it yet.

A calculator you can fill in

  1. Measure actual log volume. Take a week of production logs, including the peak day, and convert to gigabytes per day.
  2. Separate the noise. Estimate the share from health checks, access logs and debug statements. That share is your first saving.
  3. Count metric series. Query the platform for the number of active series and the top ten metrics by series count.
  4. Count request volume and spans. Use the tracing backend or the application's own request counter.
  5. Choose retention by purpose. Days for debugging, months for trend analysis, a year or more only where a regulation or contract requires it.
  6. Apply your contract rates. Use the rates in your own agreement, including any committed-use discount and overage price.
  7. Add people cost. Include the on-call allowance.

Run the worksheet at current volume and at twice current volume. The second number is the one that matters, because the architecture that is affordable today is often unaffordable after a successful campaign.

Where the savings usually are

  • Drop debug logs in production. Keep the level configurable and enforce it in the deployment template, not by memory.
  • Aggregate before ingestion. Count requests at the edge and ship a metric instead of a log line per request.
  • Sample traces. Keep all errors and a small percentage of successful requests. Tail-based sampling lets you decide after the trace completes.
  • Control cardinality. Never put user ID, order ID, email or raw URL into a metric label. Put them in logs or trace attributes instead.
  • Tier retention. Search hot data for days, not months. Move the rest to object storage with a lifecycle rule.
  • Give each team a budget. A monthly volume target per service changes behaviour faster than a central policy.

Questions to ask a vendor

  • How is each signal metered, and what exactly counts as a billable unit?
  • What happens at the retention boundary, and is retrieval charged separately?
  • Does the price change with cardinality, and is there a cap?
  • What discount applies to a committed volume, and what is the overage rate?
  • Can we export raw data without a proprietary agent?

A 90-day plan

  1. Days 1 to 30: measure volume by service and signal, identify the top three sources of ingest, and write the current monthly cost per service.
  2. Days 31 to 60: apply the cheap fixes: drop debug logs, remove health-check noise, cap cardinality, set retention tiers and delete unused dashboards.
  3. Days 61 to 90: introduce sampling, set per-team budgets, and renegotiate or resize the contract using measured volumes rather than vendor defaults.

Metrics for the observability budget

  • Cost per service per month, and cost per thousand requests.
  • Gigabytes ingested per day by signal and environment.
  • Active metric series, and the share held by the top ten metrics.
  • Trace sample rate and the percentage of errors retained.
  • Mean time to detect and mean time to resolve, which is what the spending is supposed to improve.

Local conditions

Most observability platforms bill in US dollars, while the business budgets in taka, so an exchange-rate movement changes the effective cost without any change in usage. International card limits and procurement approvals can also delay a plan upgrade during an incident. Where latency to a foreign region is a problem, check whether the vendor has a closer ingestion endpoint before assuming the application is slow.

Where this fits in our work

We set up monitoring, logging and alerting as part of Managed Technology & Cybersecurity, with scope and commercial terms on the pricing page. For the wider cost discipline, the FinOps guide for small teams covers how to run a monthly cloud cost review.

Sources

Services that deliver this

Want this implemented properly?

Pricing →

Every engagement starts with a written scope. If a requirement is outside it, we quote it separately before starting.

Responses (0)

No comments yet. Add the first one below.

Leave a response

Your email address will not be published. Required fields are marked *.

Keep reading

More field notes