Service Level Agreements

SLA vs SLO: The Legal Line

 SLO: internal target. miss it → engineering process kicks in.
 SLA: external contract. miss it → CREDITS, penalties,
      potentially lost customers.

      ┌─────────────── your service ───────────────┐
      │  internal SLO target: 99.95%               │
      │  ── buffer of silence ──                   │
      │  external SLA promise: 99.9%               │
      └────────────────────────────────────────────┘

 THE BUFFER RULE: never set SLA = SLO.
 the gap between them is where you fix problems
 before they become contractual events.

What SLAs Actually Contain

 □ COMMITTED METRICS: availability %, latency bounds, 
   support response times (often the sharpest teeth!)
 □ MEASUREMENT METHOD: whose monitoring counts?
   how measured? exclusions? (maintenance windows, force majeure)
   — vendors negotiate these hardest; buyers should read closest
 □ REMEDIES: service credits tiered by breach depth:
     <99.9%:        10% credit
     <99.0%:        25% credit
     <95%:          exit rights / full refund month
 □ REPORTING: compliance dashboards, postmortem access

The Engineer’s Role in SLAs

 engineers must shape SLAs, not receive them:

 □ PROMISE ONLY WHAT YOU MEASURE: every committed number maps
   to an existing, trusted SLI pipeline. unmeasured promises
   = blind contracts.
 □ PRICE THE NINES: each nine multiplies architecture cost;
   sales promising 99.99% commits YOU to multi-region active-
   active budgets. get a seat at that table.
 □ DEFINE EXCLUSIONS PRECISELY: customer-caused failures,
   announced maintenance, upstream provider outages —
   vague exclusions burn goodwill; precise ones protect both sides.
 □ ERROR-BUDGET ALIGNMENT: internal budget stricter than SLA
   so breaches are preceded by internal alarms.

Reading Vendor SLAs (the reverse skill)

Clause	What to check
Availability definition	request-level or ping-level? which regions?
Measurement source	vendor’s own dashboard (conflict of interest!)
Credit mechanics	caps? claim windows? automatic or on-request?
Exclusions	broad enough to swallow most outages?
Support SLAs	response vs RESOLUTION time confusion

 veteran move: ask for HISTORICAL compliance data during
 procurement. past behavior predicts future downtime
 better than any promise.

Interview Framing

“Enterprise customer demands 99.99% in contract” scored response: translate to concrete cost (multi-region redundancy math), negotiate buffer structure (internal SLO stricter), define measurement + exclusions carefully, tie credits to real error-budget mechanics. Engineers who treat SLAs as business-department trivia get surprised by their own pager when legal signs something the architecture can’t hold.