Durations & Time Intervals

TTL (Time to Live) — cache expiry

In cache configuration contexts, TTL values are almost always in seconds unless explicitly stated otherwise.

3,600 seconds = 60 minutes = 1 hour

Common TTL values and their meanings:

Value (seconds)	Duration	Typical use
60	1 minute	Rate limit windows, short-lived tokens
3,600	1 hour	API responses, user sessions
86,400	1 day	Static content, daily aggregates
604,800	1 week	User preferences, infrequently changing data

Vocabulary:

• "The cache TTL is [X] seconds — [Y hours/minutes]."
• "We increased the TTL from 300s to 3,600s — this will reduce upstream API calls significantly."
• "After TTL expiry, the next request triggers a cache miss and refreshes the value."

Exponential backoff — describing retry logic

The configuration describes exponential backoff:

• Attempt 1 fails → wait 1s (initial_backoff)
• Attempt 2 fails → wait 2s (1s × 2 = 2s)
• Attempt 3 fails → wait 4s (2s × 2 = 4s)
• All 3 retries exhausted → return error to caller

Why exponential backoff? Linear backoff (same wait every time) can cause thundering herd — all clients retrying simultaneously. Exponential backoff spreads the load by increasing wait time geometrically.

Jitter variant: Production retry logic often adds random jitter to the backoff: wait = backoff + random(0, backoff). This prevents correlated retries from multiple clients.

Retry vocabulary:

• "The client uses exponential backoff with [N] retries."
• "Retries are exhausted after [total wait time] — the request fails permanently."
• "We added jitter to the backoff to prevent correlated retries."
• "Idempotent endpoints can be retried safely; non-idempotent ones cannot."

Business hours SLA — clarifying the implication

"4 business hours" requires defining business hours to be meaningful:

• If business hours = 09:00–18:00 CET Mon–Fri:
• A P1 at 17:00 Friday: 1 remaining business hour Friday, then 3 more starting Monday at 09:00 → response by 12:00 Monday
• A P1 at 10:00 Monday: response expected by 14:00 Monday

Why the definition matters for customers: A customer experiencing a critical outage at 17:00 Friday expects to know they won't hear back until Monday morning — not that "4 hours" means tonight.

SLA time window vocabulary:

• "Business hours for this SLA are defined as [hours] [timezone] [days]."
• "4 business hours from receipt of the ticket."
• "This is a calendar-hours SLA — we respond 24x7 regardless of business hours."
• "The SLA clock starts when the ticket is created, not when the agent first reviews it."

JWT exp claim — Unix timestamp expiry

In JWTs (JSON Web Tokens), the exp claim is a Unix timestamp (seconds since epoch) indicating when the token expires.

To check expiry:

1. Get the current Unix timestamp (e.g. Math.floor(Date.now() / 1000) in JavaScript)
2. Compare: if current_time > exp, the token is expired

1716645137 ≈ May 25, 2024 — a token with this expiry has been expired for over a year (as of May 2026).

Other JWT time claims:

• iat: issued at — when the token was created
• nbf: not before — the token is not valid before this time
• exp: expiry — the token is not valid after this time

Vocabulary:

• "The JWT has expired — the exp claim was [date]."
• "Token expiry is set to [N] seconds after issuance."
• "We need to refresh the token before the exp timestamp."

Deployment observation window — what and why

Many post-deployment issues don't appear immediately:

• Gradual metric degradation: error rate or latency may increase slowly over minutes as load shifts to the new version
• Cold start effects: first few requests may be slow; the 5-minute trend shows the real steady-state
• Memory growth: a leak may not be visible in the first 2 minutes but apparent at 15
• Traffic ramp-up: if using canary or blue-green deployment, traffic may shift gradually

What to monitor during the window:

• Error rate (vs. baseline)
• p99 latency trend
• CPU and memory usage trajectory
• Alert status — any new firing alerts?

Vocabulary:

• "Initiating the 20-minute observation window."
• "Metrics are stable — declaring the deployment successful."
• "Error rate elevated during the window — rolling back."
• "Post-deploy observation: [metric] is within normal bounds."