SLAs, limits and quotas

Everything a procurement or architecture review needs, in one place. Three sections deliberately kept separate: our commitment to you, Cloudflare's commitment to us, and the hard technical limits neither of us can change by agreement.

1. Spikefrost service level agreement

Applies to app serving — your deployed application responding to requests on its platform URL or custom domain.

Term	Commitment
Monthly uptime	99.9%
Allowed downtime	43 minutes 49 seconds per month
Measurement	Monthly, per app, based on failed requests attributable to the platform
Remedy	Service credits, below

Service credits

Monthly uptime achieved	Credit against that month's fee
99.0% – 99.9%	10%
95.0% – 99.0%	25%
Below 95.0%	50%

Support response

Severity	Definition	First response	Coverage
Sev-1	Service unavailable	1 hour	24/7
Sev-2	Major feature degraded	4 hours	Business hours
Sev-3	Question or minor issue	1 business day	Business hours

What the uptime commitment excludes

Standard, and stated so there are no surprises during an incident review:

• Faults in your application code, including returning your own 5xx responses
• Your configuration — firewall rules, blocked paths, maintenance mode, misconfigured bindings or secrets, custom domain DNS
• Exceeding a documented platform limit or quota
• Suspension for non-payment, or for breach of acceptable use
• Scheduled maintenance announced in advance
• Force majeure, and failures in networks outside our and Cloudflare's control

Control-plane availability — deploys, the CLI, the dashboard, agent dispatch — is handled under the support severities above rather than the uptime credit, because a control-plane incident does not stop a deployed app from serving traffic. See Failure domains.

2. Cloudflare's underlying commitment

Spikefrost apps run on Cloudflare. Their commitment to us is a component of ours to you, not a substitute for it — and its scope is narrower than most people assume, so it's quoted rather than summarized.

Term	Value
Workers Service Level Objective	99.99% monthly uptime
How measured	Error Rate Percentage — non-custom 500, 502, 503, 504 responses from provisioned Workers, excluding throttled or rate-limited requests
Credits	99.9%–99.99% → 10% · 95%–99.9% → 25% · below 95% → 50%
Source	Cloudflare Workers SLA

The exclusion that matters most

The Service Level Objective does not apply to unavailability resulting from "…incorrect configuration of bindings (e.g., KV, Durable Objects, D1, or R2, etc.), mismanaging Workers secrets, or exceeding resource limits", nor from "unavailability or performance degradation of interdependent Cloudflare services" — including the CDN, DNS, WAF, and storage bindings.

In other words: the 99.99% covers the compute path, not the storage your app depends on. This is why our own commitment sits below it, and why Failure domains treats graceful degradation as a design requirement rather than an optimization.

3. Platform limits

Current Cloudflare limits on the paid plans Spikefrost apps run on. These are technical ceilings, not contractual terms.

Compute — Workers

Limit	Value
Memory per isolate	128 MB (JS heap + WebAssembly)
CPU time per request	30 s default, configurable to 5 min
Wall-clock time	Unlimited while the client is connected
waitUntil after response	30 s
Cron, queue consumer, DO alarm wall time	15 min
Global-scope startup budget	1 s
Script size	10 MB gzipped (64 MB uncompressed)
Subrequests per request	10,000
Environment variables / secrets	128 per Worker, 5 KB each

Database — D1

Limit	Value
Database size	10 GB
Storage per account	1 TB
Databases per account	50,000 (raisable by request)
Point-in-time recovery window	30 days
Queries per Worker invocation	1,000
Maximum query duration	30 s
Maximum row / string / BLOB size	2 MB
Columns per table	100
SQL statement length	100 KB
Rows per table	Unlimited within the size cap

Coordination — Durable Objects

Limit	Value
Storage per object (SQLite)	10 GB
Classes per account	500
Objects per class	Unlimited
CPU per request	30 s default, to 5 min
Alarm / background wall time	15 min
WebSocket message size	32 MiB (received)
Row / string / BLOB size	2 MB

Spikefrost-level quotas

Limit	Value	Why
App source tree	256 MB	Source only — media belongs in assets, data in the database
Analytics retention	3 months	Aggregate metrics, sampled
Request-log body capture	14 days, opt-in	Off by default; JSON and text only, secrets redacted
Environments per app	Unlimited	Created on first deploy
Queues per app	10	Plus a dead-letter queue each
Managed email senders per app	5	Transactional sending only

Note on the analytics dataset: Cloudflare samples Analytics Engine data automatically and does not guarantee retrieval of any individual record. Weight aggregates by the sampling interval. For exact events use request logs; for business outcomes use your app's own event feed in D1. See Observability and audit.

4. Compliance posture

Item	Status
SOC 2 Type II	Attested — report available under NDA
ISO 27001	Certified — certificate available on request
GDPR	Data processing agreement available; sub-processor list on request
Data residency guarantees	Not offered today. Compute runs in the Cloudflare location nearest the user
Bring-your-own Cloudflare account	Not available today
HIPAA	Not claimed

Where something isn't offered, it says so. If residency or dedicated tenancy is a hard requirement, raise it before you build — see Isolation and tenancy.

Recovery objectives

Summarized from Durability and recovery:

Scenario	RPO	RTO
Database corruption or bad write	~0 within the 30-day window	Minutes
Bad deploy	0 — code is versioned	~1 minute
Accidental file deletion	0 — every version retained	Seconds to minutes
Durable Object state loss	Unbounded — no point-in-time recovery	Unbounded unless mirrored to D1
KV value loss	Unbounded	Rebuild from source of truth

Verifying any of this yourself

Nothing here requires trusting the page:

sf deployments                  # deploy history and outcomes per environment
sf d1 history                   # available point-in-time bookmarks
sf logs --since 24h --errors    # what your app actually returned
sf analytics query "SELECT ..." # your own metrics
curl -sSI https://your-app.example.com/ | grep -i x-sf-   # version + cache state

Last reviewed: 30 July 2026. Cloudflare's published limits and SLA change; where this page and their documentation disagree, theirs is authoritative and we want to know so we can correct this.

Next

• Failure domains and resilience — designing for the exclusions above.
• Durability, backup and recovery — the restore drill behind the RTO figures.
• Isolation and tenancy — the security-review answers.