Inside Managed Cloud Storage on ServersAreUp: Client-Side AES-256-GCM Double-Envelope Encryption, Azure Geo-Redundancy & WORM Immutability
On this page (6 sections)
- Why Server-Side Encryption Alone Is Not Enough for Regulated Cloud Storage
- Cryptographic Pipeline: Client-Side AES-256-GCM Envelope Encryption Over Azure 256-Bit SSE
- Why Unique DEKs, 96-Bit IVs, and 128-Bit GCM Tags Matter Mathematically
- Failure-Domain Engineering: LRS vs ZRS vs GRS vs RA-GRS
- FinOps Lifecycle Tiering (Hot, Cool, Archive) & Transparent Plan Pricing
- Complete Managed Cloud Storage Plans & Overages (/cloud-storage)
- Regulatory WORM Immutability Locks, Legal Holds & Scoped HMAC-SHA256 SAS URLs
- Unifying Managed Cloud Storage with BYO Azure Storage Accounts (*.blob.core.windows.net)
Why Server-Side Encryption Alone Is Not Enough for Regulated Cloud Storage
When an engineering team uploads database backups, healthcare records, financial ledgers, or AI training corpora to standard cloud object storage, default server-side encryption (SSE) only encrypts bytes after plaintext traverses the provider's frontend load balancers and memory buffers. If an upstream storage policy is misconfigured or an operator demands cryptographic proof that raw plaintext never touched third-party storage media, single-layer server-side encryption leaves an architectural gap.
With Managed Cloud Storage hosted via Microsoft Azure (/cloud-storage)—administered visually in /dashboard/blob-storage and programmatically through /api/v1/blob—ServersAreUp combines client-side AES-256-GCM authenticated envelope encryption performed before bytes leave our boundary with Microsoft Azure 256-bit SSE containers, multi-zone and multi-region physical replication (LRS, ZRS, GRS, RA-GRS), Hot / Cool / Archive lifecycle economics, and SEC 17a-4 / HIPAA / FINRA time-bound WORM immutability locks.
Cryptographic Pipeline: Client-Side AES-256-GCM Envelope Encryption Over Azure 256-Bit SSE
Every object sealed through the ServersAreUp envelope pipeline undergoes two independent layers of authenticated encryption before coming to rest on physical storage media:
[ Raw Object Stream ]
|
v
+-----------------------------------------------------------------------+
| Layer 1: SAU Envelope Encryption Engine (Before Egress to Storage) |
| 1. Mint fresh 256-bit Data Encryption Key (DEK) via CSPRNG |
| 2. Mint unique 96-bit (12-byte) Initialization Vector (IV) |
| 3. Encrypt payload -> Ciphertext + 128-bit (16-byte) GCM Auth Tag |
| 4. Wrap DEK under tenant Vault KEK (/api/v1/vault) -> WrappedDEK |
+-----------------------------------------------------------------------+
|
| TLS 1.3 Ciphertext Stream + Detached Envelope Metadata
v
+-----------------------------------------------------------------------+
| Layer 2: Microsoft Azure Blob Storage Container |
| - 256-bit AES Server-Side Encryption (SSE) at physical media layer |
| - Replicated across LRS / ZRS / GRS / RA-GRS fault domains |
| - Optional WORM Time-Bound Immutability Policy & Legal Hold |
+-----------------------------------------------------------------------+
Why Unique DEKs, 96-Bit IVs, and 128-Bit GCM Tags Matter Mathematically
Reusing a single static AES key across millions of objects under Galois/Counter Mode (AES-GCM) invites catastrophic nonce-collision exposure (2^32 birthday bound on random 96-bit nonces under a single key). Envelope encryption eliminates key-wear exhaustion completely:
- Per-Object 256-Bit
DEK: Every blob receives its own cryptographically random32-byteData Encryption Key (crypto.randomBytes(32)). Even across a billion stored blobs, eachDEKencrypts exactly one object stream under one fresh96-bit(12-byte) IV—reducing key-nonce collision probability to zero. KEKWrapping & Fast Key Rotation: The32-byteDEKis wrapped (encrypted) using a higher-level Key Encryption Key (KEK) managed in SAU Key Vault KMS (/dashboard/key-vault), and the compactwrappedDek,iv,authTag, andkekVersionheaders store alongside the ciphertext metadata. Rotating yourKEKre-wraps tiny32-bytemetadata headers in milliseconds without re-encrypting terabytes of underlying blob payloads!- 128-Bit Galois Authentication Tag (
16 bytes):AES-256-GCMis an Authenticated Encryption with Associated Data (AEAD) cipher. We bind(containerId, blobPath, contentLength)into the GCM Additional Authenticated Data (AAD) so an attacker or storage bug cannot swap the ciphertext ofinvoice-A.pdfintoinvoice-B.pdfwithout immediately failingdecipher.final()tag verification!
Below is the exact Node.js / TypeScript streaming envelope construction pattern used when sealing and verifying objects via /api/v1/blob:
// Client-side AES-256-GCM envelope encryption + AAD binding before /api/v1/blob upload
import { createCipheriv, createDecipheriv, randomBytes } from 'node:crypto';
export interface EncryptedBlobEnvelope {
ciphertext: Buffer;
ivBase64: string; // 96-bit (12-byte) NIST SP 800-38D recommended IV
authTagBase64: string; // 128-bit (16-byte) Galois integrity & authenticity tag
wrappedDekBase64: string;
kekKeyId: string;
}
export async function sealBlobWithEnvelope(
plaintext: Buffer,
blobLogicalPath: string,
wrapKeyFn: (rawDek: Buffer) => Promise<{ wrappedDek: Buffer; kekKeyId: string }>,
): Promise<EncryptedBlobEnvelope> {
const dek = randomBytes(32); // 256-bit ephemeral Data Encryption Key
const iv = randomBytes(12); // 96-bit IV
const cipher = createCipheriv('aes-256-gcm', dek, iv, { authTagLength: 16 });
// Bind canonical blob path into AAD to prevent cross-key substitution attacks
cipher.setAAD(Buffer.from(`sau-blob-v1:${blobLogicalPath}`, 'utf8'));
const ciphertext = Buffer.concat([cipher.update(plaintext), cipher.final()]);
const authTag = cipher.getAuthTag();
const { wrappedDek, kekKeyId } = await wrapKeyFn(dek);
dek.fill(0); // Zero ephemeral DEK from V8 Buffer memory immediately
return {
ciphertext,
ivBase64: iv.toString('base64'),
authTagBase64: authTag.toString('base64'),
wrappedDekBase64: wrappedDek.toString('base64'),
kekKeyId,
};
}
export function openBlobEnvelope(
ciphertext: Buffer,
blobLogicalPath: string,
unwrappedDek: Buffer,
ivBase64: string,
authTagBase64: string,
): Buffer {
const decipher = createDecipheriv(
'aes-256-gcm',
unwrappedDek,
Buffer.from(ivBase64, 'base64'),
{ authTagLength: 16 },
);
decipher.setAAD(Buffer.from(`sau-blob-v1:${blobLogicalPath}`, 'utf8'));
decipher.setAuthTag(Buffer.from(authTagBase64, 'base64'));
// Throws immediately if a single ciphertext bit, AAD path, or tag byte was altered!
return Buffer.concat([decipher.update(ciphertext), decipher.final()]);
}
Failure-Domain Engineering: LRS vs ZRS vs GRS vs RA-GRS
Choosing a storage plan is fundamentally a question of which physical failure domain your RPO (Recovery Point Objective) and RTO (Recovery Time Objective) must survive:
| Redundancy Mode | Synchronous / Async Copies | Physical Blast Radius Survived | Read Availability During Primary Region Outage | Included In SAU Plan |
|---|---|---|---|---|
LRS (Locally Redundant) |
3 synchronous copies in 1 datacenter |
Disk, storage rack, or TOR switch failure | Waits for datacenter recovery | Custom / Dev containers |
ZRS (Zone-Redundant) |
3 synchronous copies across 3 independent Availability Zones |
Entire datacenter facility power/cooling/fiber loss (RPO = 0) |
Automatic transparent DNS/routing across surviving AZs | Starter (250 GiB — \$9/mo) |
GRS (Geo-Redundant) |
3 local copies + 3 async copies in paired region hundreds of miles away (6 total) |
Regional natural disaster or multi-AZ blackout (RPO < 15m typical) |
Accessible after Microsoft-initiated or customer geo-failover | Pro (1 TiB — \$29/mo) |
RA-GRS (Read-Access Geo-Redundant) |
6 copies across paired regions + always-live -secondary read endpoint |
Full regional loss with immediate RTO ~ 0 read failover |
Immediate: app switches read base URL to secondary endpoint without waiting for regional failover | Business (5 TiB — \$119/mo) & Enterprise (25 TiB — \$399/mo) |
For disaster-recovery runbooks, RA-GRS removes the single biggest operational trap in cloud storage: waiting for a regional DNS failover decision while customers cannot download critical state. Your application health loop can retry reads immediately against the geo-secondary replica whenever primary regional latency spikes.
FinOps Lifecycle Tiering (Hot, Cool, Archive) & Transparent Plan Pricing
Not every terabyte has the same access temperature. Inside /dashboard/blob-storage and /api/v1/blob, lifecycle policies transition aging objects automatically across three storage tiers:
HotTier: Sub-10ms first-byte latency optimized for active user uploads, CDN origins, container registries, and recent7-day database backups.CoolTier (30-day minimum retention): Lower per-GiB storage footprint for30–90day operational logs, Grafana Loki chunks, staging snapshots, and quarterly exports read infrequently.ArchiveTier (180-day minimum offline cold tier): Lowest-cost offline compliance storage for multi-year financial audit trails, medical imaging archives, and historical training datasets (rehydrated back toHotorCoolwhen an audit requires retrieval).
Complete Managed Cloud Storage Plans & Overages (/cloud-storage)
| Plan Tier | Included Capacity | Included Monthly Egress | Redundancy Architecture | Monthly Price | Overages Beyond Plan |
|---|---|---|---|---|---|
| Developer Free | 5 GiB |
25 GiB/mo |
Managed Azure SSE + SAU Envelope | \$0/mo |
Upgrade or \$0.023/GiB storage · \$0.05/GiB egress |
| Starter | 250 GiB |
Generous pooled tier | ZRS (3 Availability Zones) |
\$9/mo |
\$0.023/GiB storage · \$0.05/GiB egress |
| Pro | 1 TiB |
Generous pooled tier | GRS (6 copies across 2 regions) |
\$29/mo |
\$0.023/GiB storage · \$0.05/GiB egress |
| Business | 5 TiB |
High-volume production | RA-GRS (6 copies + live secondary read) |
\$119/mo |
\$0.023/GiB storage · \$0.05/GiB egress |
| Enterprise | 25 TiB |
Enterprise multi-TB | RA-GRS (6 copies + live secondary read) |
\$399/mo |
\$0.023/GiB storage · \$0.05/GiB egress |
Regulatory WORM Immutability Locks, Legal Holds & Scoped HMAC-SHA256 SAS URLs
Ransomware operators systematically target backup API credentials first: before encrypting a production database, attackers issue DELETE calls against cloud backup buckets.
Enabling Time-Bound WORM (Write-Once-Read-Many) Immutability or a Legal Hold on a container in /dashboard/blob-storage enforces non-erasable, non-rewritable retention aligned with SEC Rule 17a-4(f), FINRA Rule 4511, and HIPAA archival mandates:
- Locked Time-Bound Retention (
1to3,650days): Once locked, blobs can be created and read, whilePUToverwrites andDELETEcalls are rejected at the storage control plane—even if issued by an organizationOWNERAPI key! Retention intervals can only be extended, never shortened. - Litigation Legal Holds: Place an indefinite immutable hold tagged with a case identifier (
legal-matter-2026-04); objects remain undeletable until every active legal hold tag and time-bound retention window clears. - Scoped HMAC-SHA256 Pre-Signed SAS URLs: Rather than routing multi-gigabyte uploads through your Node.js/Python web workers, call
POST /api/v1/blob/sasto mint a short-lived (5m–60m), permission-scoped (r,cw), source-IP-restricted HMAC-SHA256 Shared Access Signature URL:
# 1. Create a WORM-locked compliance backup container via /api/v1/blob
curl -fsSL -X POST https://serversareup.com/api/v1/blob/containers \
-H "Authorization: Bearer ${SAU_API_TOKEN}" \
-H "Content-Type: application/json" \
-d '{
"name": "pg-wal-worm-prod",
"redundancy": "RA-GRS",
"defaultAccessTier": "Cool",
"wormImmutabilityDays": 90,
"lockImmutabilityPolicy": true
}'
# 2. Mint a 15-minute write-only HMAC-SHA256 SAS URL restricted to your server IP
curl -fsSL -X POST https://serversareup.com/api/v1/blob/sas \
-H "Authorization: Bearer ${SAU_API_TOKEN}" \
-H "Content-Type: application/json" \
-d '{
"container": "pg-wal-worm-prod",
"blobPath": "2026-11-16/basebackup.tar.zst.enc",
"permissions": ["create", "write"],
"expiresInSeconds": 900,
"allowedIpCidr": "203.0.113.42/32"
}'
Unifying Managed Cloud Storage with BYO Azure Storage Accounts (*.blob.core.windows.net)
Already operate corporate Microsoft Azure subscriptions with existing *.blob.core.windows.net storage accounts? Connect your BYO Azure Storage Account right alongside your ServersAreUp Managed Cloud Storage containers inside /dashboard/blob-storage and /dashboard/storage—giving you one unified browser explorer, background rclone cloud-to-cloud transfer engine, and Linux iscsi-block / rclone-vfs mount workflow across Managed Azure, BYO Azure, S3, Cloudflare R2, DigitalOcean Spaces, and Backblaze B2.
Explore plans or launch your free 5 GiB developer container today on Managed Cloud Storage (/cloud-storage).
Frequently Asked Questions
How does double-envelope encryption protect my files differently than standard cloud SSE?
With client-side AES-256-GCM envelope encryption, every object is encrypted with a unique 256-bit Data Encryption Key (DEK), a fresh 96-bit IV, and a 128-bit GCM authentication tag before ciphertext streams over TLS 1.3 into Microsoft Azure 256-bit SSE containers—ensuring plaintext never touches third-party disks and any bit alteration fails cryptographic verification immediately.
When should I choose ZRS vs GRS vs RA-GRS for production workloads?
Choose ZRS (250 GiB \$9/mo) to survive a complete single-datacenter power or fiber outage within one metro region (RPO = 0). Choose GRS (1 TiB \$29/mo) when backups must survive a regional catastrophe across paired regions hundreds of miles apart. Choose RA-GRS (5 TiB \$119/mo or 25 TiB \$399/mo) when your application requires an always-live secondary regional endpoint for immediate read failover without waiting for provider failover declarations.
Can a compromised admin token delete backups inside a locked WORM container?
No. Once a time-bound WORM immutability policy is locked (or a Legal Hold is active) on a container in /dashboard/blob-storage, neither API tokens nor dashboard administrators can overwrite or delete protected blobs until the retention period expires—satisfying SEC 17a-4(f), FINRA, and HIPAA ransomware-resilience controls.
Is there a free tier to test /api/v1/blob and envelope-encrypted uploads?
Yes. Every account includes a free Developer tier with 5 GiB of Managed Cloud Storage and 25 GiB/mo of outbound egress at \$0/mo, with transparent overages (\$0.023/GiB storage and \$0.05/GiB egress) and instant upgrades as your data grows.
Explore With AI & Cite This Page
Open a grounded prompt in your AI assistant or copy a clean Markdown, HTML, or APA reference link.
Related Diagnostic Tools & Engineering Guides
Validate protocols and subnets directly in your browser or continue reading related architecture notes.
Free JWT Decoder, Base64URL & SHA-256 Tool
SSH Keygen Generator & Cloud-Init YAML Builder
Cloud Cost & Infrastructure ROI Calculator
Cloud FinOps & Infrastructure ROI Engineering: Eliminating Egress Taxes, Memory Fragmentation, and SaaS Sprawl
An engineering framework for cloud FinOps: modeling hyperscaler egress and NAT taxes, reclaiming fragmented memory, replacing per-seat SaaS via /cloud, and offloading blobs to /cloud-storage.