T1 Established shops, predictable load.
Bare-metal managed
- nodes
- 1 dedicated server
- scaling
- Vertical scaling
- database
- Local MariaDB
- cache
- Varnish, Redis, OPcache
- failover
- Restore from backup
- rpo
- 24h (daily backup)
- rto
- ~90 min
Managed e-commerce hosting, since 2009
Hosting built for
Magento, PrestaShop, WooCommerce, Sylius.
Tuned LAMP stack, multi-DC architecture, automatic scaling, HA on demand. Managed end-to-end by senior engineers on hardened Debian, in European datacenters.
peak.requests
10,000+
req/s per cluster
sla.uptime
99.9 %
monthly, contractual
restore.points
9+
per site, 2 EU countries
monitoring.probes
100+
custom, 24/7
support.window
24/7
on-call, senior engineers
01, ARCHITECTURE TIERS
Four hosting tiers.
Each tier maps a real e-commerce load profile. The HA module grafts on T3 and T4 to push RPO to zero on isolated node failure.
T2 Growing shops, seasonal peaks.
Cloud, vertical scaling
- nodes
- 1 to 2 webservers, 1 dedicated DB
- scaling
- Vertical auto-scale (CPU, RAM)
- database
- External MariaDB
- cache
- Varnish, Redis, OpenSearch
- failover
- Restore from backup
- rpo
- 24h
- rto
- ~60 min
T3 Established mid-market, 24/7 ops.
Cloud, advanced vertical
- nodes
- 2 to 4 webservers, master/slave DB, 2 OpenSearch
- scaling
- Vertical auto-scale, manual horizontal
- database
- MariaDB master/slave
- cache
- Varnish, Redis, OpenSearch x2
- failover
- HA module available
- rpo
- Seconds (DB replication)
- rto
- 30 to 45 min
T4 High-traffic, drops, viral peaks.
Horizontal auto-scaling
- nodes
- 1 to 30 webservers, up to 10,000 req/s
- scaling
- Horizontal auto-scale
- database
- Galera or master/slave
- cache
- Varnish, Redis Sentinel, OpenSearch cluster
- failover
- HA module recommended
- rpo
- 0 (with HA)
- rto
- Automatic, seconds
Indicative pricing on /en/prices. Every infrastructure is sized after an audit, no off-the-shelf product.
02, HIGH-AVAILABILITY MODULE
Synchronous, multi-DC, automatic.
3 nodes minimum, spread across 2 datacenters minimum, with rack anti-affinity. Designed to keep the service running through isolated node loss without human intervention.
topology, ha module
Solid coral line: intra-DC synchronous replication. Dashed coral line: inter-DC synchronous replication. Quorum maintained as long as 2 of 3 nodes stay reachable.
- 01
Galera multi-master
Synchronous MariaDB replication. Every write is committed across all nodes before validation.
- 02
Redis Sentinel
Shared sessions, automatic failover. Cache and session continuity through node loss.
- 03
Load balancer with health checks
Automatic traffic rerouting on node failure. No human intervention for single-node loss.
- 04
Rack anti-affinity
3-node cluster minimum, spread across 2+ datacenters. Rack-aware placement to survive infrastructure-level events.
- 05
File sync
Real-time replication between nodes via lsyncd. Application files stay coherent across the cluster.
- 06
RPO close to zero
Synchronous DB replication, in-memory session replication. Designed for zero data loss on isolated node failure.
03, STACK
A tuned stack, layer by layer.
Every layer is provisioned via versioned Ansible playbooks, with project-level Unix isolation and a dedicated PHP-FPM pool per site.
| Layer | Components |
|---|---|
| Edge | Cloudflare WAF, DDoS mitigation, bot management, rate limiting |
| Load balancer | HAProxy or Nginx, health checks, sticky sessions when needed |
| Web | Nginx or Apache, PHP-FPM tuned per project, isolated pools, dedicated Unix user |
| Cache | Varnish full-page, Redis Sentinel for sessions and object cache, OPcache |
| Search | OpenSearch or Elasticsearch, dedicated nodes on T3 and T4 |
| Database | MariaDB, master/slave on T3, Galera multi-master on HA module |
| OS | Hardened Debian 12+, kernel-level isolation via LXC, project-level Unix isolation |
| Provisioning | Ansible playbooks versioned, internal inventory, lsyncd file replication |
| Monitoring | 100+ custom probes, Monit, Munin, log analysis, 24/7 alerting |
04, OBSERVABILITY AND SCALING
100+ probes. Auto-scale in seconds.
Monitoring catches anomalies before they hit the application. LXC isolation lets us add CPU and RAM without restart.
MONITORING
100+ custom probes, 24/7.
- +Monit and Munin for system metrics, services and thresholds.
- +Custom probes per CMS: cart funnel, checkout latency, payment gateway round-trip.
- +Log analysis: Nginx, PHP-FPM, MariaDB slow queries, OpenSearch errors.
- +Alert escalation: in-hours engineer first, after-hours SMS to on-call rotation.
- +Client console (console.fast-mage.com): service status, backup history.
AUTO-SCALING
Vertical in seconds. Horizontal on demand.
- +Vertical (T1 to T3): CPU and RAM allocated to the LXC container without reboot.
- +Horizontal (T4): 1 to 30 webservers behind the load balancer, sustained 10,000 req/s.
- +Trigger on real-time load: CPU pressure, request queue depth, PHP-FPM worker saturation.
- +File replication via lsyncd, sessions on Redis Sentinel, no sticky session needed.
- +Capacity planning reviewed before known peaks (Black Friday, drops, campaigns).
Console views, live probes
05, DISASTER RECOVERY
Three resilience profiles.
Targets validated during the commissioning phase, measured from incident detection. Detailed DR plan available to prospects on request.
| Scenario | Standard, RPO / RTO | Resilience, RPO / RTO | HA module, RPO / RTO |
|---|---|---|---|
| Application service crash (PHP-FPM, Redis, MariaDB, OpenSearch, Varnish) | 0 RPO / minutes | 0 RPO / seconds (LB) | 0 RPO / automatic |
| VM unreachable, hardware failure, filesystem corruption | 24h RPO / ~90 min | Seconds RPO / automatic (LB), 30-45 min (master/slave) | 0 RPO / automatic |
| Datacenter loss, prolonged provider outage | 24h RPO / 1-2h | Variable, automatic if multi-DC | 0 RPO / automatic (minority DC), 15-30 min (majority DC) |
| Database corruption | 24h RPO / 1-4h | Seconds RPO / 30-60 min (slave restore) | 0 RPO / automatic (Galera resync) |
| Configuration error (Cloudflare, system) | 0 RPO / under 1h | 0 RPO / under 1h | 0 RPO / under 1h |
BACKUP RETENTION
9+ restore points by default: 7 daily, plus 2 bi-monthly. Extended retention (30, 60, 90 days) on demand.
BACKUP LOCATION
Stored on 2+ separate servers, in 2 different European countries, with providers distinct from production.
RESTORE TESTING
Annual joint restore drill included on T3 and HA tiers. Run with you, on your infrastructure.
CONSOLE.FAST-MAGE.COM
Every backup is tracked: timestamp, scope (files, database, full), size, storage target. Restore opens as a ticket from this view, no tier 1 filter.
History retained for the full contract duration. Client-side audit available at any time, no prior request needed.
06, CONTRACTUAL SLA
Written, measurable, enforceable.
Full SLA defined in the Cloud Hosting Particular Conditions, article 5. Penalties on miss, article 5.2.2.
| Engagement | SLA |
|---|---|
| Monthly uptime | 99.9 % |
| Network restore | 2h |
| Hardware restore | 2h |
| System restore | 3h |
| Functional anomaly restore | 12h |
| High-priority intervention, business hours | 30 min |
| High-priority intervention, after hours | 6h |
REAL-WORLD
Field response, not just the contract.
On real incidents, in-hours pickup is typically under 10 min, well below the 30 min contractual SLA. After hours, automated SMS escalation reaches an on-call engineer with end-to-end pickup under 20 to 30 min, then resolution under the 6h contractual window.
Every high-priority incident gets a written post-mortem on request: timeline, root cause, corrective actions.
07, SECURITY AND SOVEREIGNTY
European, hardened, PCI DSS.
Production stays in the EU, on infrastructure providers we audit for performance and isolation. PCI DSS available on dedicated offers.
DEFENSE IN DEPTH
- +Cloudflare WAF, DDoS mitigation, bot management, rate limiting.
- +Hardened Debian 12+, security patches applied on day 0.
- +LXC isolation per project. Dedicated Unix user, isolated PHP-FPM pool.
- +Custom firewalls, restricted access controls, audit logs.
- +Cloudflare change protocol: peer review, impact assessment, rollback ready.
DATA SOVEREIGNTY
- +European datacenters only, no US providers in production path.
- +GDPR compliant by design.
- +PCI DSS on dedicated infrastructure offers.
- +Backups stored in 2 different European countries.
- +Optional private AI inference on your own GPUs (see /en/ai).
INFRASTRUCTURE PROVIDERS
Production runs on a mix of:
- >OVH (FR, DE, PL)
- >Scaleway (FR, NL)
- >Hetzner (DE, FI)
- >UpCloud (FI, NL, DE)
Provider selection per project: latency, location, capacity, regulatory constraints. GCP and AWS available when client requirements demand it.
08, SUPPORT MODEL
Senior engineers. No tier 1.
Direct contact with the engineers who run your infrastructure. No qualification script, no bot, no disposable first-level filter.
01
System level
Stack maintenance, kernel and OS patching, MariaDB tuning, Varnish VCL, Redis, OpenSearch indexing, PHP-FPM pool sizing. Provisioning via Ansible.
02
Application level
Performance bottleneck triage, plugin conflict diagnosis, CMS hack containment, indexer freeze, queue stalls. Direct collaboration with your dev team or agency.
03
Infrastructure ops
Provider escalation, hardware replacement coordination, DNS cutover, capacity planning, backup restore drills. PowerSupport option covers client-owned infrastructure.
60 % of new clients come via referrals from agencies and IT directors. The DR plan (PRA) is shared with prospects under NDA, contact us for the full version.
06, FREQUENTLY ASKED
Frequently asked questions.
Five key points.
Answers reviewed by our engineers. If your question is missing, just write to us, quick reply, no sales script.
Q01 Do you guarantee 99.9 % uptime in writing?
+
Do you guarantee 99.9 % uptime in writing?
Yes. 99.9 % monthly uptime is the contractual SLA written in our Cloud Hosting Particular Conditions, article 5. Penalties apply if the threshold is missed. Higher targets are achievable on the HA module by design (synchronous Galera, automatic failover), validated during the one-month commissioning phase.
Q02 How many restore points do you keep, and where?
+
How many restore points do you keep, and where?
Minimum 9 restore points available at any time: 7 daily rolling backups, plus 2 bi-monthly snapshots (1st and 16th of each month, kept until the next cycle). Backups live on at least 2 separate servers, in 2 different European countries, with infrastructure providers distinct from the production server. Extended retention (30, 60, 90 days) and Tier-3 backup to your own storage are available.
Q03 What does the HA module actually guarantee?
+
What does the HA module actually guarantee?
Designed RPO of zero on isolated single-node failure: synchronous Galera writes are committed across the cluster, Redis Sentinel handles session failover, the load balancer reroutes traffic automatically. On loss of a majority DC (2 nodes out of 3), the remaining node loses quorum and refuses writes by default (split-brain protection); a manual bootstrap reinstates service in degraded mode within 15 to 30 min in business hours. Targets are validated during the one-month commissioning acceptance test before production.
Q04 Where is my data stored and who can access it?
+
Where is my data stored and who can access it?
European datacenters only: OVH, Scaleway, Hetzner, UpCloud. No US providers in the production path. Each project runs under a dedicated Unix user with isolated PHP-FPM pool. Configuration and credentials are centralized in our internal inventory; access is restricted to authenticated Fast-Mage engineers and logged. GDPR-compliant. PCI DSS available on dedicated offers.
Q05 Can you take over an existing infrastructure I own?
+
Can you take over an existing infrastructure I own?
Yes. The managed-services-only model is supported. RTO targets then apply from the moment you provision an operational server on your account. The PowerSupport option lets Fast-Mage interact directly with your provider's console and ticket system, otherwise hardware escalation stays on your side. Audit, takeover playbook, and a one-month observation period are included.
Q06 What is included in 24/7 on-call?
+
What is included in 24/7 on-call?
Automated monitoring (100+ custom probes) detects incidents around the clock. In business hours, the technical team on duty acts immediately. After hours, the alerting system escalates by SMS to the first-level on-call engineer, with automatic escalation if no acknowledgement. End-to-end pickup time after hours can reach 20 to 30 minutes. Effective intervention then starts under the contractual 6h after-hours SLA, in practice much faster on real incidents.
Question not answered?
Contact us →CONTACT, FREE AUDIT
Let's talk about your project.
Free audit, reply within 24 business hours.
Tell us about your stack and situation, a Fast-Mage engineer will call you back. No salesperson, no script. The audit stays yours, even if you don't migrate.