Home / Services / Database

Services / Database

Your database is the one system that can't stay down.

When the database fails, everything fails, the app, the checkout, the dashboard your customers are refreshing right now. Erzon puts a senior database engineer on it: stabilize first, recover the data, fix the root cause, and prove the backups actually restore.

Database emergencies get triaged within one business hour.

Symptoms we fix

Call us if you're seeing…

SYMPTOMS · DATABASE
  • The database is down or refusing connections, and restarts aren't bringing it back.
  • Queries that took milliseconds now take seconds or time out, and it's getting worse.
  • Replication lag growing, or a replica that silently stopped replicating days ago.
  • A migration failed halfway and now the schema is in a state nobody trusts.
  • Deadlocks or lock contention freezing writes during your busiest hours.
  • Suspected data corruption, checksum errors, a crashed table, an index returning wrong results.
  • Disk filling up faster than it should, bloat, unvacuumed tables, runaway WAL/binlogs.
  • A restore you attempted isn't working, or you've just discovered the backups are empty.

All of these are fixable. Most within hours. If any is happening in production right now, skip the rest of this page and book the fix.

The work

What we do.

Outage recovery

Get the database answering queries again, safely. We stabilize before we experiment, never against a primary holding your only copy of the data.

Corruption repair & data recovery

Crashed tables, corrupted indexes, torn pages, accidental deletes. We recover what's recoverable and tell you plainly what's gone, before anything irreversible.

Slow queries & performance

Query plans, missing indexes, N+1 patterns from the app, table bloat, misconfigured memory. We make the slow parts fast and leave you the analysis.

Migrations & schema changes

Un-stick failed migrations and plan zero-downtime schema changes on large tables, off the "afraid to touch the schema" treadmill.

Replication & high availability

Broken replicas, failover that didn't fail over, split-brain cleanup, and replication you can actually trust during the next incident.

Backups & restore verification

The backup that matters is the one that restores. We fix backup pipelines and run real restore drills so "we have backups" becomes "we've tested them."

Stacks: PostgreSQL, MySQL/MariaDB, SQL Server, MongoDB, Redis, and the managed flavors (RDS, Aurora, Cloud SQL, Azure Database, Atlas). If yours isn't listed, ask, the failure modes rhyme.

How you engage

Two ways to engage.

Production is down

Emergency fix

A senior engineer on your incident fast, triage first, stabilize, fix, then a written root-cause report. Flat scope agreed before any work starts.

Response: within 1 business hour

Get emergency help
Keep it from breaking

Reliability retainer

A monthly engagement: we monitor, patch, harden, and handle incidents with a guaranteed SLA, from an engineer who already knows your stack.

Monitoring · reviews · on-call

See retainer plans
Every engagement

How a fix works.

01

Triage

02

Fix

03

Root-cause report

04

Prevent

See the full process →

Questions

Mini-FAQ.

Can you work on a managed database like RDS or Cloud SQL?

Yes, most of our database work is on managed services. Managed hosting removes the hardware, not the failure modes: bad queries, bloat, replication lag, failed migrations, and untested backups are all still yours. And ours.

Will you need production access?

For emergencies, read access plus a screen share is usually enough to diagnose; we agree on write access explicitly before changing anything. Everything we run is logged and included in the report. NDA on request, before you share anything.

The data might be partially lost. Should we keep trying things?

Stop. Every recovery attempt on a corrupted system risks overwriting what's still recoverable. Take a snapshot if you safely can, stop the experiments, and get a senior engineer on it. This is the single most expensive hour to DIY.

Down, slow, or corrupted, we've recovered from this one before.