Luthfi Farabi Curriculum vitae, October 2026 Download CV

Luthfi Farabi

Senior Database Engineer & DBA

Open to full-time roles (remote or hybrid) · also select contract work

10+ years in production databases, most recently running MySQL, PostgreSQL and MongoDB fleets on Alibaba Cloud and AWS. I tune queries, cut cloud bills, run zero-downtime migrations, build near-real-time CDC pipelines into the data warehouse, and automate routine DBA work with Claude + MCP.

Currently
Senior DBA, MekariPT Mid Solusi Nusantara · since March 2025
Previously
Data Engineering & Architect LeadKoltiva · 2024–2025Database Engineer LeadInvestree · 2021–2024
Based in
Jakarta, IndonesiaGMT+7 · works async across time zones

Key figures

10+years1
of production database and data work, since 2013
1TB2
MySQL table, 500M+ rows, changed online with zero downtime
30–55%3
lower database cost, at three employers
99.9%4
uptime held in regulated fintech lending
  1. July 2013 to present, from Sybase and DB2 on Linux and AIX to cloud fleets. Experience
  2. Lock-free schema change via Alibaba Cloud DMS: shadow-table copy, binlog catch-up, atomic rename; reads and writes continued throughout. Case 01
  3. ~30% Alibaba Cloud RDS at Mekari, since 2025; 40% AWS RDS at Koltiva, 2024–25; 40–55% at REA Group, 2018–19.
  4. As Database Engineer Lead at Investree, 2021–24.
Case 01

Migration

1TB

Migration · 500M+ rows · zero downtime

Zero-downtime schema change on a 1 TB MySQL table

Context
A cross-service integration needed new columns on a 1 TB, 500M+ row MySQL table. At that size, a table-rebuilding ALTER TABLE means hours of heavy I/O, replica lag and a metadata-lock risk at the end.
Action
Used Alibaba Cloud DMS lock-free schema change: a shadow table with the new columns, filled in batches and kept in sync from the binlog, then an atomic RENAME cutover that needs only a brief metadata lock.
Result
Zero downtime and no maintenance window. Reads and writes continued throughout the copy, and the integration shipped on schedule.
Stack
MySQL · Alibaba Cloud DMS · binlog-based online schema change
Read the full write-up

A native ALTER TABLE that rebuilds a 1 TB, 500M+ row MySQL table runs for hours: heavy I/O, replicas that only start the change after the primary finishes, and an exclusive metadata lock at the end that a single long transaction can stall, blocking every query behind it. Instead I used Alibaba Cloud DMS lock-free schema change. DMS creates a shadow table with the new schema, copies the existing rows in batches, replays ongoing writes from the binlog, then performs an atomic RENAME cutover. Apart from short shared locks on each copy batch, the only blocking step is a brief metadata lock for the swap.

Reads and writes continued normally through the copy and catch-up. Zero downtime, no maintenance window, and the service integration shipped on schedule.

Case 02

Cost

40–55%

Cost · lower database bill

A MySQL database bill cut by 40–55%, by fixing the queries first

Context
A multi-node MySQL setup on large instances was struggling: 800K+ slow queries in about a day, CPU at 94–98%, and over 99% of open connections sitting idle.
Action
Redesigned the indexes to remove a filesort from the worst query, trimmed the idle connections, then right-sized the instances.
Result
40–55% off the database bill, with headroom to spare.
Stack
MySQL · composite indexes · connection management · right-sizing
Read the full write-up

The instinct on a CPU-pinned database is to scale up. I scaled down, after finding why it was hot. 800K+ slow queries in about a day and 94–98% CPU were not a capacity problem; they were a query problem. The worst offender ran a filesort on every execution; a composite index that returns rows already in sort order removed the sort entirely.

On top of that, over 99% of connections were idle: bloat, not load. Idle sessions cost per-thread memory and max_connections headroom rather than CPU; the CPU was the queries. With the queries fixed and the idle connections trimmed, the original instances were far more than the workload needed. Right-sizing them took 40–55% off the bill.

Case 03

Incident

95%

Incident · memory back to stable within 2 hours

A production PostgreSQL pinned at 95% memory, stabilized

Context
A production PostgreSQL was pinned at 95% memory and falling over.
Action
Traced it past the obvious to ~500 idle app connections with no pooler, lock contention and 23-second sequential scans. Shipped PgBouncer transaction pooling, idle timeouts, work_mem tuning and an index plan.
Result
Memory stable again within 2 hours, with a 5-Whys RCA so the root cause, not just the symptom, is on record.
Stack
PostgreSQL · PgBouncer · index design · 5-Whys RCA
Read the full write-up

The page said "high memory"; the cause was three things stacked. No connection pooler in front of the app meant ~500 idle connections, each a separate backend process with its own memory, and each able to claim work_mem for every sort or hash once it got busy. Long-running transactions created lock contention. And a few hot queries were doing 23-second sequential scans on tables missing the right index.

The fix went in as layers: PgBouncer in transaction-pooling mode to multiplex the app's connections onto a small server pool, idle timeouts to reap the stragglers, work_mem dialed back because it is a per-sort, per-hash limit that multiplies across concurrent queries, and a targeted index plan to replace the sequential scans with index scans, so the heavy sorts shrank or disappeared instead of needing more memory. Then I wrote the 5-Whys RCA in Confluence so the root cause is on record.

Case 04

Incident

100GB+

Incident · table recovered inside the maintenance window

A database failover loop, broken inside the maintenance window

Context
After a lock-free schema change on a 100 GB+ MySQL table, an engine bug in an index on a virtual generated column sent the instance into a failover loop.
Action
Dropped the offending index to stop the loop, built a clean copy of the table with DTS, and swapped it in with an atomic rename. The same week, tracked down and indexed a MongoDB COLLSCAN on ApsaraDB.
Result
Loop broken, table healthy, and the whole recovery finished inside the agreed maintenance window.
Stack
MySQL · Alibaba Cloud DTS · MongoDB on ApsaraDB
Read the full write-up

A lock-free schema change on a 100 GB+ table tripped an engine bug in an index on a virtual generated column, and the instance started looping through failovers. The safe path: drop the offending index to stop the bleeding, build a clean copy of the table with DTS, then swap it in with a single atomic rename (RENAME TABLE), all inside the agreed maintenance window.

The same week, on another engine: a MongoDB query on ApsaraDB running as a full collection scan (COLLSCAN in its plan), tracked down and indexed. Different database, same instinct: find the missing index, stop the scan.

Case 05

Observability

15+gaps

Observability · alerting gaps found across a ~20-host fleet

A ~20-host fleet's monitoring, audited panel by panel

Context
A ~20-host fleet with a sprawling dashboard of 30+ panels that nobody had fully audited.
Action
Reviewed every panel and wrote down what each one was actually showing. Found a host running 10K+ QPS with 20+ slow queries per second, another near connection saturation, and 15+ signals that were graphed but never alerted on.
Result
An alerting specification covering every gap, each with its alert, threshold and owner, designed to surface problems before they become incidents.
Stack
Monitoring dashboards · alert design
Read the full write-up

You can't alert on what you've never looked at. I read every one of the 30+ panels across the ~20-host fleet and wrote down what each one was telling me. That surfaced a host quietly running 10K+ QPS with 20+ slow queries per second, another near connection saturation, and 15+ alerting gaps: signals we were graphing but not alerting on.

The output was not a complaint but a specification: for each gap, the alert, its threshold and its owner, designed to surface the problem before it turns into an incident.

Case 06

Automation

3bots

Automation · shipped for RCA, slow-query triage and standups

Incident RCA and triage bots built on Claude + MCP

Context
RCAs, slow-query triage and daily standups were eating hours of repeatable work.
Action
Shipped an auto-RCA agent (Grafana and Confluence via MCP, 5-Whys), a Bash slow-query monitor that sends normalized query fingerprints, with literal values stripped, to Claude and posts the analysis to Google Chat, and daily Jira standups generated via MCP.
Result
The first pass on each routine job is now automatic, and the draft RCA is written before the adrenaline wears off.
Stack
Claude · MCP · Bash · Grafana · Confluence · Jira · Google Chat
Read the full write-up

Auto-RCA agent: pulls the signals from Grafana, walks a 5-Whys analysis, and writes the draft RCA straight into Confluence via MCP, so the post-incident document exists before the adrenaline wears off.

Slow-query monitor: a Bash watcher that catches slow queries, normalizes them into fingerprints so no literal values (and no customer data) leave the database, hands them to Claude for an explanation and a suggested fix, and posts the analysis to Google Chat. Daily standups: generated from Jira via MCP, so the status update writes itself.

Experience

Eight roles since 2013, from Sybase and DB2 on Linux and AIX to cloud database fleets and team leadership.

  1. – Present

    Senior DBA

    PT Mid Solusi Nusantara (Mekari)

    Administer and scale a multi-tenant database fleet on Alibaba Cloud (RDS, DAS, DTS, DMS) with no data-loss incidents. Lead query tuning and cost optimization (~30% lower RDS spend), zero-downtime migrations, and Claude + MCP automation that cut time on routine DBA tasks by an estimated 40%.

  2. –

    Data Engineering & Architect Lead

    PT Koltiva · AgriTech, global operations

    Led a cross-timezone team of data engineers and DBAs. Built a near-real-time streaming pipeline (OLTP → data warehouse), rolled out observability with Datadog, Grafana and AWS CloudWatch that cut MTTR by ~35%, and drove a 40% AWS RDS cost reduction.

  3. –

    Database Engineer Lead

    PT Investree Radhika Jaya · fintech lending

    Owned the data operations strategy and led a team of engineers. Built CI/CD-integrated schema-migration pipelines and designed the backup, PITR and DR plans; uptime held at 99.9% in a regulated environment.

  4. –

    Senior DBA

    PT Investree Radhika Jaya · fintech lending

    Automated data masking and server alerting with Bash and cron. Handled PostgreSQL performance tuning and incident response: connection pooling, memory configuration and index design.

  5. –

    System Engineer

    REA Group Asia (rumah123.com) · property portal

    Managed AWS infrastructure (EC2, RDS, S3, ELB) with KMS encryption, containerized apps with Docker, and set up alerting with PagerDuty and New Relic. Index and query redesign cut the cloud database bill by 40–55%.

Earlier roles

  1. –

    Staff Database Administrator

    PT Web Marketing Indonesia (rumah123.com)

    MySQL, Cassandra and Solr for a high-traffic property portal: access control, query tuning and daily data provisioning for engineering teams.

  2. –

    Technical Quality Assurance

    PT Adidata · project for a major Indonesian bank

    Planned and ran testing of product deliverables against enterprise banking compliance standards.

  3. –

    Staff Database Administrator

    PT Collega Inti Pratama

    Linux and AIX servers with Sybase and DB2: daily health checks, backup verification, and replication and job-scheduling troubleshooting.

Education

  1. –

    Bachelor of Computer Science

    Budi Luhur University, Jakarta

Skills

Tools that have been in production with me, grouped by the job they do.

DatabasesEngines run in production
MySQL · PostgreSQL · MongoDB · MariaDBEarlier: Oracle · Sybase · DB2 · Cassandra · Solr (search)
PerformanceMaking slow queries fast
EXPLAIN / ANALYZE · index design · online DDL (INSTANT / INPLACE) · PgBouncer · work_mem and buffer tuning · slow-query log analysis
ReliabilityKeeping data safe and online
Streaming and logical replication · HA / DR and failover · point-in-time recovery · pg_basebackup / barman · xtrabackup / mysqldump · zero-downtime migration
Cloud and infrastructureWhere the fleets run
Alibaba Cloud (RDS / ApsaraDB, DAS, DTS, DMS) · AWS (EC2, RDS, S3, CloudWatch) · Terraform · Docker · GNU/LinuxEarlier: AIX · Windows · networking
Data engineeringMoving data into analytics
ETL / ELT pipelines · CDC with Debezium · near-real-time streaming (OLTP → data warehouse) · data warehouse design
ObservabilitySeeing problems early
Grafana · Prometheus · Mimir · Datadog · New Relic · PagerDuty
AutomationRemoving routine work
Bash · Python · CI/CD pipelines · Claude + MCP · Jira and Confluence automation
PracticeHow incidents get closed
5-Whys RCA · runbooks

How I work

I'm the person who gets paged when the database is on fire, and the person who, a week later, has built something so it never catches fire the same way twice.

Day to day that means execution-plan and index work, online schema changes on tables up to a terabyte, connection-pool sanity, cost optimization, and HA/DR that actually fails over when it has to. On the data side I build near-real-time CDC pipelines with Debezium, streaming from OLTP systems into the data warehouse for analytics. And I wire Claude + MCP into the routine work (RCA drafts, slow-query analysis, reporting) so the repeatable parts take far less of my time.

Hands-on. Ships in production. Remote-ready across time zones. Writes the runbook after.

Now

What I'm focused on, updated . A now page (opens in a new tab).

  • Connection management at scale

    Going deeper on PgBouncer transaction pooling and its trade-offs, and on MySQL connection tuning (wait_timeout, stale-connection diagnosis).

  • AI-assisted DBA automation

    Prototyping where an LLM can help run databases: anomaly detection, index recommendations tested against real traffic before rollout, automated restore drills and schema-change checks in CI. First up: anomaly detection and a better slow-query analyzer.

  • Making the bots reliable unattended

    Hardening my RCA-drafting and daily-standup automation until a team can trust it to run on its own.

  • Refining my Claude + MCP skill library

    The toolkit my DBA workflows are built on, always being sharpened.

Hiring for a senior database role? Let's talk.

I'm open to full-time roles, remote or hybrid, and to select contract work. The quickest way to reach me is LinkedIn; I usually reply within 24 hours.

Senior Database Engineer & DBA with 10+ years in production databases, most recently MySQL, PostgreSQL and MongoDB on Alibaba Cloud and AWS: query tuning, 30–55% database cost reductions, zero-downtime migrations, CDC streaming pipelines and AI-assisted automation. Open to full-time roles (remote or hybrid) and select contract work.
One-liner for your shortlist or intro email
Full-time
Senior DBA, database engineering or data engineering roles. Remote, or hybrid from Jakarta.
Contract
Hourly, fixed-scope project, or ongoing retainer. Good fits: performance and cost audits, zero-downtime migrations, HA/DR and backups, CDC to data-warehouse pipelines, monitoring and alerting setup.
Hours
Jakarta, GMT+7, async-first. About four hours of daily overlap with Europe, partial overlap with the US.
Response
Usually within 24 hours, in English.