Fadhel Hasbullah
← Engineering experience

DBO
Order Service

Senior Backend Engineer · Section Head · 2024–now

DBO is a B2B construction-materials marketplace. I moved its order reads to MySQL, tuned latency and memory, and decommissioned OpenSearch.

−55.3%Order-list median latency · controlled A/B
−38%Peak memory (RSS) · bounded 5-user load test
One datastoreOpenSearch decommissioned

Simplified read path

REST + gRPC clients
Go order serviceIndexed IDs · batched reads
MySQLOrder data
Other servicesMetadata
OpenSearchDecommissioned

Order-list median latency

Before · 1,080 ms After · 482.5 ms
Controlled A/B: 8 interleaved rounds, 80 calls per version, median of round medians. Not a production-wide latency claim.

Performance tuning

Indexed MySQL reads, batched hydration and bounded downstream calls. Checked latency, memory and cancellation behaviour across REST and gRPC.

OpenSearch decommissioned

Migrated read paths to MySQL first, then removed search projections and client dependencies, then decommissioned the OpenSearch infrastructure. Rollback stayed possible at each stage.

Data integrity

Validated order-closing jobs across transactional records, history and downstream processing, including excluded-account handling.

Measurement scope

The target endpoint improved in all 8 rounds. Eleven other endpoints showed no systematic regression; occasional latency outliers remained. Peak memory comes from a bounded five-user test and includes profiling overhead. These numbers describe the experiments, not the whole production system.

Stack: Go, MySQL, Kafka, gRPC, REST, Kubernetes.