Traiecto
A resilience control plane for organisations that run critical applications across several data centres or clouds.
The problem
Existing orchestration tools act on the last state they observed. In multi-cluster environments that state is often stale: reports arrive late, sites are partitioned, backups are restored, controllers restart. Automation acting on stale state turns a fault into an outage: evicting the last healthy replicas, treating an empty endpoint list as truth, rolling out to two regions at once. Under DORA (finance, in force since January 2025) and NIS2 (critical infrastructure), such outages are regulatory events, not just incidents.
The approach
The operator declares the desired state; Traiecto continuously reconciles it with reality and treats uncertainty as a first-class part of the control loop.
- Version-bound lifecycle statesAccepted, allocated, running, ready and serving are tracked separately. Readiness counts only when tied to the exact configuration version and a fresh node session.
- Stale-report protectionNode sessions carry epochs and strictly increasing sequence numbers, so late, replayed or out-of-order reports cannot overwrite newer facts.
- Fleet-wide disruption budgetOne availability budget across all clusters, split into exact per-cluster portions. An unreachable cluster counts as unavailable: the system fails closed.
- Fail-safe service discoveryA sudden mass “not ready” signal is treated as suspect: the last good endpoint set is kept and marked degraded instead of wiping a working service.
Evidence today
A working prototype on the founder's own platform, written in C++ with gRPC and Protobuf.
- Working multi-node stand: 3 control-plane replicas and 5 node agents.
- 2-hour fault-injection rehearsals on that stand, each with about 116 injected faults: process kills, network partitions, storage pause and restore, disk exhaustion.
- Automated invariant checks; violations per run reduced from 686 to 0–1 over successive iterations.
- About 5,400 automated tests and 18 fuzz targets; CI gated by ASan, UBSan and TSan sanitizers.
Next milestones. The 6-hour soak and the multi-cluster soak are not yet green; they are the next milestones. There are no customers yet.
Roadmap
Kubernetes interoperability, then sovereign AI inference (customers' own models on their own or EU-sovereign GPU capacity) on the same core.
Live demo under NDA on request — m.grafov2004@outlook.com.
Experience
- 2026 –Founder & sole engineer · Traiecto
Designed and built a declarative multi-cluster orchestration platform and its fault-injection soak harness.
- 2024 –Chief Technology Officer · ARBUZ
Cross-border remittance app; lead technology and engineering for the product.
- 2023Lead Developer / Tech Lead · Deltatechservice LLC
Led development of algorithms for hardware-software information-security systems.
- 2022Head of IT Infrastructure / Chief Technical Officer · AKIB Pochtobank JSC
Responsible for the bank's IT infrastructure: availability, security and infrastructure operations.
- 2020Senior Engineer, Payment Processing Systems Design · Forbank JSC
Designed high-load payment transaction processing systems.
Research
Grafov M.D. Hybrid Cloud Solutions for Processing High-Load Financial Systems. Industrial Automated Control Systems and Controllers, 2025, No. 11, pp. 24–32. DOI 10.25791/asu.11.2025.1623. Sole author, peer-reviewed.
The paper models hybrid public/private cloud architectures for high-load transaction processing in a simulation at 20,000 TPS under combined DDoS attacks, assessing performance, resilience and security under variable load. It derives recommendations on Zero Trust, micro-segmentation, automated incident response and DevSecOps. Traiecto is the engineering continuation of this work.
Education
Ivanovo State University of Chemistry and Technology
- Master's programme (final year)
- BSc Information Systems and Technologies (09.03.02)