Proposal: Add Volcengine self-hosting support

Proposal: Add Volcengine (火山引擎) self-hosting support

Summary

Hi E2B team,

Thank you for building and open-sourcing E2B. The project is extremely valuable for AI code execution and sandboxed environments. We have seen strong interest from Volcengine (ByteDance Cloud / 火山引擎) users in China and Asia-Pacific who would like to run E2B in their own cloud environment.

We are currently evaluating a self-hosted E2B deployment on Volcengine, and we would like to contribute Volcengine self-hosting support back upstream if this aligns with your roadmap.

Similar to the Alibaba Cloud proposal (#3035), our approach is to keep the upstream architecture and core runtime behavior unchanged, and contribute the Volcengine-specific work as an isolated provider/deployment layer.

We would not change:

• The public E2B API contract
• Firecracker/orchestrator/template build core logic
• Scheduler, NBD, networking, or other high-blast-radius runtime internals

Background: Why Volcengine

Volcengine (火山引擎) is ByteDance's enterprise cloud platform, widely adopted in China and Asia-Pacific, particularly strong in AI/ML workloads. It provides:

• ECS (Elastic Compute Service) — bare-metal and virtualization instances with KVM/nested virtualization support (required for Firecracker)
• TOS (Torch Object Storage) — S3-compatible object storage with multi-region support
• CR (Container Registry) — Docker image registry with enterprise features
• VPC / CLB / ALB — networking and load balancing
• Redis & RDS (PostgreSQL) — managed database services
• IAM — identity and access management with role-based credentials

Key advantage: Volcengine TOS is fully S3-compatible (endpoint: tos-s3-{region}.volces.com), which means the existing AWS S3 storage provider (storage_aws.go) can potentially be reused with custom endpoint configuration, significantly reducing implementation effort.

Proposed scope

If this direction makes sense, we can contribute the work incrementally:

1. Volcengine deployment support (Terraform)

• Terraform provider volcengine/volcengine based infrastructure templates
• ECS instance pools for control/client/orchestrator nodes (with KVM support)
• CLB/ALB exposure for API and client-proxy
• VPC, security group, and subnet configuration
• IAM Role configuration for service authentication

2. Volcengine TOS storage provider

• Object storage implementation for template/snapshot/rootfs artifacts
• Leverage S3-compatible API — potentially reuse storage_aws.go with custom endpoint configuration
  • External endpoint: tos-s3-{region}.volces.com
  • Internal endpoint: tos-s3-{region}.ivolces.com
• Range reads and large-object multipart upload support
• IAM Role / STS token based credentials
• Alternative: dedicated TOS SDK (Go SDK available) implementation if S3 compatibility is insufficient

3. Volcengine CR (Container Registry) provider

• Store template build images in Volcengine CR
• Pull/push support using IAM Role / CR temporary token flow
• Repository creation or documented prerequisite

4. Nomad cluster on Volcengine

• Packer-based disk image for Volcengine ECS
• Nomad server/client cluster provisioning
• Consul service discovery integration
• Auto-scaling group configuration

5. Self-hosting documentation

• Volcengine prerequisites (account, quotas, KVM-enabled instance types)
• Required IAM permissions
• Network architecture diagram
• Step-by-step deployment guide
• Smoke test and validation steps

Proposed contribution approach

We do not want to submit a large, hard-to-review PR.

If the maintainers are open to this direction, we would follow a phased approach:

1. Design/docs PR — Architecture overview and Volcengine-specific design decisions
2. TOS storage provider — With unit tests (may extend existing AWS provider with configurable endpoint, or implement standalone)
3. CR artifacts registry provider — With tests
4. Terraform deployment skeleton — iac/provider-volcengine/ directory structure
5. Nomad cluster disk image — Packer template for Volcengine ECS
6. Self-hosting documentation — Complete deployment guide and smoke test

Proposed directory structure

iac/
provider-volcengine/
main.tf
variables.tf
Makefile
init/
modules/
nomad/
nomad-cluster/
nomad-cluster-disk-image/

packages/shared/pkg/storage/
storage_volcengine.go # TOS storage provider (or extend AWS with endpoint config)
storage_volcengine_test.go

Validation plan

We plan to validate the provider with:

• API health check
• Sandbox create / connect / delete
• Pause / resume
• Template build
• Template build from image
• Container registry pull path (Volcengine CR)
• Basic concurrent sandbox creation
• Storage read/write/range-read correctness
• End-to-end latency benchmarks

Technical notes

• Firecracker compatibility: Volcengine ECS supports bare-metal instances and nested virtualization (KVM), which is required for Firecracker microVMs.
• S3 compatibility: Volcengine TOS supports the S3-compatible API, which may allow reusing the existing storage_aws.go with a custom endpoint. Available regions include cn-beijing, cn-shanghai, cn-guangzhou, ap-southeast-1, etc. (full list). We will evaluate whether this is sufficient or if a dedicated TOS SDK implementation is needed.
• Terraform provider: The official volcengine/volcengine Terraform provider is available on the Terraform Registry with comprehensive resource coverage.
• Go SDK: Official TOS Go SDK is available if native implementation is preferred over S3 compatibility layer.

Question

Would Volcengine (火山引擎) self-hosting support be something you would be open to accepting upstream, if we keep it isolated to provider, storage, registry, deployment, and documentation layers?

If yes, we can prepare a small initial design/docs PR for review before starting the implementation.

---

References:

• Volcengine official site: https://www.volcengine.com/
• Volcengine Terraform provider: https://registry.terraform.io/providers/volcengine/volcengine/latest
• Volcengine TOS S3 compatibility: https://www.volcengine.com/docs/6349/75069
• Volcengine TOS regions & endpoints: https://www.volcengine.com/docs/6349/107356
• Volcengine TOS Go SDK: https://www.volcengine.com/docs/6349/93450
• Related: #3035 (Alibaba Cloud proposal)

Start by reviewing the proposed provider layout under iac/provider-volcengine/, the existing packages/shared/pkg/storage/storage_aws.go, and related proposal #3035. Done would be a maintainer-approved design/docs PR that defines the isolated Volcengine deployment, storage, registry, and documentation scope before implementation begins.