Single service comparison

Both services are FaaS with per-request, per-duration billing, but their event-source and integration surfaces differ materially. Lambda triggers natively on dozens of AWS event sources (S3, DynamoDB Streams, Kinesis, SQS, SNS, EventBridge, API Gateway) using AWS resource policies and IAM. FunctionGraph triggers on Huawei-native sources (OBS, DIS, SMN, APIG, LTS, Timer) with Huawei IAM and agency-based delegation. Existing AWS event wiring does not translate directly; each trigger binding, event payload schema, and permission model must be re-mapped against FunctionGraph's supported trigger catalogue and documented event contracts.

Runtime and packaging boundaries differ in practice. Lambda offers its own managed runtimes, container-image packaging, Lambda Layers, and Extensions API. FunctionGraph documents Node.js, Python, Java, Go, C#, PHP, Cangjie, and custom runtimes, with inline/ZIP/JAR OBS packaging and a separate container-image deployment path; supported versions, regional availability (e.g. some runtimes only in ME-Riyadh/TR-Istanbul), and deprecation timelines are Huawei-specific. Concurrency, cold-start (FunctionGraph advertises snapshot-based warm starts), instance, and timeout quotas are governed by FunctionGraph's Notes and Constraints rather than Lambda's account-level concurrency controls.

Operational responsibility separates along service-adjacent lines. Lambda observability leans on CloudWatch Logs, X-Ray, and Lambda Insights. FunctionGraph integrates with Log Tank Service (LTS) for logs and Application Operations Management (AOM) for metrics, and orchestration uses FunctionGraph's visual workflow engine (aligned to the CNCF Serverless Workflow spec) instead of AWS Step Functions. There is no documented Huawei one-click Lambda-to-FunctionGraph migration tool; re-platforming requires rebuilding functions, triggers, IAM-equivalent agencies, and observability against Huawei's services.

Start with an assessment that inventories every function, its runtime and version, packaging format (ZIP/JAR/container), concurrency and timeout settings, memory/cpu shape, and the full set of triggers and downstream integrations (S3, DynamoDB, SQS, EventBridge, API Gateway, X-Ray). Map each Lambda runtime to a supported FunctionGraph runtime where one exists, flagging unsupported versions or container-only workloads for a custom-runtime or container-image path. Confirm target regional availability, since some FunctionGraph runtimes are limited to specific regions. Treat FunctionGraph as the core equivalent, with APIG for HTTP endpoints, OBS for object triggers, DIS for stream triggers, and SMN for notifications covering integration gaps.

Re-platform function code rather than lift-and-shift: rewrite handler signatures to FunctionGraph's event/context/callback contract, re-package as inline/ZIP/JAR or as a container image per Huawei's packaging rules, and recreate each trigger against the Huawei equivalent event source (OBS for S3, DIS for Kinesis/DynamoDB Streams, SMN for SNS, APIG for API Gateway). Re-implement IAM roles as FunctionGraph agencies and reconfigure environment variables, layers (no direct Layers equivalent, so vendor dependencies into the package), VPC/networking, and reserved/provisioned concurrency using FunctionGraph concurrency controls.

Validate before cutover with parallel shadow runs: replay recorded events through both Lambda and FunctionGraph and diff outputs, latency, and error semantics. Recreate observability by piping FunctionGraph logs to LTS and metrics to AOM, and define alarms equivalent to your CloudWatch alarms. Confirm quotas (max instances, timeout, package size) and run load tests at expected peak concurrency to surface throttling or cold-start behavior before switching traffic at the APIG edge.

Mind the cost-model and governance changes. Both services bill per request and execution duration, but FunctionGraph pricing tiers, free-tier allowance (1 million monthly requests advertised), memory/cpu granularity, and data-transfer/egress rates differ from Lambda plus you lose Lambda's integrations with AWS-only edge and data services. Recompute TCO using your actual invocation volume, duration distribution, payload sizes, and cross-service egress. Plan for runtime-deprecation deadlines published by FunctionGraph, and budget for re-development effort since no automated source-to-target migration path is documented.