Traditional systems centralize execution in long-running servers. This makes failures contagious and deployment heavy. Graph-native computation takes the opposite path: each function is an isolated executable.
The Single-Function Executable
Each function:
- Reads rows from its input table
- Processes them
- Writes to its output table
- Logs its behavior
- Exits
There is no shared runtime. There is no global state.
Natural Parallelism
Because functions are isolated, you can run them in parallel across machines, cores, or accelerators. Parallelism is not a special optimization. It is the default.
Fault Isolation
If a function crashes, it doesn’t crash the system. Its input rows remain and can be retried. Other functions continue unaffected.
Elastic Distribution
You can scale out by launching more workers for popular functions. You can scale down by leaving idle workers asleep. Execution becomes a set of pulses, not a constant hum.
Why It Matters
- Operational simplicity: No global server to babysit.
- Flexibility: Different functions can run on different platforms or languages.
- Resilience: Failures are local, not systemic.
In this model, a system is not a single running process. It is a village of independent workers connected by shared tables and graph routes.