Why I Separated MIB Compilation from SNMP Execution in Python

When building network management tools, the author initially did not plan to create a MIB compiler or implement SNMP runtime. However, as the applications grew in complexity, maintaining MIB compilation and SNMP integration separately became increasingly problematic. This led the author to separate these responsibilities into two distinct components: MIB compilation and SNMP execution.

The author began by using PySMI and PySNMP, which were capable projects but faced challenges in how their integration fit into their applications. The existing architecture centered around compiling MIBs into Python modules for PySNMP, which created issues around error handling, dependencies, and runtime integration. The author sought a more stable and portable data contract for applications that did not require the same Python runtime.

To address these issues, the author made two key architectural decisions. First, they chose to make JSON the primary MIB artifact. The Trishul SMI compiler emitted versioned JSON modules, bundles, and OID indexes, allowing applications to consume compiled MIB knowledge without embedding a compiler or depending on PySNMP-generated Python modules. This separation of compilation and application code made the boundary explicit and enabled MIB artifacts to be created as part of a build or management operation.

Second, the author created Trishul SNMP, an asynchronous SNMP manager with its own protocol handling, notification handling, offline decoding, and simulation functionality. Trishul SNMP did not require Trishul SMI for numeric OID operations, allowing compiled MIB bundles to be optional. This kept the protocol runtime independent and avoided making a simple polling request dependent on a separate MIB compilation subsystem.

The architectural relationship between the components is as follows: MIB source files -> Trishul SMI -> Versioned JSON bundles (optional enrichment) -> Trishul SNMP -> SNMP devices -> Consumer applications. The runtime has deliberate limits, such as not being a full SNMP agent framework and not supporting SET operations.

The results of implementing these architectural decisions were significant. The Trishul SNMP Suite experienced a median GET operation time improvement from 93.9 ms to 3.0 ms on the Trishul integration compared to the legacy integration. While the performance gains varied depending on the specific use case, the most valuable outcome was having a reusable contract, clearer responsibilities, and a foundation that could evolve across projects.

In summary, separating MIB knowledge from SNMP execution proved to be a crucial architectural boundary, allowing for more maintainable and adaptable network management platforms. The choice between compiled Python MIB modules, portable JSON schemas, or runtime MIB parsing ultimately depends on the specific requirements and constraints of each project.