Agent Development

By Lex Li

This page shows you how the agent sample works.

In this article:

Dependencies

When you open snmpd.csproj in Visual Studio (or another IDE), you should first check what are its references.

Below are several references you should pay attention to,

  • SharpSnmpLib.Portable.dll and SharpSnmpLib.Full.dll are our #SNMP Library components.

Note

You might use iOS or Android specific files if targeting those platforms.

Facade and Pipeline

We no longer need to construct an Agent object, because that class is too old to be used. We try to construct an SnmpEngine object instead.

SnmpEngine is the facade that you should put on top. It is a very complex class that replaces our old Agent class. To construct it, we need: an EngineGroup object, which contains all engine global objects; A Listener object, which monitors incoming SNMP message; An SnmpApplicationFactory, who manages the pipelines.

SnmpApplicationFactory creates new pipelines when its pipeline pool is full of busy pipelines. The objects we pass to its constructor finally goes into each pipeline as they are shared among them (at this moment).

A pipeline is in fact an SnmpApplication object. It has several processing phases and each phase utilizes a few helper classes, which you can extend.

An SNMP message captured by Listener will be packaged as an SnmpContext object. This context object is passed to the pipeline and processed in each phase to generate a correct response message.

About how to construct instances of each classes, you can refer to snmpd sample source code.

Pipeline in Details

Currently we only have four phases in the pipeline,

  • Authentication
  • Request handler mapping
  • Request handler executing
  • Logging request

Authentication

By checking the sample code you can see how the primary membership provider (ComposedMembershipProvider) is created and added into the pipeline. The request is checked by the provider and dropped if it does not contain a proper community name.

ComposedMembershipProvider is a special membership provider, who allows you to support different SNMP versions. If you only target a specific version, you can use the version specific provider as primary one.

To customize authentication, make use of the existing providers or write your own.

Request handler mapping

For authenticated message, in this phase it is verified again and mapped to a message handler.

Message handlers are injected to the pipeline, too. So you can analyze the sample code to know how many handlers are there already, and how each registers its interested message (SNMP version and verb).

For example, GetV1MessageHandler is only interested in v1 GET messages, while GetMessageHandler in v2 and v3 GET messages.

Carefully configure the existing handlers, you can achieve different SNMP engine configurations, so as to meet different requirements. You can write your own handlers to further customize the pipeline.

Request handler executing

Once a message handler is found for the message, in this phase the handler performs the requested operation and generate a response message.

You should notice that *V1MessageHandler classes follow RFC 1157 specification to handle v1 messages, while other handler classes follow RFC 3416 to handle v2 and v3 messages.

Logging request

In this phase the response message is sent back, while the logger logs the processing into log files.

After phase 4, the pipeline is reused by SnmpApplicationFactory for future messages.

We may add an authorization phase to achieve user based authorization, but it is not yet designed and implemented.

ObjectStore and ISnmpObject

When a handler tries to do a typical SNMP operation, it looks into the ObjectStore object to locate the specified object.

Currently we only have a few sample objects created, to test out the pipeline. You can find them under Lextm.SharpSnmpLib.Objects namespace.

If you want to write more objects, you can follow our sample ones.

ObjectStore is not yet thread safe, which will be improved in the future.

Performance Tuning

The SNMP engine is multi-threading by nature. Both the ListenerBinding and SnmpApplication instances utilize the default thread pool to handle requests asynchronously. Thus, it is a must to make sure the thread pool is optimized before requests come in.

Before calling SnmpEngine.Start, it is recommended that the below code to be executed, which sets the minimal worker thread count to a suitable value.

int minWorker, minIOC;
// Get the current settings.
ThreadPool.GetMinThreads(out minWorker, out minIOC);
var threads = engine.Listener.Bindings.Count;
ThreadPool.SetMinThreads(threads + 1, minIOC);

If not tuned, the very first request to this agent will cost extra time (noticeably several seconds if there are too many bindings), as the operating system needs to create new threads before putting them into the thread pool.

Limitation Explained

You should take a look at MainForm.cs and read what extra lines are required to configure the SnmpEngine object, how to start and stop it. SNMP tables can be quite complex, while this sample only shows simple tables such as IfTable for simplicity.

As the sample is released under MIT/X11 license. The snmptrapd sample also uses the pipeline to handle trap messages, and once you are familiar with snmpd, you can switch to it to learn how to construct a browser side pipeline accordingly.

This sample is provided to demonstrate how the library might be used. If you want to build a full feature SNMP agent based on this sample, then many changes (mult-threading and security related) are mandate.