Skip to content

Latest commit

 

History

History
1229 lines (1014 loc) · 52.6 KB

README.md

File metadata and controls

1229 lines (1014 loc) · 52.6 KB
Raw

tnm4j

Build Status Maven Central

A simplified SNMP API for Java, inspired by Jürgen Schönwälder's Tnm extension for Tcl.

The original Tnm made it easy to write network management applications using simple Tcl scripts. Tnm4j attempts to bring this same simplicity to the task of writing network management applications in Java or in Java-based scripting languages such as Groovy.

Running the Examples Using Docker

The src/examples/java subdirectory contains several examples.

Running the examples is easy if you have Maven and if you have or (or are willing to install) Docker Desktop and Docker Compose on your workstation.

The src/examples/docker subdirectory contains a Dockerfile that can be used to create a Linux-based container image that runs Net-SNMP, with configuration that matches up with the src/examples/java/ExampleTargets. At the base directory for this project, there is a docker-compose.yml that can build and run the container for you.

Steps for running the examples:

  1. Install Docker Desktop

  2. Install Docker Compose

  3. Open a shell and navigate to the base directory for this project.

  4. Run Docker Compose in the base directory.

    docker-compose up --build

    The first time you start up, you'll see the container image being built. Subsequently, when you run docker-compose up --build it should reuse the cached image.

    After the image is built, it will run and you'll see the Net-SNMP console output.

  5. In another shell, navigate to the base directory for this project. In this shell you'll run an example using Maven as follows.

    mvn -Pexamples clean compile exec:java -Dexec.mainClass=Example01_GetAndGetNext

    Maven will build the project and run the example class you specified. You can run any of the examples in src/examples/java in this same manner.

  6. After you're done playing with the examples, go back to the first shell, hit Ctrl-C on the keyboard and then

    docker-compose down

  7. You can save a little disk space by getting rid of the container image, too.

    docker image rm tnm4j-netsnmp

Architecture

Tnm4j provides a lightweight façade over an SNMP adapter and a MIB parser adapter. These adapters each implement a Tnm4j service provider interface to adapt a third-party library for use with Tnm4j. The JDK's ServiceLoader mechanism is used to locate adapters for SNMP and MIB parsing support.

The default SNMP adapter uses Snmp4j by Frank Fock and Jochen Katz. Snmp4j is an outstanding library providing comprehensive support for SNMP communications in Java.

The default MIB adapter uses Per Cederberg's excellent Mibble MIB parser.

Other SNMP providers or MIB parsers could be easily adapted for use with Tnm4j by implementing the necessary SPI.

Trivial Example

The following example illustrates just a few of the features of Tnm4j. This snippet retrieves and displays the name, description, and up time of an SNMP-enabled network device.

MIB mib = MIBFactory.getInstance().newMib();
mib.load("RFC1213-MIB");

SnmpV2cContext snmp = SnmpFactory.getInstance().newSnmpV2cContext(mib);
snmp.setAddress("10.0.0.1");
snmp.setCommunity("public");

VarbindCollection varbinds = snmp.getNext("sysName", "sysDescr", "sysUpTime").get();
for (Varbind varbind : varbinds) {
  System.out.format("%s=%s\n", varbind.getName(), varbind.toString());
}

Notice how the MIB and SNMP context objects are designed to work together? This is arguably the most salient concept of Tnm4j. Tnm4j fully exploits the MIB to make it easy for the developer to access management objects in an SNMP agent. The developer can use MIB object names (instead of SNMP object identifiers) when getting or setting object values, reducing the time and effort required to get the desired management information.

The syntax and textual convention details from the MIB are used when converting object values to strings. This is illustrated in the preceding example -- converting a retrieved value to a string in the appropriate format is as simple as calling Varbind.toString().

Getting Started: Targets, Contexts, and Operations

In using Tnm4j, there are three fundamental objects you will use to interact with SNMP agents: targets, contexts, and operations.

A target is a simple object that describes the relevant characteristics of a remote agent necessary for communication. A target implements either the SnmpV3Target, SnmpV2cTarget, or SnmpV1Target interface, depending on whether the remote agent supports the SNMPv3, SNMPv2c, or SNMPv1 protocol, respectively. These interfaces describe properties such as network address and security characteristics of the remote agent. Tnm4j provides simple concrete target implementations -- SimpleSnmpV3Target, SimpleSnmpV2cTarget, and SimpleV1Target -- that your application can construct and configure directly. Alternatively, your domain model objects representing network devices could easily implement these interfaces, allowing your model objects to be used directly as targets.

A context provides the ability to invoke SNMP operations on a given target. You create a context using SnmpFactory and providing the target to the factory method:

import org.soulwing.snmp4j.*;

SimpleSnmpV2cTarget target = new SimpleSnmpV2cTarget();
target.setAddress("10.0.0.1");
target.setCommunity("public");

SnmpContext context = SnmpFactory.getInstance().newContext(target);
try {
  // perform some SNMP operations
}
finally {
  context.close();
}

As shown in the preceding snippet, a context must (eventually) be closed, in order to avoid a resource leak. Contexts are, however, lightweight objects and it is quite reasonable to create and retain a context for as long as you need to continue communicating with the target SNMP agent. Depending on the needs of your application, this could be as short as the time needed to perform a few SNMP operations, or perhaps for as long as the remote agent exists and your application continues running. Tnm4j is used in applications that retain thousands of context objects on the heap.

If you neglect to close a context, it will eventually be closed when your code no longer holds any references to it -- the underlying implementation implements finalize to close the context if necessary. However, as it is difficult to predict when a discarded context will be reclaimed by the JVM's garbage collector, it is a best practice to close a context before you discard the last reference to it.

Performing SNMP Operations

Once you have a context for a given target, you can use the context to perform SNMP operations on the targeted SNMP agent:

SnmpContext context = SnmpFactory.getInstance().newContext(target);
try {
  VarbindCollection result = context.get("1.3.6.1.2.1.1.3.0").get();
  System.out.println(result.get(0));
}
finally {
  context.close();
}

The preceding snippet uses the context to invoke a GET operation on the remote agent and stores a reference to the retrieved variable bindings -- varbinds in SNMP speak. The numeric object ID used here (in case you didn't recognize it) is the SNMP sysUpTime object. When invoking an operation, we can request an arbitrary number of SNMP objects; the operation methods include variants which take a variable number of arguments or a List.

If you're wondering why we're using numeric OIDs here instead of names, just hang on... we'll get there shortly!

In addition to the get, the context provides methods to support all of the fundamental SNMP operations: GET, GETNEXT, GETBULK, and SET. Moreover, it provides methods to support easy and efficient SNMP table walks, which we'll cover later. See the javadoc for the full details.

You might have noticed there are quite a few gets in those two lines of code inside of the try block. The snippet is written in the idiomatic style recommended for Tnm4j, which is lean on syntax, and hides a lot of the underlying details. Let's break it down a little more to help you understand what's going on. We could rewrite the snippet a little more verbosely, like this:

SnmpContext context = SnmpFactory.getInstance().newContext(target);
try {
  SnmpResponse<VarbindCollection> response = context.get("1.3.6.1.2.1.1.3.0");
  VarbindCollection result = response.get();
  Varbind sysUpTime = result.get(0);
  System.out.println(sysUpTime);
}
finally {
  context.close();
}

Now we can see that when we use get to perform a GET operation, the return value is an SnmpResponse. If you check out the javadoc for SnmpResponse you'll see that it has a single method (get) that retrieves the result of the SNMP operation. The response object is patterned after the JDK's Future object -- the get method will block until the result of the operation is available. If the operation fails, the relevant exception will be thrown when you try to get the result from the response object.

Assuming that the operation succeeds, the result we retrieve from the response object is a VarbindCollection. This object is not a subtype of the JDK's Collection type. However, it has an interface with methods that have familiar signatures supporting both list-like and map-like access to the varbinds in the collection. In this example, we're using a list-like getter that takes an index -- since we requested only one varbind in the GET operation, there is exactly one varbind in the result (and it has index 0).

In addition to the access methods provided on the VarbindCollection interface, you can use the asList or asMap methods to efficiently coerce the varbind collection to a JDK List or Map. This feature is especially handy when using Tnm4j with JavaEE's expression language (EL) or in a scripting language such as Groovy, which provides a lot of sweet syntax sugar for lists and maps.

Using MIBs and Named Objects

One of the major benefits of Tnm4j is that it fully integrates the textual description of SNMP managed objects known as the Management Information Base or MIB. By loading one or more MIB module definitions, you can specify SNMP objects by name when invoking operations and retrieving results.

In Tnm4j the Mib object is used to load MIB module definitions. You create a Mib using MibFactory. When creating context objects you can specify the MIB to be used by the context. All contexts sharing the same MIB have access to the MIB modules loaded into the MIB.

Let's rework our previous example, to make use of a MIB.

import org.soulwing.snmp4j.*;

Mib mib = MibFactory.getInstance().newMib();
mib.load("SNMPv2-MIB");

SimpleSnmpV2cTarget target = new SimpleSnmpV2cTarget();
target.setAddress("10.0.0.1");
target.setCommunity("public");

SnmpContext context = SnmpFactory.getInstance().newContext(target, mib);
try {
  VarbindCollection result = context.getNext("sysUpTime").get();
  System.out.println(result.get("sysUpTime"));
}
finally {
  context.close();
}

Note that not only do we ask for the sysUpTime object by name when invoking the operation, we can also use the name to retrieve the value from the resulting varbind collection.

If you're reading closely, you might have noticed that we used getNext to invoke a GETNEXT operation -- in the previous example we used a GET operation. The reason for this change involves something fundamental about the way management objects are represented by SNMP.

A Brief Diversion on the topic of Object Indexes

In SNMP, every managed object type has a unique object identifier. For example, the standard MIB includes an interface table with an object named ifDescr that describes a network interface. The unique identifier of the ifDescr object type is 1.3.6.1.2.1.2.2.1.2. Similarly, the sysUpTime object type is identified as 1.3.6.1.2.1.1.3.

Every SNMP object instance has an index that is appended to the object identifier for the object type. For an object like ifDescr this makes sense -- a network device often has many network interfaces, so SNMP needs to be able to uniquely identify every instance of ifDescr (one per network interface). The third instance of ifDescr (corresponding to the third network interface managed by the agent) might have an index of 3, resulting in an object instance identifier of 1.3.6.1.2.1.2.2.1.2.3; i.e. the type identifier with .3 appended to the end.

What is often surprising to newcomers to SNMP, is that even object types with singleton instances have an index. So sysUpTime, for which a given agent only ever has one instance, must have an index. In SNMP, the index for singleton object instances is 0. This index is appended to the end of the object type identifier, just like any other index. In our first examples, we used 1.3.6.1.2.1.1.3.0 as the object identifier to GET from the target agent. Now we know that this identifier refers to the singleton instance of sysUpTime.

In our last example, we used a GETNEXT operation. Extracting out just the two most relevant lines of code, we had this:

VarbindCollection result = context.getNext("sysUpTime").get();
System.out.println(result.get("sysUpTime"));

So why did we call getNext instead of get? In SNMP, the GETNEXT operation returns the object instance whose identifier is the successor of the identifier specified in the operation. Due to the fact that the index subidentifier is appended to the object type identifier, the type identifier is always the predecessor of the first instance of that type.

For example the type identifier for sysUpTime is 1.3.6.1.2.1.1.3, which is always the predecessor of object 1.3.6.1.2.1.1.3.0 (the singleton instance of sysUpTime). When we issue a GETNEXT for sysUpTime, the agent returns its successor; namely sysUpTime.0.

You might be wondering if we could have also used a GET operation and specified the object as sysUpTime.0. Indeed we could have done so, and the reason we didn't is mostly a question of style -- using GETNEXT to retrieve instances of singleton object types is the idiomatic choice in SNMP.

The other to thing to note about our example is that even though the retrieved sysUpTime object instance has an index of 0, we retrieved it from the result object (VarbindCollection) without specifying the index. The collection could only contain one instance of any given object type, so making you specify the index would be redundant. Coming up shortly, we learn how to retrieve MIB tables (such as the standard MIB's interface table). We'll find it very convenient that we don't have to know the index of the row in order to retrieve a row's columns.

Setting Object Values

In addition to retrieving object values from a remote agent, we can set object values. For example, we can set the value for the sysContact object by invoking the set method on the context:

import org.soulwing.snmp4j.*;

Mib mib = MibFactory.getInstance().newMib();
mib.load("SNMPv2-MIB");

SimpleSnmpV2cTarget target = new SimpleSnmpV2cTarget();
target.setAddress("10.0.0.1");
target.setCommunity("public");

SnmpContext context = SnmpFactory.getInstance().newContext(target, mib);
try {
  Varbind vb = context.newVarbind("sysContact.0", "nobody@nowhere.net");
  VarbindCollection result = context.set(vb).get();
  System.out.println(result.get("sysContact"));
}
finally {
  context.close();
}

Notice that when specifying a variable binding to set an object value, we must use an object instance identifier -- in this case we are specifying the identifier for the singleton instance of the sysContact object.

We pass one or more variable bindings to the set method. Each variable binding associates a value with an object identifier. The context has a factory method that creates Varbind objects given an object identifier and an object value. The value you pass must be compatible with the data type of the associated SNMP object.

We can also pass a VarbindCollection to the set method. This allows you to fetch a current value for an object, change the value, and update it on the remote agent:

try {
  VarbindCollection vbs = context.getNext("sysContact").get();
  result.get("sysContact").set("nobody@nowhere.net");
  VarbindCollection result = context.set(vbs).get();
  System.out.println(result.get("sysContact"));
}
finally {
  context.close();
}

Retrieving Rows from Conceptual Tables

SNMP represents tabular information by allowing a particular object type to have many instances, each identified with a unique subidentifier appended to the object type's identifier. For example, in A Brief Diversion on the topic of Object Indexes we learned that the standard MIB's network interface table has perhaps many instances of the ifDescr object type; one for each network interface managed by an agent.

They're called "conceptual" tables, because their representation isn't really all that tabular. A table "column" is really just an object type, and because of the way object instances are identified, all of the columns of a conceptual table are siblings in the MIB tree, and all of the values of a column are siblings in a subtree under the column's type object. The concept of a "row" really only arises from the way we specify which objects to retrieve in a GET/GETNEXT operation.

Table Retrieval Using GETNEXT

Let's start with an example that shows the most primitive (and slowest!) way to retrieve all of the rows of a table -- using the GETNEXT operation.

import org.soulwing.snmp.*;

Mib mib = MibFactory.getInstance().newMib();
mib.load("SNMPv2-MIB");
mib.load("IF-MIB");

SimpleSnmpV2cTarget target = new SimpleSnmpV2cTarget();
target.setAddress("10.0.0.1");
target.setCommunity("public");

SnmpContext context = SnmpFactory.getInstance().newContext(target, mib);
try {
  final String[] columns = {
      "sysUpTime", "ifName", "ifDescr", "ifAdminStatus", "ifOperStatus",
      "ifInOctets", "ifOutOctets"
  };
  VarbindCollection row = context.getNext(columns).get();
  while (row.get("ifName") != null) {
    if (row.size() < columns.length) {
      System.err.println("truncated row; too many objects requested");
      break;
    }
    System.out.format("%14s %-8s %-20s %-4s %-4s %,15d %,15d\n",
        row.get("sysUpTime"),
        row.get("ifName"),
        row.get("ifDescr"),
        row.get("ifAdminStatus"),
        row.get("ifOperStatus"),
        row.get("ifInOctets").asLong(),
        row.get("ifOutOctets").asLong());
    row = context.getNext(row.nextIdentifiers("sysUpTime")).get();
  }
}
finally {
  context.close();
}

Assuming you're already familiar with basic SNMP, this example should be fairly obvious. We start with a getNext using the object type identifiers for some of the columns of the standard MIB's network interface table. Assuming that there is at least one network interface managed by the target agent, we'll enter the while loop -- if the result contains an ifName object then we have a table row. In the loop body, after we extract and print the column values for the current "row", we ask the row object to produce a list of object identifiers that will be used to request the next row and then perform another GETNEXT operation.

This example also illustrates the common practice of including sysUpTime in the list of object identifiers for each GETNEXT operation. This provides an agent-centric time basis for computing rates for counters such as the input and output octet counters. Including a singleton object type like sysUpTime means that we can't simply use all of the object identifiers for the current row in the request for the next row -- the successor to sysUpTime.0 is sysContact.0! The nextIdentifiers method we call on the row object (a VarbindCollection) provides a handy solution. We can provide it a list of the non-repeating (singleton) objects in the collection, and it will produce a list of object identifiers consisting of the specified non-repeating object identifiers followed by the identifiers associated with the other (repeating) objects in the collection.

It's possible to ask for too many objects in a single GETNEXT request. We check for that case by comparing the size of the row to the number of objects we requested. Requesting too many objects in a single request is a programming error, so we simply print an error message and exit the loop.

Table Retrieval Using GETBULK

If you ran the previous example to collect the interface table from a real network device, you probably noticed that it's a bit slow. Using GETNEXT, we have to make a full round trip over the network for each entry in the target SNMP agent's interface table. You can probably imagine that this approach is not going to work very well for a something like an IP route table that could contain hundreds of thousands of rows.

In SNMPv2, the GETBULK operation was introduced to the protocol to provide a solution for improving table retrieval performance. Conceptually, the GETBULK operation allows us to perform the while loop in the previous example on the remote agent itself. Just like GETNEXT, the GETBULK operation takes a list of object identifiers. As we've seen, when executing the GETNEXT operation, the agent returns the successor object instance for each of the specified object identifiers. With GETBULK, the agent returns the next N successor object instances for each of the specified identifiers.

We specify the maximum value for N as one of the parameters to the GETNEXT operation; the protocol specification calls this parameter max-repetitions. The value we specify indicates the maximum number of "rows" we want to retrieve in a single GETBULK operation. Of course, the agent may choose to return fewer than the number of rows we specify.

In our previous example using GETNEXT, we included sysUpTime in the request. Since GETBULK effectively performs N GETNEXT operations, we need a way to tell it that some of the object identifiers in the request are for non-repeating objects. The GETBULK operation includes a non-repeating parameter for this purpose -- this parameter is used to indicate that the first k identifiers in the list are for non-repeating object types.

Performing a GETBULK operation with Tnm4j is easy enough:

List<VarbindCollection> rows = context.getBulk(1, 10, 
    "sysUpTime", "ifName", "ifInOctets", "ifOutOctets").get();

The first parameter to getBulk indicates that there is one non-repeating object type in the list (sysUpTime). The second parameter indicates that we want the agent to return as many as 10 rows for the specified columns of the interface table. The remaining parameters identify the object names we wish to retrieve. The return value from getBulk is a list of VarbindCollection objects. Based on the parameters specified in this example, the list will contain as many as 10 collections -- one per retrieved row.

By now you're probably wondering, How will I know what to choose as the maximum number of repetitions? In general, there is no way to know the number of rows in a conceptual table without actually retrieving it. So the answer is that you're going to have to make a guess. Guessing will have two potential consequences. Our guess could be too small, in which case there will be more rows in the table that our GETBULK operation did not retrieve -- in this case we'll want to do one or more subsequent GETBULK operations to complete the retrieval. Our guess could be too big, in which case GETBULK might end up retrieving objects that aren't part of the table we wanted to retrieve -- in this case we'll need to discard the stuff we didn't really want.

Complicating matters even further, the number we choose is just a hint to the remote agent. It may choose to return fewer than the number we specified. Even worse, the remote agent is allowed to return a partial last row -- i.e. it is allowed to truncate the last row if the entire row won't fit neatly into an SNMP PDU.

If all of this is making you feel like you'll just deal with the poor performance of table retrieval using GETNEXT, we have good news! In the next section, we discuss using Tnm4j's high level table walk operation, which allows you to get all of the performance benefits of using GETBULK to retrieve tables, without the complexity of using GETBULK directly.

If we take all of these facts into consideration, we can write a table retrieval loop that uses GETBULK something like this:

Mib mib = MibFactory.getInstance().newMib();
mib.load("SNMPv2-MIB");
mib.load("IF-MIB");

SimpleSnmpV2cTarget target = new SimpleSnmpV2cTarget();
target.setAddress("10.0.0.1");
target.setCommunity("public");

SnmpContext context = SnmpFactory.getInstance().newContext(target, mib);
try {
  final int maxRepetitions = 10;

  final String[] columns = {
      "sysUpTime", "ifName", "ifDescr", "ifAdminStatus", "ifOperStatus",
      "ifInOctets", "ifOutOctets"
  };

  List<VarbindCollection> rows =
      context.getBulk(1, maxRepetitions, columns).get();

  outer:
  while (!rows.isEmpty()) {
    VarbindCollection lastRow = null;
    inner:
    for (VarbindCollection row : rows) {

      if (row.get("ifName") == null) break outer;    // doesn't look like ifTable columns

      if (row.size() < columns.length) break inner;  // truncated row... discard it and try again

      lastRow = row;
      System.out.format("%14s %-8s %-20s %-4s %-4s %,15d %,15d\n",
          row.get("sysUpTime"),
          row.get("ifName"),
          row.get("ifDescr"),
          row.get("ifAdminStatus"),
          row.get("ifOperStatus"),
          row.get("ifInOctets").asLong(),
          row.get("ifOutOctets").asLong());
    }
    if (lastRow == null) {
      // if this is the first row, we're asking for too much in a single row
      System.err.println("truncated first row; too many objects requested");
      break outer;
    }
    rows = context.getBulk(1, maxRepetitions,
        lastRow.nextIdentifiers("sysUpTime")).get();
  }
}
finally {
  context.close();
}

(You know an algorithm is messy when you find yourself needing labeled break statements!)

In the first couple of lines inside of the try block, we set up for the first getBulk request. We put the identifiers into an array so we can later use the array's length to check for a truncated row. Our array contains sysUpTime so in the call to getBulk we indicate that there is one non-repeating object at the front of the list.

The loop labeled outer is going to be used to keep doing getBulk calls until there are no more rows available (or we fall out the loop because we've reached the end of the table).

The loop labeled inner inspects each of the rows to make sure it is still actually looking at the table we want (the interface table) and to make sure the current row hasn't been truncated by the agent. Assuming the row is good, we print its contents. In order to be ready to do the next call to getBulk we need to keep track of the last complete row we processed, so we can use the identifiers in it as the starting point for the next GETBULK operation -- we assign the row to lastRow for this purpose.

If we fall out of the inner loop without having processed at least one full row, that means we're asking for too many objects in a single request -- we did something similar in our example using GETNEXT.

At the bottom of the outer loop we use the identifiers in the last full row for the next call to getBulk. This means that the next GETBULK operation will pick up right where the last one left off.

Table Retrieval Using Tnm4j's High-Level Walk Operation

As we saw in the previous section, retrieving a full table using GETBULK is tricky. For this reason, Tnm4j includes a high-level walk operation that uses GETBULK under the covers, but presents an iterator-like interface to your code.

Here's an example of the same retrieval of the standard interface table we used in the preceding examples, using Tnm4j's walk operation:

Mib mib = MibFactory.getInstance().newMib();
mib.load("SNMPv2-MIB");
mib.load("IF-MIB");

SimpleSnmpV2cTarget target = new SimpleSnmpV2cTarget();
target.setAddress("10.0.0.1");
target.setCommunity("public");

SnmpContext context = SnmpFactory.getInstance().newContext(target, mib);
try {
  SnmpWalker<VarbindCollection> walker = context.walk(1, "sysUpTime", "ifName",
      "ifDescr", "ifAdminStatus", "ifOperStatus", "ifInOctets", "ifOutOctets");
  VarbindCollection row = walker.next().get();
  while (row != null) {
    System.out.format("%14s %-8s %-20s %-4s %-4s %,15d %,15d\n",
        row.get("sysUpTime"),
        row.get("ifName"),
        row.get("ifDescr"),
        row.get("ifAdminStatus"),
        row.get("ifOperStatus"),
        row.get("ifInOctets").asLong(),
        row.get("ifOutOctets").asLong());
    row = walker.next().get();
  }
}
finally {
  context.close();
}

As you can see, using walk is the easier than using either GETBULK or GETNEXT. We obtain an SnmpWalker object from the context, using the walk method with a parameter list that is essentially the same as what we used for getBulk; the first parameter is the number of non-repeating object types in the list, followed by the names of the non-repeating objects (just sysUpTime), followed by the names of the columns we want to retrieve from a table. The walker has a next method that executes GETBULK operations with the remote agent as needed to retrieve all of the rows from the table.

While not shown here, the return value from next is an SnmpResponse just as it is for any other operation. We invoke get on the return value from next in order to get the VarbindCollection representing the next row. When get returns null the walk is complete. Just as it does for other operations, get will throw an exception if an error occurs in performing the underlying operations with the SNMP agent.

Because the high-level walk operation is implemented directly on top of Tnm4j's SNMP provider, it performs better than any table retrieval operation that you could write using the getBulk operation.

If you thought writing table retrieval using GETBULK was tricky, imagine writing an asynchronous implementation that doesn't block while waiting for the response to a GETBULK request! The good news is that you won't have to write that code yourself, because Tnm4j has done it for you. Just as it does for all of the native SNMP operations, Tnm4j provides an asynchronous variant of the high level walk operation. This allows you to walk arbitrarily large tables without blocking, and allows concurrent retrieval of tables from multiple agents without requiring a retrieval thread per agent. See Asynchronous Operations for details.

Accessing Table Index Objects

In SNMP version 2, the management information structure for tables was changed such that all index objects for tables are specified as not-accessible. Tables that were defined prior to this change (such as the interface table of the standard MIB) continue to provide accessible index objects, but all other tables are subject to this restriction. This restriction makes the agent implementation more efficient, but greatly complicates the work of the network management application.

Table index objects often contain information that is important to network management applications. For example, suppose that our management application wants to use the IPV6-MIB to obtain the IPv6 addresses that are configured on the interfaces of a network device. The ipv6AddrTable contains the information that we'd like to retrieve. If you review the definition of this table, you'll note that the most important pieces of information in this table -- the interface index and the address itself -- cannot be retrieved from the table; the ACCESS-TYPE of these objects is not-accessible.

Tnm4j makes it easy to access this information (which is encoded in the object identifiers of the other objects in the table), as shown in the following example:

MIB mib = MIBFactory.getInstance().newMib();
mib.load("IPV6-MIB");

SimpleSnmpV2cTarget target = new SimpleSnmpV2cTarget();
target.setAddress("10.0.0.1");
target.setCommunity("public");

SnmpContext context = SnmpFactory.getInstance().newContext(target, mib);
try {
  SnmpWalker<VarbindCollection> walker = context.walk("ipv6AddrPfxLength",
      "ipv6AddrType", "ipv6AddrAnycastFlag", "ipv6AddrStatus");
  VarbindCollection row = walker.next().get();
  while (row != null) {
    System.out.format("%d %s %d %s %s %s\n",
        row.get("ipv6IfIndex").asInt(), row.get("ipv6AddrAddress"),
        row.get("ipv6AddrPfxLength").asInt(), row.get("ipv6AddrType"),
        row.get("ipv6AddrAnycastFlag"), row.get("ipv6AddrStatus"));
    row = walker.next().get();
  }
}
finally {
  context.close();
}

When using any of the context's low-level or high-level operation methods, the index objects are implicitly available in the resulting VarbindCollection, even though they were not (and cannot) be among the objects retrieved from the table. Note that this feature requires that you load the relevant MIB(s) and provide the resulting Mib object to your context object(s).

The Varbind.getIndexes method also provides access to a table object's indexes.

Asynchronous Operations

Tnm4j fully supports asynchronous SNMP operations. When using asynchronous operations, your application does not block while waiting for a response from an SNMP agent. Tnm4j's asynchronous operations support can support concurrent communication with hundreds or even thousands of SNMP agents using just a few service threads.

Asynchronous operations are a key ingredient for writing responsive SNMP network management applications that scale to support real networks. Communication with SNMP agents can be handled in the background, while application users continue to interact with the application's user interface.

Effective use of asynchronous SNMP operations in Tnm4j requires a solid understanding of Java's facilities supporting concurrency: threads, locks, conditions, and the like.

The SnmpContext interface provides async- methods for each of the low level SNMP operations and the high-level walk operation: asyncGet, asyncGetNext, asyncGetBulk, asyncSet, and asyncWalk.

Example: An asynchronous GETNEXT operation.

Let's look at example that uses an asynchronous GETNEXT operation to retrieve the sysName and sysUpTime objects from a target SNMP agent. The setup looks very similar to our prior examples, with one salient difference:

Mib mib = MibFactory.getInstance().newMib();
mib.load("SNMPv2-MIB");

SimpleSnmpV2cTarget target = new SimpleSnmpV2cTarget();
target.setAddress(System.getProperty("tnm4j.agent.address", "10.0.0.1"));
target.setCommunity(System.getProperty("tnm4j.agent.community", "public"));

SnmpContext context = SnmpFactory.getInstance().newContext(target, mib);

SnmpCallback<VarbindCollection> callback = new ExampleCallback();
context.asyncGetNext(callback, "sysName", "sysUpTime");

// go do something else while waiting for the callback

When using an asynchronous SNMP operation, we must provide a callback that will be notified when a response is available from the remote SNMP agent. In the example we construct an ExampleCallback (that we'll see in just a moment). We invoke asyncGetNext on the context, passing a reference to our callback and the list of objects we wish to retrieve. After we initiate the GETNEXT request our application can go on to do some other work -- the callback will be invoked later (via a different thread) when a response is available.

The callback implements the SnmpCallback<VarbindCollection> interface. In our example implementation below, we simply print the objects received from the remote agent.

class ExampleCallback implements SnmpCallback<VarbindCollection> {

  @Override
  public void onSnmpResponse(SnmpEvent<VarbindCollection> event) {
    try {
      VarbindCollection result = event.getResponse().get();
      System.out.format("%s uptime %s\n",
          result.get("sysName"),
          result.get("sysUpTime"));
    }
    catch (SnmpException ex) {
      ex.printStackTrace(System.err);
    }
    finally {
      event.getContext().close();
    }
  }

}

When we previously looked at the get method in detail, we observed that its return value is SnmpResponse<VarbindCallback> -- when an invoking a synchronous SNMP operation, we get an SnmpResponse which contains a VarbindCollection result. In an asynchronous operation, the callback is provided an event object that contains an SnmpResponse (which in turn contains a VarbindCollection). The event object also provides a reference to the same context object that was used to invoke the operation.

As we noted previously, the response object is patterned after the JDK's Future object. When we invoke get the response object can block until a response is available. However, when we invoke get on the response object inside of our callback, it will never block -- the callback is not invoked until a response is available.

If a timeout or other error occurs while executing the asynchronous operation, the call to get on the response object will throw an appropriate exception. This allows our callback to handle the exception using an ordinary try block.

In this example, the context object is closed before the callback returns. As we discussed previously, context objects are lightweight, but should be closed when no longer needed. In our example, after the callback is finished, we don't need the context any more so we close it. In your design, you might choose to retain the context for subsequent operations, and that's fine too -- you just need to close the context when it really is no longer needed.

If you plop this example into a main method, as is, you'll probably find that it exits without printing anything. Because our example doesn't really have anything else to do, it terminates before the response is received from the remote agent. In a real application, this won't be a problem -- you wouldn't use asynchronous operations if the application didn't have better things to do than hang around waiting for a response from the remote SNMP agent.

To make the example work, you could easily add a Thread.sleep after the call to asyncGetNext to put the main thread to sleep for long enough for a response to be received. This is imprecise, but good enough for this silly example. In the example code provided with Tnm4j, the callback is modified so that the main method can block until a response is received.

Using a Completion Service to Collect Results

Often, in creating network management applications you will write code that simply needs to collect and record information from many different SNMP agents. An SnmpCompletionService can be used to simplify this work. The completion service provides provides a submit method to which you can submit SnmpOperation object instances for asynchronous execution. The service has a queue like interface that you can use to retrieve SnmpEvent objects for completed operations, in either a blocking on non-blocking manner.

In addition to the methods such as getNext and asyncGetNext that directly execute SNMP operations, SnmpContext provides factory methods that create SnmpOperation objects. An operation object provides two overloads of its invoke method, providing for either synchronous or asynchronous execution. You can create operation objects for operations you wish to perform, and delegate the handling of the callback. SnmpCompletionService is designed around this concept.

Suppose we wanted to collect sysName and sysUpTime from many different network devices. Here's an example of how we might accomplish this using a completion service:

Mib mib = MibFactory.getInstance().newMib();
mib.load("SNMPv2-MIB");

SnmpCompletionService<VarbindCollection> completionService =
    new BlockingQueueSnmpCompletionService<VarbindCollection>();

completionService.submit(newOperation("10.0.0.1", "public", mib));
completionService.submit(newOperation("10.0.0.2", "public", mib));
completionService.submit(newOperation("10.0.0.3", "public", mib));

while (!completionService.isIdle()) {
  SnmpEvent<VarbindCollection> event = completionService.take();
  VarbindCollection result = event.getResponse().get();
  System.out.format("%s: sysName=%s sysUpTime=%s\n",
      event.getContext().getTarget().getAddress(),
      result.get("sysName"),
      result.get("sysUpTime"));

  event.getContext().close();
}

The basic structure of the main body of our example is quite simple. We submit operations on different network devices to the completion service, and then remain in the loop until all of the results have been obtained. While the example simply uses a few hard-coded agent addresses, we could easily imagine looking those details up in a database and submitting operations to the completion service for all of the devices in our network.

In the loop body, we can see that the take method on the completion service returns an SnmpEvent object, which we saw in the callback of our previous example. The take method blocks until the next event describing a completed operation becomes available. We extract the result from the event and print the information requested by the operation. Since we're done with the context associated with the event, we close it.

The newOperation method is shown below.

private static SnmpOperation<VarbindCollection> newOperation(String address,
    String community, Mib mib) {
  SimpleSnmpV2cTarget target = new SimpleSnmpV2cTarget();
  target.setAddress(address);
  target.setCommunity(community);

  return SnmpFactory.getInstance().newContext(target, mib)
      .newGetNext("sysName", "sysUpTime");
}

As we've seen in previous examples, we create a target that describes the SNMP agent on which we wish to execute an operation. We use SnmpFactory to obtain a context for our target, and we use the newGetNext factory method to obtain an SnmpOperation that will perform a GETNEXT for the sysName and sysUpTime objects.

The context interface provides factory methods such as newGetNext for all of the SNMP operations. An SnmpOperation encapsulates a context, an operation, and a list of SNMP objects on which the operation is to be performed. An operation can be invoked either synchronously or asynchronously. When you submit an operation to an SnmpCompletionService, the service invokes the operation asynchronously, providing a callback that will be used to queue the responses, making them available via the public API of the service.

Receiving SNMP Notifications (Traps, Informs)

SNMP agents can be configured to send notifications to a management application when certain events occur. For example, the standard MIB defines events that can notify a management application when a network link managed by an agent transitions to an up or down state, or to notify the application when an agent or the network device it is managing is restarted. Tnm4j makes it easy to respond to notifications received from SNMP agents in some manner that is appropriate for your network application.

SNMP defines two different notification types; TRAP and INFORM. An INFORM must be acknowledged by the recipient, allowing the agent to be assured of its delivery. For the agent, a TRAP is purely fire and forget -- the agent does not concern itself with whether a TRAP is actually received by any management application. In Tnm4j, the underlying SNMP provider takes care of acknowledging receipt of INFORM events, so the difference between TRAP and INFORM notifications is completely transparent to your application.

Tnm4j defines two fundamental objects for handling notifications from SNMP agents; listeners and handlers.

A listener is an instance of SnmpListener and is responsible for collaborating with the underlying SNMP provider to receive and route inbound notifications from remote agents to your application. Listener objects are created by SnmpFactory. A listener is configured to listen for notifications addressed to a particular port address. If your application needs to listen on more than one port address, you create a listener for each address.

The default port used by a listener is UDP port 162, which is defined as the standard notification port for SNMP. On most operating systems, this is a privileged port and can only be used by processes with superuser access. If a BindException is thrown when attempting to add a listener, try specifying a port number greater than or equal to 1024 (and less than 65536).

A handler is responsible for doing something useful with a received notification. Handlers implement the SnmpNotificationHandler interface. This interface declares a single handleNotification method that receives an event object that describes a notification and the agent from which it was received.

Here's an example of the basic setup:

Mib mib = MibFactory.getInstance().newMib();
mib.load("SNMPv2-MIB");

SnmpListener listener = SnmpFactory.getInstance().newListener(10162, mib);
try {
  listener.addHandler(new SnmpNotificationHandler() {
    @Override
    public Boolean handleNotification(SnmpNotificationEvent event) {
      System.out.println("received a notification: " + event);
      return true;
    }
  });

  Thread.sleep(60000L);     // wait for some notifications to arrive
}
finally {
  listener.close();         // listeners must be closed when no longer needed
}

The notification event object contains an SnmpNotification that provides the details of the received INFORM or TRAP. The getType method can be used to determine the notification type. Legacy SNMPv1 traps are a little different than SNMPv2 INFORM and TRAP notifications. When the trap type is TRAPv1, you may safely cast the notification object to SnmpV1Trap to access the additional properties defined for an SNMPv1 trap.

On most Unix hosts, you can use the snmptrap and snmpnotify commands (which are part of the Net-SNMP package) to test your notification handler. For example, the following shell commands can be used to send our example handler an INFORM, TRAP, and a legacy SNMPv1 TRAP, respectively:

snmpinform -v 2c -c public localhost:11162 {} 1.2.3.4 sysUpTime.0 t 218128
snmptrap -v 2c -c public localhost:11162 {} 1.2.3.4 sysUpTime.0 t 218128
snmptrap -v 1 -c public localhost:11162 enterprises.1446 10.0.0.1 0 0 ''

See the man pages for these commands for more details on how to use them to send notifications.

A given listener supports an arbitrary number of handlers and orders them according to a priority specified when each handler is registered. The listener-handler interaction is designed around the strategy pattern. When a notification is received, handlers are notified in priority order. Each handler is allowed to inspect the notification and decide whether to handle it. The first handler that returns true is the last handler that will receive the notification. See the javadoc for SnmpListener for more information.

The strategy-based design allows you to easily write highly cohesive notification handlers, each focused on handling a particular kind of notification, rather than having a single handler with hard-to-test (and hard-to-debug) conditional logic for sorting out what kind of notification was received and what do about it.

Best Practices

When using Tnm4j in an application, there are a few other best practices that you should observe.

Close the SnmpFactory Before Exit

SNMP operations invariably involve waiting for things to happen, and in Java, waiting invariably involves thread management. The singleton SnmpFactory instance holds references to a couple of thread pools used for making SNMP requests and scheduling timeouts. In order for your application to exit cleanly, your code must eventually invoke close on the SnmpFactory instance.

In a JavaSE application you can register a shutdown hook via the Runtime class to close the factory instance as shown here.

Runtime.getRuntime().addShutdownHook(new Thread(new Runnable() {
  public void run() {
    try {
      SnmpFactory.getInstance().close();
    }
    catch (InterruptedException ex) {
      Thread.currentThread().interrupt();
    }
  }
}));

In a JavaEE application or in similar frameworks such as Spring, you can use a method annotated using @PreDestroy in one of your application beans to close the SnmpFactory, as shown below.

Inject SnmpFactory as a Dependency

SnmpFactory provides a singleton instance via its getInstance() method, to make it relatively easy to use in simple Java SE applications. However, when using Tnm4j in an application framework such as Java EE, CDI, or Spring that supports dependency injection, you should create some form of producer method and use it to inject an SnmpFactory instance into beans that require it. This simplifies application design and makes it possible to mock the SnmpFactory for testing.

Using CDI (with or without Java EE), you can easily create a bean that acts as a producer for the SnmpFactory instance.

@ApplicationScoped
public class SnmpFactoryProducerBean {

  private SnmpFactory snmpFactory;
  
  @PostConstruct
  public void init() {
    snmpFactory = SnmpFactory.newInstance();
  }
  
  @PreDestroy
  public void destroy() {
    try {
      snnpFactory.close();
    }
    catch (InterruptedException ex) {
      Thread.currentThread().interrupt();
    }
  }
  
  @Produces
  public SnmpFactory snmpFactory() {
    return snmpFactory;
  }
}

In addition to providing an init method to create and store a reference to the SnmpFactory instance, this bean also illustrates the best practice of closing the factory when it is no longer needed.

Having defined a producer bean, you can then use standard dependency injection to get the SnmpFactory instance into any bean that needs it.

public class MySnmpAppBean {

  @Inject
  SnmpFactory snmpFactory;
  
  // ...
}

Use a ManagedThreadFactory in Java EE Applications

Java EE containers generally want to manage all resources used by an application, including any threads that the application needs to create. The SnmpFactory instance in Tnm4j needs to create threads to make SNMP requests, wait for results, and to dispatch control to user-provided callbacks when performing asynchronous SNMP operations.

As of Java EE 7, a Java EE container is obligated to provide a default ManagedThreadFactory instance that can be injected into application beans using an @Resource annotation. Most containers will also allow an administrator to define additional managed thread factory instances that can be injected into an application bean by specifying a JNDI name with the @Resource annotation.

It is a best practice, when using Tnm4j in a Java EE environment, to use a ManagedThreadFactory instance when creating the SnmpFactory instance. A simple approach to ensuring that Tnm4j uses a managed thread factory is to define a singleton EJB that is instantiated at application startup, as follows.

@EJB
@Singleton
public class SnmpFactoryConfiguratorBean {
  
  @Resource
  ManagedThreadFactory threadFactory;
  
  private SnmpFactory snmpFactory;
  
  @PostConstruct
  public void init() {
    snmpFactory = SnmpFactory.getInstance(threadFactory);
  }
  
  @PreDestroy
  public void destroy() {
    try {
      snnpFactory.close();
    }
    catch (InterruptedException ex) {
      Thread.currentThread().interrupt();
    }
  }
    
  @Produces
  public SnmpFactory snmpFactory() {
    return snmpFactory;
  }
  
}

Notice that this approach extends the prior example of using dependency injection to provide the SnmpFactory instance. It differs only in that it is an EJB and it initializes the SnmpFactory using a managed thread factory.