PySNMP tutorial

by Ilya Etingof, May 2006

Table of contents

  • 1. Network management basics
  • 2. Programming with PySNMP
    • 2.1 One-line Applications

      1. Network management basics

      As networks become more complex, in terms of device population, topology and distances, it has been getting more and more important for network administrators to have some easy and convenient way for controlling all pieces of the whole network.

      Basic features of a network management system include device information retrieval and device remote control. Former often takes shape of gathering device operation statistics, while latter can be seen in device remote configuration facilities.

      For any information to be exchanged between entities, some agreement on information format and transmission procedure should be settled beforehand. This is what is conventionally called a Protocol.

      Large networks nowdays, may host thousands of different devices. To benefit network manager's interoperability and simplicity, any device on the network should carry out most common and important management operations in a well known, unified way. Therefore, an important feature of a network management system would be a Convention on management information naming and presentation.

      Sometimes, management operations should be performed on large number of managed devices. For a network manager to complete such a management round in a reasonably short period of time, an important feature of a network management software would be Performance.

      Some of network devices may run on strictly limited resources what require another property of network management facility: Low resource consumption.

      In practice, the latter requirement translates into low CPU cycles and memory footprint for management software aboard device being managed.

      As networking becomes a more crucial part of our daily lives, security issues have become more apparent. As a side note, even Internet technologies, having military roots, did not pay much attention to security initially. So, the last key feature of network management appears to be Security.

      Data passed back and forth through the course of management operations should be at least authentic and sometimes hidden from possible observers.

      All these problems were approached many times during about three decades of networking history. Some solutions collapsed over time for one reason or another, while others, such as Simple Network Management Protocol (SNMP), evolve into an industry standard.

      1.1 SNMP management architecture

      The SNMP management model includes three distinct entities -- Agent, Manager and Proxy talking to each other over network.

      Agent entity is basically a software running somewhere in a networked device and having the following distinguishing properties:

      • SNMP protocol support
      • Access to managed device's internals

      The latter feature is a source of management information for Agent, as well as a target for remote control operations.

      Modern SNMP standards suggest splitting Agent functionality on two parts. Such Agents may run SNMP for local processes called Subagents, which interface with managed devices internals. Communication between Master Agent and its Subagents is performed using a simplified version of original SNMP protocol, known as AgentX, which is designed to run only within a single system.

      Manager entity is usually an application used by humans (or daemons) for performing various network management tasks, such as device statistics retrieval or remote control.

      Sometimes, Agents and Managers may run peer-to-peer within a single entity that is called Proxy. Proxies can often be seen in application-level firewalling or may serve as SNMP protocol translators between otherwise SNMP version-incompatible Managers and Agents.

      For Manager to request Agent for an operation on a particular part of managed device, some convention on device's components naming is needed. Once some components are identified, Manager and Agent would have to agree upon possible components' states and their semantics.

      SNMP approach to both problems is to represent each component of a device as a named object, similar to named variables seen in programming languages, and state of a component maps to a value associated with this imaginary variable. These are called Managed Objects in SNMP.

      For representing a group of similar components of a device, such as network interfaces, Managed Objects can be organized into a so-called conceptual table.

      Manager talks to Agent by sending it messages of several types. Message type implies certain action to be taken. For example, GET message instructs Agent to report back values of Managed Objects whose names are indicated in message.

      There's also a way for Agent to notify Manager of an event occurred to Agent. This is done through so-called Trap messages. Trap message also carries Managed Objects and possibly Values, but besides that it has an ID of event in form of integer number or a Managed Object.

      For naming Managed Objects, SNMP uses the concept of Object Identifier. As an example of Managed Object, .iso.org.dod.internet.mgmt.mib-2.system.sysName.0 represents human-readable name of a device where Agent is running.

      Managed Objects values are always instances of ASN.1 types (such as Integer) or SNMP-specific subtypes (such as IpAddress). As in programming languages, type has an effect of restricting possible set of states Managed Object may ever enter.

      Whenever SNMP entities talk to each other, they refer to Managed Objects whose semantics (and value type) must be known in advance by both parties. SNMP Agent may be seen as a primary source of information on Managed Objects, as they are implemented by Agent. In this model, Manager should have a map of Managed Objects contained within each Agent to talk to.

      SNMP standard introduces a set of ASN.1 language constructs (such as ASN.1 subtypes and MACROs) which is called Structure of Management Information (SMI). Collections of related Managed Objects described in terms of SMI comprise Management Information Base (MIB) modules.

      Commonly used Managed Objects form core MIBs that become part of SNMP standard. The rest of MIBs are normally created by vendors who build SNMP Agents into their products.

      More often then not, Manager implementations could parse MIB files and use Managed Objects information for names resolution, value type determination, pretty printing and so on. This feature is known as MIB parser support.

      1.2 The history of SNMP

      First SNMP version dates back to 1988 when a set of IETF RFC's were first published ( RFC1065, RFC1066, RFC1067 ). These documents describe protocol operations (in terms of message syntax and semantics), SMI and a few core MIBs. The first version appears to be lightweight and easy to implement. Although, its poor security became notorious over years (Security? Not My Problem!), because cleartext password used for authentication (AKA Community String) is extremely easy to eavesdrop and replay, even after almost 20 years, slightly refined standard ( RFC1155, RFC1157, RFC1212 ) still seems to be the most frequent encounter in modern SNMP devices.

      In effort to fix security issues of SNMPv1 and to make protocol faster for operations on large number of Managed Objects, SNMP Working Group at IETF came up with SNMPv2. This new protocol offers bulk transfers of Managed Objects information (by means of new, GETBULK message payload), improved security and re-worked SMI. But its new party-based security system turned out to be too complicated. In the end, security part of SNMPv2 has been dropped in favor of community-based authentication system used in SNMPv1. The result of this compromise is known as SNMPv2c (where "c" stands for community) and is still widely supported without being a standard ( RFC1902, RFC1903, RFC1904, RFC1905, RFC1906, RFC1907, RFC1908 ).

      The other compromise targeted at offering greater security than SNMPv1, without falling into complexities of SNMPv2, has been attempted by replacing SNMPv2 party-based security system with newly developed user-based security model. This variant of protocol is known as SNMPv2u. Although neither widely implemented nor standardized, User Based Security Model (USM) of SNMPv2u got eventually adopted as one of possibly many SNMPv3 security models.

      As of this writing, SNMPv3 is current standard for SNMP. Although it's based heavily on previous SNMP specifications, SNMPv3 offers many innovations but also brings significant complexity. Additions to version 3 are mostly about protocol operations. SMI part of standard is inherited intact from SNMPv2.

      SNMPv3 system is designed as a framework that consists of a core, known as Message and PDU Dispatcher, and several abstract subsystems: Message Processing Subsystem (MP), responsible for SNMP message handling, Transport Dispatcher, used for carrying over messages, and Security Subsystem, which deals with message authentication and encryption issues. The framework defines subsystems interfaces to let feature-specific modules to be plugged into SNMPv3 core thus forming particular feature-set of SNMP system. Typical use of this modularity feature could be seen in multiprotocol systems -- legacy SNMP protocols are implemented as version-specific MP and security modules. Native SNMPv3 functionality relies upon v3 message processing and User-Based Security modules.

      Besides highly detailed SNMP system specification, SNMPv3 standard also defines a typical set of SNMP applications and their behavior. These applications are Manager, Agent and Proxy ( RFC3411, RFC3412, RFC3413, RFC3414, RFC3415, RFC3416, RFC3417, RFC3418 ).

      2. Programming with PySNMP

      PySNMP stands for a pure-Python SNMP implementation. This software deals with darkest corners of SNMP specifications all in Python programming language.

      This paper is dedicated to PySNMP revisions from 4.1.x and up. Previous PySNMP versions do not follow the architecture and interfaces described in this tutorial.

      From Programmer's point of view, the layout of PySNMP software reflects SNMP protocol evolution. It has been written from ground up, from trivial SNMPv1 up to fully featured SNMPv3. Therefore, several levels of API to SNMP functionality are available:

      • The most ancient and low-level is SNMPv1/v2c protocol scope. Here programmer is supposed to build/parse SNMP messages and their payload -- Protocol Data Unit (PDU), handle protocol-level errors, transport issues and so on.

        Although considered rather complex to deal with, this API probably gives best performance, memory footprint and flexibility, unless MIB access and/or SNMPv3 support is needed.

      • Parts of SNMPv3 standard is expressed in terms of some abstract API to SNMP engine and its components. PySNMP implementation adopts this abstract API to a great extent, so it's available at Programmer's disposal. As a side effect, SNMP RFCs could be referenced for API semantics when programming PySNMP at this level.

        This API is much more higher-level than previous; here Programmer would have to manage two major issues: setting up Local Configuration Datastore (LCD) of SNMP engine and build/parse PDUs. PySNMP system is shipped multi-lingual, thus at this level all SNMPv1, SNMPv2c and SNMPv3 features are available.

      • At last, the highest-level API to SNMP functionality is available through the use of standard SNMPv3 applications. These applications cover the most frequent needs. That's why this API is expected to be the first to start with.

        The Applications API further simplifies Programmer's job by hiding LCD management issues (contrary to SNMPv3 engine level). This API could be exploited in a one-liner fashion, for quick and simple prototyping.

      The following figure draws major components of PySNMP system along with standard Applications.

      PySNMP architecture One-line Applications

      These standard SNMP applications, such as GET/SET command generators and responders or TRAP notificators and receivers, translate into a set of classes designed by the Visitor pattern. Application classes implement concrete SNMP operations in terms of specific PDU handling, while SNMP Engine class acts as a Visitor. A single SNMP Engine can serve many Applications of different types at the same time.

      One of the design choices of SNMPv3 standard is to use a set of dedicated Managed Objects for SNMP engine internal purposes. One reason for that design involves making SNMP engine remotely configurable via SNMP. These internally used Managed Objects are collectively called Local Configuration Datastore (LCD). In PySNMP, all SNMP engine configuration and statistics is kept in LCD. LCD Configurator is a wrapper aimed at simplifying LCD operations. Technically, LCD Configurator is a set of functions whose names clearly reflect their semantics.

      SNMP Engine, on the above figure, is a Composite class holding references to all components of SNMP system. Typical user application has a single instance of SNMP Engine class possibly shared by many SNMP Applications of all kinds.

      Transport subsystem is used for sending SNMP messages to and accepting them from network. The I/O subsystem consists of a an abstract Dispatcher and one or more abstract Transport classes. Concrete Dispatcher implementation is I/O method-specific, consider BSD sockets for example. Concrete Transport classes are transport domain-specific. SNMP frequently uses UDP Transport but others are also possible. Dispatcher/Transport classes are designed after the Visitor pattern -- Transport instances are Dispatcher visitors. Transport Dispatcher interfaces are mostly used by Message And PDU Dispatcher. However, when using the SNMPv1/v2c-native API (the lowest-level one), these interfaces would be invoked directly.

      The rest of components are not normally accessed directly. They're mentioned here for clarification purposes.

      Message And PDU Dispatcher is a heart of SNMP system. Its main responsibilities include dispatching PDUs from SNMP Applications through various subsystems all the way down to Transport Dispatcher, and passing SNMP messages coming from network up to SNMP Applications. It maintains logical connection with Management Instrumentation Controller which carries out operations on Managed Objects, here for the purpose of LCD access.

      Message Processing Modules handle message-level protocol operations for present and possibly future versions of SNMP protocol. Most importantly, these include message parsing/building and possibly invoking security services whenever required. All MP Modules share standard API used by Message And PDU Dispatcher.

      Message Security Modules perform message authentication and/or encryption. As of this writing, User-Based (for v3) and Community (for v1/2c) modules are implemented in PySNMP. All Security Modules share standard API used by Message Processing subsystem.

      Access Control subsystem uses LCD information to authorize remote access to Managed Objects. This is used when serving Agent Applications or Trap receiver in Manager Applications.

      2.1 One-line Applications

      As of this writing, one-line Applications currently cover Manager-side operations. Agent and Proxy roles could be implemented on top of native Applications API.

      There're two kinds of APIs to one-line Applications: synchronous and asynchronous. Both are implemented within the pysnmp.entity.rfc3413.oneliner.cmdgen module.

      2.1.1 Synchronous One-line Applications

      This is the simplest and the most high-level API to standard SNMP Applications. It's advised to employ for singular and blocking operations as well as for rapid prototyping.

      All Command Generator Applications are implemented within a single class:

      class CommandGenerator([snmpEngine])

      Create a SNMP Command Generator object.

      Methods of the CommandGenerator class instances implement specific request types.

      getCmd( authData, transportTarget, *varNames )

      Perform SNMP GET request and return a response or error indication.

      The authData is a SNMP Security Parameters object, transportTarget is a SNMP Transport Configuration object and *varNames are Managed Objects names (ASN.1 OIDs).

      The getCmd method returns a tuple of errorIndication, errorStatus, errorIndex, varBinds.

      Non-empty errorIndication string indicates SNMP engine-level error.

      The pair of errorStatus and errorIndex variables determines SNMP PDU-level error. If errorStatus evaluates to true, this indicates SNMP PDU error caused by Managed Object at position errorIndex-1 in varBinds.

      The varBinds is a tuple of Managed Objects. Managed Objects found in response are position-bound to Managed Object names passed in request. Each Managed Object is a tuple of Object Name and Object Value.

      The following code performs SNMP GET operation over SNMPv1: 

      >>> from pysnmp.entity.rfc3413.oneliner import cmdgen
>>> errorIndication, errorStatus, errorIndex, varBinds = cmdgen.CommandGenerator().getCmd(
... cmdgen.CommunityData('my-agent', 'public', 0),
... cmdgen.UdpTransportTarget(('localhost', 161)),
... (1,3,6,1,2,1,1,1,0)
... )
>>> print errorIndication
None
>>> print errorStatus
0
>>> print varBinds
[(ObjectName('1.3.6.1.2.1.1.1.0'), OctetString("'Linux my.domain.com 2.6.21 #2 Mon Mar 19 17:07:18 MSD 2006 i686'"))]
      setCmd( authData, transportTarget, *varBinds )

      Perform SNMP SET request and return a response or error indication.

      The authData and transportTarget parameters have he same semantics as in getCmd method.

      The *varBinds input parameter is a tuple of Managed Objects to be applied at Agent. The syntax of *varBinds is the same as in getCmd.

      The setCmd method returns a tuple of errorIndication, errorStatus, errorIndex, varBinds.

      The errorIndication, errorStatus and errorIndex parameters have the same meaning as in getCmd method.

      The following code performs SNMP SET operation over SNMPv2c: 

      >>> from pysnmp.entity.rfc3413.oneliner import cmdgen
>>> from pysnmp.proto import rfc1902
>>> errorIndication, errorStatus, errorIndex, varBinds = cmdgen.CommandGenerator().setCmd(
... cmdgen.CommunityData('my-agent', 'public', 1),
... cmdgen.UdpTransportTarget(('localhost', 161)),
... ((1,3,6,1,2,1,1,1,0), rfc1902.OctetString('my system description'))
... )
>>> print errorIndication
None
>>> print errorStatus
17
>>> print errorStatus.prettyPrint()
notWritable(17)
      nextCmd( authData, transportTarget, *varNames )

      Perform SNMP GETNEXT request and return a response or error indication. The GETNEXT request type implies referring to Managed Objects whose Object Names are next to those used in request.

      Input parameters to the nextCmd method are the same as to getCmd.

      The nextCmd method returns a tuple of errorIndication, errorStatus, errorIndex, varBindTable.

      The errorIndication, errorStatus and errorIndex parameters have the same meaning as in getCmd method.

      The varBindTable parameter is a tuple of varBinds. Each varBind of varBinds in varBindTable represent a set of Managed Objects whose Object Names reside inside OID sub-tree of Managed Object name passed in request. In other words, with this oneliner API, an invocation of nextCmd method for a single Managed Object might return a sequence of Managed Objects so that Object Name passed in request would be a prefix for Object Names returned in response (as a side note, the same method in Applications API would return varBinds as held in a single response, and regardless of the prefix property). Properties of the varBinds parameter is the same as in getCmd method.

      The following code performs SNMP GETNEXT operation against a MIB subtree over SNMPv3: 

      >>> from pysnmp.entity.rfc3413.oneliner import cmdgen
>>> errorIndication, errorStatus, errorIndex, varBinds = cmdgen.CommandGenerator().nextCmd(
... cmdgen.UsmUserData('my-user', 'my-authkey', 'my-privkey'),
... cmdgen.UdpTransportTarget(('localhost', 161)),
... (1,3,6,1,2,1,1)
... )
>>> print errorIndication
None
>>> print errorStatus
0
>>> for varBindTableRow in varBindTable:
...     print varBindTableRow
...
[(ObjectName('1.3.6.1.2.1.1.1.0'), OctetString("'Linux my.domain.com 2.6.21 #2 Mon Mar 19 17:07:18 MSD 2006 i686'"))]
[(ObjectName('1.3.6.1.2.1.1.2.0'), ObjectIdentifier('1.3.6.1.4.1.8072.3.2.10'))]
[ skipped ]
[(ObjectName('1.3.6.1.2.1.1.9.1.4.9'), TimeTicks('17'))]
>>>
      bulkCmd( authData, transportTarget, nonRepeaters, maxRepetitions, *varNames )

      Perform SNMP GETBULK request and return a response or error indication. The GETBULK request type has the same semantics as GETNEXT one except that the latter queries a bulk of Managed Objects at once.

      The authData, transportTarget, *varNames input parameters to the bulkCmd method are the same as to nextCmd.

      The nonRepeaters parameter indicates how many of *varNames passed in request should be queried for a single instance with in a request.

      The maxRepetitions parameter indicates for how many instances of Managed Objects in the rest of *varNames, besides first nonRepeaters ones, should be queried with single request.

      The bulkCmd method returns a tuple of errorIndication, errorStatus, errorIndex, varBindTable.

      The errorIndication, errorStatus, errorIndex and varBindTable parameters have the same meaning as in getCmd method.

      The following code performs SNMP GETBULK operation against a MIB subtree over SNMPv3: 

      >>> from pysnmp.entity.rfc3413.oneliner import cmdgen
>>> errorIndication, errorStatus, errorIndex, varBinds = cmdgen.CommandGenerator().bulkCmd(
... cmdgen.UsmUserData('my-user', 'my-authkey', 'my-privkey'),
... cmdgen.UdpTransportTarget(('localhost', 161)),
... 0, 25, # nonRepeaters, maxRepetitions
... (1,3,6,1,2,1,1)
... )
>>> print errorIndication
None
>>> print errorStatus
0
>>> for varBindTableRow in varBindTable:
...     print varBindTableRow
...
[(ObjectName('1.3.6.1.2.1.1.1.0'), OctetString("'Linux my.domain.com 2.6.21 #2 Mon Mar 19 17:07:18 MSD 2006 i686'"))]
[(ObjectName('1.3.6.1.2.1.1.2.0'), ObjectIdentifier('1.3.6.1.4.1.8072.3.2.10'))]
[ skipped ]
[(ObjectName('1.3.6.1.2.1.1.9.1.4.9'), TimeTicks('17'))]
>>>

      Notification Originator Applications are implemented within a single class:

      class NotificationOriginator([snmpContext])

      Create a SNMP Notification Originator object.

      The following method of NotificationOriginator class instance implements specific notifications types.

      sendNotification( authData, transportTarget, notifyType, notificationType, *varBinds )

      Send either unconfirmed (TRAP) or confirmed (INFORM) SNMP notification and possibly return an error indication.

      The authData and transportTarget parameters have the same semantics as in CommandGenerator.getCmd method.

      The notifyType parameter determines the type of notification to be generated. Supported values include "trap" for unconfirmed notification or "inform" for a confirmed one.

      Be advised, that when using confirmed notification, Notification Receiver must know ContextEngineID of Notification Originator to be able to process and acknowledge confirmed notification.

      The notificationType parameter indicates the kind of event to notify Manager about in form of SMI NOTIFICATION-TYPE object name. For instance, ('SNMPv2-MIB', 'coldStart') is a value of coldStart notification type as defined in SNMPv2-MIB module.

      The *varBinds input parameter is a tuple of Managed Objects to be passed over to Manager along with Notification. The syntax of *varBinds is the same as in CommandGenerator.getCmd.

      The sendNotification method returns an errorIndication parameter which has the same meaning as in CommandGenerator.getCmd.

      The following code sends SNMP TRAP over SNMPv3: 

      >>> from pysnmp.entity.rfc3413.oneliner import cmdgen, ntforg
>>> from pysnmp.proto.api import v2c
>>> errorIndication = ntforg.NotificationOriginator().sendNotification(
... cmdgen.UsmUserData('my-user', 'my-authkey', 'my-privkey'),
... cmdgen.UdpTransportTarget(('localhost', 162)),
... 'trap',
... ('SNMPv2-MIB', 'coldStart'),
... ((1,3,6,1,2,1,1,3,0), v2c.TimeTicks(44100))
)
>>> print errorIndication
None
>>> print errorStatus
0

      2.1.2 Asynchronous One-line Applications

      Asynchronous API to one-line Applications is actually a foundation for Synchronous version, so they're very similar. This Asynchronous API is useful for purposes such as running multiple, possibly different, SNMP Applications at the same time or handling other activities inside user's program while SNMP Application is waiting for input/output.

      All Command Generator Applications are implemented within a single class:

      class AsynCommandGenerator([snmpEngine])

      Create an asynchronous SNMP Command Generator object.

      Methods of the AsynCommandGenerator class instances implement specific request types. These methods are similar to those described in the CommandGenerator class section except that asynchronous interface uses a callback function for delivering responses.

      asyncGetCmd( authData, transportTarget, varNames, (cbFun, cbCtx) )

      Prepare SNMP GET request to be dispatched. Return the sendRequestHandle value.

      The cbFun parameter is a reference to a callable object (such as Python function) that takes the following parameters:

      cbFun( sendRequestHandle, errorIndication, errorStatus, errorIndex, varBinds, cbCtx )

      Where sendRequestHandle is an integer value used for matching response to request. Its counterpart is returned on request submission by the asyncGetCmd method.

      The cbCtx parameter is a reference to the cbCtx object being passed to asyncGetCmd method. Its purpose is to carry opaque application's state from request through response methods.

      The errorIndication, errorStatus, errorIndex and varBinds parameters have the same meaning as in CommandGenerator.getCmd method.

      If cbFun has no more requests pending and want to complete, it must return a true value. Otherwise, it returns false.

      The authData, transportTarget and varNames parameters have the same meaning as in CommandGenerator.getCmd method.

      The asyncGetCmd method returns unique sendRequestHandle integer value used for matching subsequent response to this request.

      asyncSetCmd( authData, transportTarget, varBinds, (cbFun, cbCtx) )

      Prepare SNMP SET request to be dispatched. Return the sendRequestHandle value.

      The authData and transportTarget parameters have the same meaning as in CommandGenerator.setCmd method.

      The cbFun and cbCtx parameters have the same meaning as in AsynCommandGenerator.asyncGetCmd method.

      The varBinds parameter has the same meaning as in CommandGenerator.setCmd method except that here it is passed in as a tuple.

      asyncNextCmd( authData, transportTarget, varNames, (cbFun, cbCtx) )

      Prepare SNMP GETNEXT request to be dispatched. Return the sendRequestHandle value.

      The authData and transportTarget parameters have the same meaning as in CommandGenerator.nextCmd method.

      The cbFun and cbCtx parameters have the same meaning as in AsynCommandGenerator.asyncGetCmd method.

      The varNames parameter has the same meaning as in CommandGenerator.nextCmd method except that here it is passed in as a tuple.

      asyncBulkCmd( authData, transportTarget, nonRepeaters, maxRepetitions, varNames, (cbFun, cbCtx) )

      Prepare SNMP GETBULK request to be dispatched. Return the sendRequestHandle value.

      The authData, transportTarget, nonRepeaters and maxRepetitions parameters have the same meaning as in CommandGenerator.nextCmd method.

      The cbFun and cbCtx parameters have the same meaning as in AsynCommandGenerator.asyncGetCmd method.

      The varNames parameter has the same meaning as in CommandGenerator.bulkCmd method except that here it is passed in as a tuple.

      After one or more requests have been submitted by calling one or more of the methods above, Transport Dispatcher must be invoked to get SNMP engine running. This is done by calling:

      asynCommandGenerator.snmpEngine.transportDispatcher.runDispatcher ()

      Where asynCommandGenerator is AsynCommandGenerator class instance.

      The runDispatcher() method terminates when no pending requests left for running Applications.

      The following code performs SNMP GET operation asynchronously through SNMPv3: 

      >>> from pysnmp.entity.rfc3413.oneliner import cmdgen
>>>
>>> def cbFun(sendRequestHandle, errorIndication, errorStatus, errorIndex, varBinds, cbCtx):
...     print 'sendRequestHandle =', sendRequestHandle
...     print 'errorIndication =', errorIndication
...     print 'errorStatus =', errorStatus
...     print 'varBinds =', varBinds
...     print 'cbCtx =', cbCtx
...
>>> asynCommandGenerator = cmdgen.AsynCommandGenerator()
>>> # This is a non-blocking call
>>> sendRequestHandle = asynCommandGenerator.asyncGetCmd(
... cmdgen.UsmUserData('my-user', 'my-authkey', 'my-privkey'), 
... cmdgen.UdpTransportTarget(('localhost', 161)), 
... ((1,3,6,1,2,1,1,1,0),), 
... (cbFun, None))
>>> print sendRequestHandle
1
>>> asynCommandGenerator.snmpEngine.transportDispatcher.runDispatcher()
sendRequestHandle = 1
errorIndication = None
errorStatus = 0
varBinds = [(ObjectName('1.3.6.1.2.1.1.1.0'), OctetString("'Linux my.domain.com 2.6.21 #2 Mon Mar 19 17:07:18 MSD 2006 i686'"))]
cbCtx = None
>>>

      The AsynNotificationOriginator class implements specific notification types.

      class AsynNotificationOriginator([snmpContext])

      Create an asynchronous SNMP Notification Originator object.

      The only method of AsynNotificationOriginator class is similar to that described in the NotificationOriginator class section except that asynchronous interface uses a callback function for delivery confirmation when confirmed notification are used.

      asyncSendNotification( authData, transportTarget, notifyType, notificationType, varBinds, (cbFun, cbCtx) )

      Prepare SNMP TRAP or INFORM notification to be dispatched. Return the sendRequestHandle value.

      The cbFun parameter is a reference to a callable object (such as Python function) that takes the following parameters:

      cbFun( sendRequestHandle, errorIndication, cbCtx )

      Where the sendRequestHandle, errorIndication and cbCtx parameters have the same meaning as in callback function in AsynCommandGenerator.asynGetCmd method.

      The cbCtx parameter has the same meaning as in AsynCommandGenerator.asyncGetCmd method.

      The notifyType, notificationType and varBinds parameters have the same meaning as in NotificationOriginator.sendNotification method except that here it is passed in as a tuple.

      The asyncSendNotification method returns unique sendRequestHandle integer value used for matching subsequent delivery confirmation response to arbitrary notification.

      After one or more notifications have been submitted by calling the sendNotification method, Transport Dispatcher must be invoked to get SNMP engine running. This is done by calling:

      asynNotificationOriginator.snmpEngine.transportDispatcher.runDispatcher ()

      Where asynNotificationOriginator is AsynNotificationOriginator class instance.

      The runDispatcher() method terminates when no unconfirmed notifications left for running Applications.

      The following code sends SNMP INFORM notification asynchronously through SNMPv3: 

      >>> from pysnmp.entity.rfc3413.oneliner import cmdgen, ntforg
>>> from pysnmp.proto.api import v2c
>>>
>>> def cbFun(sendRequestHandle, errorIndication, cbCtx):
...     print 'sendRequestHandle =', sendRequestHandle
...     print 'errorIndication =', errorIndication
...     print 'cbCtx =', cbCtx
...
>>> asynNotificationOriginator = ntforg.AsynNotificationOriginator()
>>> # This is a non-blocking call
>>> sendRequestHandle = asynNotificationOriginator.asyncSendNotification(
... cmdgen.UsmUserData('my-user', 'my-authkey', 'my-privkey'),
... cmdgen.UdpTransportTarget(('localhost', 162)),
... 'inform',
... ('SNMPv2-MIB', 'coldStart'),
... ((1,3,6,1,2,1,1,1,0), v2c.TimeTicks(44100)),
... (cbFun, None))
>>> print sendRequestHandle
1
>>> asynNotificationOriginator.snmpEngine.transportDispatcher.runDispatcher()
sendRequestHandle = 1
errorIndication = None
cbCtx = None
>>>

      2.1.3 Security configuration

      Calls to one-line Applications API require Security Parameters and Transport configuration objects as input parameters. These classes serve as convenience shortcuts to SNMP engine configuration facilities and for keeping persistent authentication/transport configuration between SNMP engine calls.

      Security Parameters object is Security Model specific. UsmUserData class serves SNMPv3 User-Based Security Model configuration, while CommunityData class is used for Community-Based Security Model of SNMPv1/SNMPv2c.

      class UsmUserData( securityName, authKey='', privKey='', authProtocol=usmNoAuthProtocol, privProtocol=usmNoPrivProtocol )

      Create an object holding User-Based Security Model specific configuration parameters.

      Mandatory securityName parameter is SNMPv3 USM username passed in as a string.

      Optional authKey parameter is a secret key used within USM for SNMP PDU authorization. It's string type. Setting it to a non-empty value implies MD5-based PDU authentication to take effect. Default hashing method may be changed by means of further authProtocol parameter.

      Optional privKey parameter is a secret key used within USM for SNMP PDU encryption. It's string type. Setting it to a non-empty value implies MD5-based PDU authentication and DES-based encryption to take effect. Default hashing and/or encryption methods may be changed by means of further authProtocol and/or privProtocol parameters.

      Optional authProtocol parameter may be used to specify non-default hash function algorithm. Possible values include usmHMACMD5AuthProtocol, usmHMACSHAAuthProtocol and usmNoAuthProtocol. These symbols are defined in pysnmp.entity.rfc3413.oneliner.cmdgen module.

      Optional privProtocol parameter may be used to specify non-default ciphering algorithm. Possible values include usmDESPrivProtocol and usmNoPrivProtocol.

      class CommunityData( securityName, communityName, mpModel=1 )

      Create an object holding Community-Based Security Model specific configuration parameters.

      Mandatory securityName parameter is Community-Based Security Model username passed in as a string. For most purposes this can be an arbitrary string.

      Mandatory communityName parameter is SNMPv1/SNMPv2c Community name passed as a string.

      Optional mpModel parameter indicates whether SNMPv2c (mpModel=1, default) or SNMPv1 (mpModel=0) protocol should be used.

      2.1.4 Transport configuration

      Transport configuration object is Transport domain specific. UdpTransportTarget class represents an Agent accessible through UDP domain transport.

      class UdpTransportTarget( transportAddr, timeout=1.5, retries=3 )

      Create an object representing a single Agent accessible through UDP socket.

      Mandatory transportAddr parameter indicates destination Agent address in form of tuple of FQDN, port where FQDN is a string and port is an integer.

      Optional timeout and retries parameters may be used to modify default response timeout (1.5 seconds) and number of succesive request retries (3 times).

      2.2 Managed Objects names and values

      In PySNMP programming context, Managed Object term (also called Variable-Binding in protocol specifications) refers to a tuple of Managed Object (or Managed Object Instance) Name and Managed Object Instance Value.

      Managed Objects Names and Values are ASN.1 types. Both Names and Values are derived from PyASN1 classes. These are defined in pysnmp.proto.rfc1902 module.

      Managed Object Name is a tuple or tuple-like Object Identifier class instance.

      class ObjectIdentifier( objectIdentifier )

      Create an ASN.1 Object Identifier object. The objectIdentifier parameter represents Object Identifier value. It should be either a tuple or tuple-like object or a string representing Object Identifier in dotted notation (like "1.3.6.1").

      Instances of this class mimic basic properties of Python tuple. Sub-OIDs of an OID translate into tuple components.

      For more information on ObjectIdentifier class properties, refer to PyASN1 documentation.

      It's PySNMP the design decision to always use SMIv2 definitions for Managed Objects regardless of SNMP protocol version being used.

      Managed Object Instance Value is an instance of some PyASN1 class or its SNMP-specific derivative. The latter case reflects SNMP-specific ASN.1 sub-type. The list of Managed Object Instance Value classes follows.

      class Integer( value )

      Create a SMIv2 Integer object. The value parameter should be an integer value. Instances of this class mimic basic properties of a Python integer.

      class Integer32( value )

      Create a SMIv2 Integer32 object. This object is similar to Integer class instance.

      class OctetString( value )

      Create a SMIv2 OctetString object. The value parameter should be a string value. Instances of this class mimic basic properties of a Python string.

      class IpAddress( value )

      Create a SMIv2 IpAddress object. The value parameter should be an IP address expressed in quad-dotted notation (e.g. "127.0.0.1").

      class Counter32( value )

      Create a SMIv2 Counter32 object. Besides different value constraints, this object is similar to Integer class instance.

      class Gauge32( value )

      Create a SMIv2 Gauge32 object. Besides different value constraints, this object is similar to Integer class instance.

      class Unsigned32( value )

      Create a SMIv2 Unsigned32 object. Besides different value constraints, this object is similar to Integer class instance.

      class TimeTicks( value )

      Create a SMIv2 TimeTicks object. Besides different value constraints, this object is similar to Integer class instance.

      class Opaque( value )

      Create a SMIv2 Opaque object. This object is similar to OctetString class instance.

      class Counter64( value )

      Create a SMIv2 Counter64 object. Besides different value constraints, this object is similar to Integer class instance.

      class Bits( value )

      Create a SMIv2 Bits object. The value parameter should be sequence of names of bits raised to one. Unmentioned bits default to zero.

      For more information on SNMP Managed Value objects properties, refer to their base classes in PyASN1 documentation.

      2.3 MIB services

      PySNMP supports both Manager and Agent-side operations on Managed Objects, including MIB lookup and custom Managed Objects implementation.

      Managed Objects, implemented in Python code, is the basis for PySNMP MIB services. Managed Objects are collected into a pool and then managed by a MIB builder. Both Manager and Agent applications deal with their Managed Objects through role-specific MIB view and MIB instrumentation. The same set of Managed Objects could serve both Manager and Agent purposes within a single SNMP entity.

      2.3.1 Data model for Managed Objects

      In PySNMP, Managed Objects take shape of Python class instances that implement various SMIv2 items. Collections of Managed Objects, or MIBs, translate, in a one-to-one fashion, into Python modules.

      Automated conversion of MIB text files into Python modules can be done through the use of smidump tool of libsmi package and "build-pysnmp-mib" script shipped with PySNMP.

      The pysnmp.smi.mibs.SNMPv2-SMI module implements the following classes:

      class MibScalar( name, syntax )

      Create a definition of scalar Managed Object with name name and associated value of type syntax.

      The name parameter represents an Object Identifier which can be expressed as either a tuple of integers or tuple-like Object Identifier class instance.

      The syntax parameter represents Managed Object's value type.

      The MibScalar class implements the following methods:

      getName()
      getSyntax()
      getMaxAccess()
      getUnits()
      getStatus()
      getDescription()

      Each of these methods return certain property of Managed Object.

      class MibScalarInstance( name, syntax )

      Create an instance of scalar Managed Object or Conceptual Table element with name name and associated value syntax.

      The name of Managed Object instance is a concatination of name of Managed Object definition and some instance identifier. For scalar types, instance identifier is a single zero (0,). For Conceptual Table elements instance identifier is a concatination of table indices.

      The name and syntax parameters have the same meaning as in MibScalar class.

      class MibTableColumn( name, syntax )

      Create a definition of Conceptual Table Column with name name and associated value of type syntax.

      The name parameter has the same meaning as in MibScalar class.

      The syntax parameter represents type of the value associated with columnar Managed Object.

      The MibTableColumn class implements the following methods:

      setProtoInstance( instanceClass )

      Configure MibTableColumn object to instantiate instanceClass when creating Columnar Objects. By default, MibScalarInstance is instantiated.

      class MibTableRow( name )

      Create a definition of Conceptual Table Row with name name.

      The name parameter has the same meaning as in MibScalar class.

      The MibTableRow class implements the following methods:

      getInstIdFromIndices( *indices )

      Compute and return Conceptual Table Column instance identifier from *indices using MIB Table Index definition.

      Types of *indices must coerce into Table Index syntax.

      getIndicesFromInstId( instanceId )

      Compute and return a tuple of Conceptual Table Index values from Column instance identifier instanceId using MIB Table Index definition.

      The number of types of returned index values depend on MIB Table definition.

      class MibTable( name )

      Create a definition of Conceptual Table with name name.

      The name parameter has the same meaning as in MibScalar class.

      The following examples explain how MIB text could be expressed in terms of PySNMP SMI data model. First example is on a scalar: 

      myManagedObject = MibScalar((1, 3, 6, 1, 4, 1, 20408, 2, 1), OctetString()).setMaxAccess("readonly")

      Managed Object Instance can be put into a stand-alone PySNMP SMI module or be implemented inside Agent application. Managed Object Instance will be associated with its parent Managed Object, by the MIB building part of PySNMP, on the basis of their names relation. 

      myManagedObjectInstance = MibScalarInstance(myManagedObject.getName() + (0,), myManagedObject.getSyntax().clone('my string'))

      Let's consider SNMP Conceptual Table created in an "MY-MIB.py" file: 

      myTable = MibTable((1, 3, 6, 1, 4, 1, 20408, 2, 1))
myTableEntry = MibTableRow(myTable.getName() + (1,)).setIndexNames((0, "MY-MIB", "myTableIndex"))
myTableIndex = MibTableColumn(myTableEntry.getName() + (1,), Integer())
myTableValue = MibTableColumn(myTableEntry.getName() + (2,), OctetString())

      Populate Managed Objects table with Managed Objects Instance in the first column. 

      myTableValueInstance = MibScalarInstance(myTableValue.getName() + (1,), myTableValue.getSyntax().clone('my value'))

      For more real-life cases, refer to modules in pysnmp.smi.mibs sub-package.

      2.3.2 MIB builder

      The pythonized MIB modules are then managed by the MibBuilder class from pysnmp.smi.builder module.

      class MibBuilder()

      Create MIB modules loader/evaluator/indexer.

      loadModules( *modNames )

      Locate in search path and evaluate each of *modNames through Python execfile() passing a reference to MibBuilder class instance to module's global scope. Evaluating modules might register their objects at MibBuilder through exportSymbols() call.

      MIB builder would then create an in-memory index of registered MIB objects by MIB names.

      Search path is managed by the getMibPath() and setMibPath() methods.

      The loadModules method may be further invoked recursively on dependent MIB modules import.

      unloadModules( *modNames )

      Drop all references to Python objects previously created through calling loadModules() method against [here optional] *modNames. This method would invoke unexportSymbols() against MIB symbols previously registered under each of *modNames.

      Missing *modNames implies all currently loaded modules.

      importSymbols( modName, *symNames )

      Return a tuple of Managed Objects looked up by their MIB names *symNames. Managed Objects returned in tuple are position-bound to *symNames parameters.

      If MIB module modName is not yet loaded, the importSymbols() method would be invoked implicitly.

      exportSymbols( modName, *anonymousSyms, **namedSyms )

      Register Managed Objects *anonymousSyms and/or **namedSyms at MibBuilder within MIB module modName scope.

      Managed Objects defined in MIB are always named. These are exported using **namedSyms parameter(s). Managed Objects Instances don't have to have MIB names, unless Application wants to access Managed Objects Instances by MIB name, so these may be exported through *anonymousSyms.

      unexportSymbols( modName, *symNames )

      Drop all references to Python objects previously registered under *symNames within modName through exportSymbols() call.

      Missing *symNames implies all symbols currently registered within modName module.

      In the following example MIB builder will be created, MIB modules loaded up and Managed Object definition looked up by symbolic name: 

      >>> from pysnmp.smi import builder
>>>
>>> # create MIB builder
... mibBuilder = builder.MibBuilder().loadModules('SNMPv2-MIB', 'IF-MIB')
>>>
>>> # get Managed Object definition by symbol name
... mibNode, = mibBuilder.importSymbols('SNMPv2-MIB', 'sysDescr')
>>> print mibNode.getName()
(1, 3, 6, 1, 2, 1, 1, 1)
>>> print repr(mibNode.getSyntax())
DisplayString('')
>>>

      2.3.3 MIB view controller

      The following facilities are intended for Manager-side access to MIB definitions. The pysnmp.smi.view module contains the following items:

      class MibViewController(mibBuilder)

      The MibViewController class instance tackles Managed Objects, constructed by MibBuilder, for their properties and provide efficient/ordered access to Managed Objects properties. Most important of these are OID names and labels.

      The mibBuilder argument is an instance of MibBuilder class.

      The MibViewController class implements the following methods:

      getNodeName(name)

      The name parameter is Managed Object name. It can be either a tuple representing sub-OIDs or Object Identifier class instance. Sub-OIDs can be a mix of integers and string labels. For example, the following are valid values of name:

      • (1, 3, 6, 1)
      • ('iso', 'org', 'dod', 'internet')
      • ('iso', 2, 'dod', 1)
      • pysnmp.proto.rfc1902.ObjectIdentifier("1.3.6.1")

      The getNodeName method returns a tuple of (oid, label, suffix) where:

      • The oid and label are tuples of sub-OIDs of best (longest) matched Managed Object in integer and label forms respectively.
      • The suffix parameter is the unmatched, trailing part of original name parameter.

        If a Managed Object is looked up with getNodeName method and an exact match occured, suffix would be an empty tuple.

        If suffix is not empty, it indicates either an index part of Conceptual Table instance name (which can be further parsed into index values by MibTableRow class methods) or a partial Managed Object name match.

        In order to distinguish MIB Table element match from a failure, see if closest matched Managed Object oid (MIB symbol label[-1]) is an instance of MibTableColumn class.

        If even partial match fails, the SmiError exception is raised. 

    >>> from pysnmp.smi import builder, view
>>>
>>> mibBuilder = builder.MibBuilder().loadModules('SNMPv2-MIB')
>>> mibViewController = view.MibViewController(mibBuilder)
>>> 
>>> oid, label, suffix = mibViewController.getNodeName((1,3,6,1,2,'mib-2',1,'sysDescr'))
>>> print oid
(1, 3, 6, 1, 2, 1, 1, 1)
>>> print label
('iso', 'org', 'dod', 'internet', 'mgmt', 'mib-2', 'system', 'sysDescr')
>>> print suffix
()

    getNextNodeName( name, modName='' )

    The getNextNodeName method works the same as getNodeName but it deals with Managed Object whose name appears to be next to the name given on input.

    The modName parameter is MIB module name as seen by MibBuilder. Use this parameter to restrict by-name to particular MIB module's scope.

    getFirstNodeName(modName='')

    The getFirstNodeName method works the same as getNodeName but it returns Managed Object whose name appears to be the first among others within MIB module modName.

    If no modName is given, the whole OID namespace is assumed.

    getNodeLocation(name)

    The getNodeLocation method returns MIB location of Managed Object by OID name as a tuple of (modName, mibName, suffix).

    The modName and mibName parameters are as used in MibBuilder interface. The suffix parameter is as described in getNodeName() method. 

    >>> from pysnmp.smi import builder, view
>>>
>>> mibBuilder = builder.MibBuilder().loadModules('SNMPv2-MIB')
>>> mibViewController = view.MibViewController(mibBuilder)
>>> 
>>> modName, symName, suffix = mibViewController.getNodeLocation((1,3,6,1,2,1,1,1,123))
>>> print modName
SNMPv2-MIB
>>> print symName
sysDescr
>>> print suffix
(123,)

    2.3.4 Implementing Managed Objects Instances

    The following chapter explains SNMP Agent-controlled Managed Object Instances to real-life objects mapping.

    SNMP defines four types of operations on Managed Objects Instances. For scalars, these are:

    • Get Managed Object Instance value (though SNMP GET request)
    • Modify Managed Object Instance value (though SNMP SET request)

    Conceptual Tables additionaly support:

    • Table row creation (through SNMP SET against a special-purpose RowStatus column instance)
    • Table row removal (similary, through SNMP SET against RowStatus column instance)

    PySNMP Managed Objects Instances are implemented by the MibScalarInstance objects while a value associated with Managed Object Instance is represented by its syntax initialization parameter.

    There are two distinct approaches to Managed Objects Instances implementation in PySNMP. The first one is simpler to use but it only works for relatively static Managed Objects. The other is universal but it is more complex to deal with.

    2.3.4.1 Associated value gatewaying

    This method only works for scalars and static tables (meaning no row creation and deletion is performed through SNMP). Also, it is not safe with this method to modify dependent values though a single request as failed modification won't roll back others in the bulk.

    Whenever SNMP Agent receives read or modification request against arbitrary Managed Object Instance, it ends up clone()'ing syntax parameter of MibScalarInstance object. Read queries (e.g. GET/GETNEXT/GETBULK) trigger clone method invocation without passing it new value, while new value will be fed to the clone method on modification request.

    This value-based gatewaying method works by listening on the clone() method of MibScalarInstance associated value thus fetching current or applying new state of some outer system represented by arbitrary Managed Object Instance.

    Consider SMI-to-filesystem gateway for example, where a Managed Object Instance would represent particular file contents. File contents would be solely dependent on SNMP updates. 

    class MyFile(OctetString):
  def clone(self, value=None):
    if value is not None:
      # SNMP SET
      open('/tmp/myfile', 'w').write(value)

    # SNMP S/GET*
    return OctetString.clone(self, open('/tmp/myfile', 'r').read())

mibBuilder.exportSymbols(
  'MYFILE-MIB', MibScalarInstance((1, 3, 6, 1, 4, 1, 20408, 1), MyFile())
)

    A variation of this through-value SMI gatewaying method would be for a third-party system to keep Managed Object Instance value synchronized with system's current state. Take file size monitor for instance -- the following code would be run periodically to measure most recent file size and re-build its SMI projection: 

    myManagedObjectInstance = MibScalarInstance(
  (1, 3, 6, 1, 4, 1, 20408, 1), Integer(os.stat('/var/adm/messages')[6])
)

mibBuilder.exportSymbols(
  'FILESIZE-MIB', myManagedObjectInstance=myManagedObjectInstance
)

    2.3.4.2 Tapping on Management Instrumentation API

    This is a generic SMI Managed Objects Instances to real-life objects mapping method. It works for scalars and tables of any origin, though, programming with it involves customization of PySNMP SMI base classes what adds up to usage complexity.

    A single SNMP request may invoke an operation on multiple Managed Objects Instances. In SNMP design, it must either succeed on all Managed Objects Instances or be rolled back and reported as a failure otherwise.

    SNMP engine talks to its Managed Objects through a protocol which is comprised from a collection of API methods (further refered to as Management Instrumentation API), implemented by Managed Objects classes and a definite sequence of their invocation. Default handlers implemented in Managed Objects classes read/modify/create the syntax parameter, passed on instantiation, to MibScalarInstance objects for scalars and MibTableColumn for tables. The essence of this Management Instrumentation Tapping technique is to listen on Management Instrumentation API methods for gaining control over particular Managed Object at request processing points.

    Formal parameters of Management Instrumentation API methods don't make much sense to custom implementation, so they are partially documented here and, in most cases, should be blindly passed down as-is to the overloaded method to not to interfere with behind-the-scene SMI workings.

    Value read methods implemented by Managed Objects and invoked by SNMP engine in response to SNMP GET/GETNEXT/GETBULK requests are:

    readTest( *args )

    The readTest method is invoked by SNMP engine prior to performing actual Managed Object Instance value read to give implementation a chance to ensure that subsequent value read is likely to succeed.

    readGet( *args )

    The readGet method is invoked by SNMP engine to fetch Managed Object Instance's value. This method must return a tuple of (name, value) which is returned by overloaded method invocation. Custom implementation may replace the value part by its own version taken from third-party sources.

    readTestNext( *args )

    The readTestNext method is invoked by SNMP engine prior to performing actual Managed Object Instance value read to give implementation a chance to ensure that subsequent value read is likely to succeed.

    readGetNext( *args )

    The readGetNext method is invoked by SNMP engine to fetch Managed Object Instance's value. This method must return a tuple of (name, value) which is returned by overloaded method invocation. Custom implementation may replace the value part by its own version taken from third-party sources.

    The following is a re-implementation of file size monitor: 

    class FileWatcherInstance(MibScalarInstance):
  def readTest(self, name, val, idx, (acFun, acCtx)):
    MibScalarInstance.readTest(self, name, val, idx, (acFun, acCtx))
    try:
      os.stat('/var/adm/messages')
    except StandardError, why:
      raise ResourceUnavailableError(idx=idx, name=name)

  def readGet(self, name, val, idx, (acFun, acCtx)):
    name, val = MibScalarInstance.readGet(self, name, val, idx, (acFun, acCtx))
    try:
      return name, val.clone(os.stat('/var/adm/messages')[6])
    except StandardError, why:
      raise ResourceUnavailableError(idx=idx, name=name)

mibBuilder.exportSymbols(
  'FILESIZE-MIB', FileWatcherInstance((1,3,6,1,4,1,20408,1), Integer())
)

    Value modification methods implemented by Managed Objects and invoked by SNMP engine in response to SNMP SET request:

    writeTest( name, value, *args )

    The writeTest method is invoked by SNMP engine prior to performing actual Managed Object Instance value modification to give implementation a chance to ensure that subsequent value modification is likely to succeed.

    Upon successful completion, this method brings Managed Object Instance into a state of pending modification which ends through either calling writeCleanup() on success or writeUndo() on failure.

    writeCommit( *args )

    The writeCommit method is invoked by SNMP engine by way of request processing in attempt to apply pending value, previously passed to Managed Object Instance through writeTest method. Custom implementation may attempt to apply pending value to a third-party system.

    writeCleanup( *args )

    The writeCleanup method is invoked by SNMP engine by way of request processing to bring Managed Object Instance out of pending value modification state. Custom implementation may attempt to bring a third-party system out of value modification state.

    writeUndo( *args )

    The writeUndo method is invoked by SNMP engine by way of request processing to drop the value applied to Managed Object Instance by the previously called writeCommit() method and re-assign previous value. This method also brings Managed Object Instance out of pending value modification state. Custom implementation may attempt to bring a third-party system out of value modification state.

    The following is a re-implementation of SMI-to-filesystem binding for file modification: 

    class MyFileInstance(MibScalarInstance):
  def writeTest(self, name, val, idx, (acFun, acCtx)):
    MibScalarInstance.writeTest(self, name, val, idx, (acFun, acCtx))
    try:
      open('/tmp/myfile.new', 'w').write(val)
    except StandardError, why:
      raise ResourceUnavailableError(idx=idx, name=name)

  def writeCommit(self, name, val, idx, (acFun, acCtx)):
    MibScalarInstance.writeCommit(self, name, val, idx, (acFun, acCtx))
    try:
      os.rename('/tmp/myfile', '/tmp/myfile.old')
      os.rename('/tmp/myfile.new', /tmp/myfile')
    except StandardError, why:
      raise ResourceUnavailableError(idx=idx, name=name)

  def writeCleanup(self, name, val, idx, (acFun, acCtx)):
    MibScalarInstance.writeCleanup(self, name, val, idx, (acFun, acCtx))
    try:
      os.unlink('/tmp/myfile.old')
    except StandardError, why:
      raise ResourceUnavailableError(idx=idx, name=name)

  def writeUndo(self, name, val, idx, (acFun, acCtx)):
    MibScalarInstance.writeUndo(self, name, val, idx, (acFun, acCtx))
    try:
      os.rename('/tmp/myfile.old', '/tmp/myfile')
    except StandardError, why:
      raise ResourceUnavailableError(idx=idx, name=name)

mibBuilder.exportSymbols(
  'MYFILE-MIB', MyFileInstance((1,3,6,1,4,1,20408,1), OctetString())
)

    Table row creation methods implemented by Managed Objects and invoked by SNMP engine in response to SNMP SET request against a non-existent or SNMPv2-TC::RowStatus type Table Column Instance (table cell) object:

    createTest( name, value, *args )

    The createTest method is invoked by SNMP engine as a first step of Columnar Instance (e.g. Managed Object Instance) creation to make sure the column instance could be created and optionally supplied value is good. Custom implementation may attempt to create a new object at a third-party system.

    The name and value parameters hold OID/value pair as arrived in request.

    Upon successful completion, this method brings Managed Object Instance into a state of pending creation which ends through either calling createCleanup() on success or createUndo() on failure.

    createCommit( *args )

    The createCommit method is invoked by SNMP engine by way of Columnar Object creation to indicate that newly created Columnar Object has been brough on-line and in attempt to apply [optional] pending value, as passed through createTest() method. Custom implementation may bring previously created object on-line at a third-party system.

    createCleanup( *args )

    The createCleanup method is invoked by SNMP engine by way of Columnar Instance creation to indicate a success. Custom implementation may pass this information to a third-party system.

    createUndo( *args )

    The createUndo method is invoked by SNMP engine by way of Columnar Instance creation to indicate a failure. Custom implementation may destroy previously created object at a third-party system.

    The following is a SMI-to-filesystem binding for file creation: 

    class MyFileInstance(MibScalarInstance):
  def createTest(self, name, val, idx, (acFun, acCtx)):
    MibScalarInstance.createTest(self, name, val, idx, (acFun, acCtx))
    # Build path to file to create from column index
    myFileEntry, = mibBuilder.importSymbols('MYFILE-MIB', 'myFileEntry')
    indices = myFileEntry.getIndicesFromInstId(name[myFileEntry.getName()+1:])
    self.__myFile = apply(os.path.join, indices)

    try:
      open('%s.new' % self.__myFile, 'w')
    except StandardError, why:
      raise ResourceUnavailableError(idx=idx, name=name)

  def createCommit(self, name, val, idx, (acFun, acCtx)):
    MibScalarInstance.createCommit(self, name, val, idx, (acFun, acCtx))
    try:
      os.rename(self.__myFile, '%s.old' % self.__myFile)
      os.rename('%s.new' % self.__myFile, self.__myFile)
    except StandardError, why:
      raise ResourceUnavailableError(idx=idx, name=name)

  def createCleanup(self, name, val, idx, (acFun, acCtx)):
    MibScalarInstance.createCleanup(self, name, val, idx, (acFun, acCtx))
    try:
      os.unlink('%s.old' % self.__myFile)
    except StandardError, why:
      raise ResourceUnavailableError(idx=idx, name=name)

  def createUndo(self, name, val, idx, (acFun, acCtx)):
    MibScalarInstance.createUndo(self, name, val, idx, (acFun, acCtx))
    try:
      os.rename('%s.old' % self.__myFile, self.__myFile)
    except StandardError, why:
      raise ResourceUnavailableError(idx=idx, name=name)

# Register custom Managed Object Instance at Column
myFileColumn, = mibBuilder.importSymbols('MYFILE-MIB', 'myFileColumn')
myFileColumn.setProtoInstance(MyFileInstance)

    In the above example, it is assumed that there is a MIB module named MYFILE-MIB where a MIB table column named myFileColumn is defined.

    Appendixies

    ASN.1 standard

    SNMP relies on Abstract Syntax Notation One (ASN.1) ITU-T standard . It is actually a family of standards targeting network systems interoperability and protocols development automation.

    In theory, ASN.1 technology provides a complete solution for protocol development: new protocol could be expressed in terms of data structures described in a specialized formal language.

    The ASN.1 notation is designed purely for data description. All data structures there are based on a small set of elementary data types, such as INTEGER or SEQUENCE OF some other types.

    Whenever protocol designer wants to define a more precise, narrow set of valid values for a field, a subtype can be created from a base ASN.1 type or another subtype by tearing up a constraint on various data properties to parent ASN.1 type. For example, a subtype of in INTEGER may allow only arbitrary values of an integer.

    Another way to create a subtype from existing type is to add or replace ASN.1 tag, which serves like an ID for a type. In this new type has all the same properties of its parent type but is now known under a different name.

    Once something gets expressed in ASN.1 notation, it could then be automatically translated into a variety of platform-specific implementations. They are often take shape of a program written in some common programming language like C or Python.

    This is where the major feature of ASN.1 emerges. ASN.1 text could be automatically compiled into a high-quality code, that handles all the nightmares of platform-specifics, virtually for free. This code would handle byte-ordering and value ranges, data structures validations and consistency issues.

    But the most useful feature is its ability to represent data in a way suitable for transmission over a communication medium. This is called encoding in ASN.1, and also known as concrete or transfer syntax in computer science.

    SNMP uses these features of ASN.1 for handling Managed Objects and guiding protocol operations.

    Object Identifier

    This technique is a simple, unambiguous, decentralized and extensible method of naming anything. It was developed within ASN.1 standard as one of its build-in data types.

    An Object Identifier consists of a sequence of integers. Each integer in this sequence maps to a node in a tree, so iterating an OID traverses this tree from root to leaf, forming a branch. Nodes in OID tree hold a group of conceptually related objects. Nodes become more specific from root to leaves. Sub-trees, or parts of OID space, often become a courtesy of various organizations and individuals.

    OIDs are conventionally written as a dot-separated sequence of integers, from left to right as from root to leaves. For example, .1.3.6.1 is an arbitrary OID.

    For the purpose of making OIDs human-readable, integers in OIDs (AKA sub-OIDs) can be replaced with a textual labels. Consider .org.iso.dod.internet as a labeled version of the previous example. The numeric and labeled OID representations are invariant and may mix within a single OID.

    ASN.1 data encoding

    For several entities to exchange ASN.1 data items some common transmission protocol is needed. This protocol would have to be able to represent ASN.1 values in a platform-native way. This might require handling hardware and/or software specific issues such as varying integer sizes, byte ordering, character encoding and so on.

    Besides data representation issues, this communication protocol would have to break up data being transmitted into small chunks. The reason is that most data transmission technologies handle only a few bits in a channel at any moment of time. After buffering and packing up few bits into larger chunks, most link-level protocols still handle information in small grains. Typical measurement is eight bit or octet.

    For all the reasons mentioned above, ASN.1 family of standards suggests several methods of two-way ASN.1 data conversion protocols. They are sometimes referred to as data encoding or serialization.

    SNMP uses somewhat restricted flavor of Basic Encoding Rules (BER) for its ASN.1 data serialization purposes. The SNMP-specific restrictions make BER encoding deterministic -- with these restrictions applied, there is a one-to-one mapping between ASN.1 value and octet-stream produced by BER encoder. Determinism in encoding makes it possible for trivial SNMP entities to reduce their SNMP engine implementation to opaque octet-streams manipulations.

    Warning! This document is a draft. It is neither complete nor accurate. Take it with a grain of salt!