Home > Articles > Operating Systems, Server > Microsoft Servers

  • Print
  • + Share This
This chapter is from the book

How Does OpsMgr Do It?

So far in this chapter, we have covered what a management group is, and how the components, or computer roles, of a management group communicate with one another—the macro view. Now we shift our focus to the micro view of the management pack—the computer and device management work the whole OpsMgr infrastructure was deployed for. The management group is the framework within which management packs do that work.

Operations Manager 2007 is a product established on the concept of model-based management. The abstraction of services into models is needed to describe and act on physical entities such as routers, and logical entities such as distributed applications, using software tools that by definition exist in cyberspace. Using models is a way to transform human knowledge and experience into something machines can operate with. In OpsMgr, service models live inside management packs. The management pack author or vendor encapsulates service health knowledge into the redistributable management pack.

Having a solid, accurate model of an object's health lets OpsMgr 2007 present information to the operator in the most immediately useful way. As you will see, the models underpin both the OpsMgr 2007 application, with a workflow framework, and the OpsMgr 2007 operator, with augmented and accelerated decision making.

Operations Manager 2007 introduces an architecture that sets the foundation for a new, broader spectrum of monitoring capabilities and extensibility than has ever been available before using Microsoft management technologies. OpsMgr 2007 fundamental concepts include service and health modeling (we will explain and differentiate between those terms). We'll briefly cover the schema of a management pack so that you understand how a service model is distilled into actionable components such as monitors and tasks. In addition, we will illustrate how monitors are the intersection between the models, and how health information progresses inside the OpsMgr workflow engine to its presentation in the OpsMgr console.

Service Modeling

One can capture knowledge through models! Service modeling in Operations Manager 2007 is rooted in the well-known Service Modeling Language (SML) used by Microsoft developers in the .NET development environment. SML is an extensible language, built for describing the cooperating systems found not just inside the computer, but also inside an entire datacenter. SML provides a way to think about computer systems, operating systems, application-hosting systems, and application systems—as well as how they interact and are combined, connected, deployed, and managed. SML is used to create models of complex IT services and systems.

A software engineer authoring in Visual Studio Team Edition for Software Architects uses SML to define how an application interacts with various layers of the datacenter, such as the hardware layer, where the servers and routers live, and the operating system layer, which is "hosted" by the hardware layer. The SML concept of one layer hosting another is used in OpsMgr service modeling when relationships are defined between objects managed by OpsMgr, such as a hard drive that hosts a website.

OpsMgr 2007 operates on a class-based structure. When the monitoring infrastructure discovers an "object" (or entity), it assigns a set of logical classes to the object. These classes serve as descriptors for the managed object. The SML for a managed object is imported into OpsMgr using the vehicle of the management pack. Specifically, the management pack adds the formal definitions of "types of objects" (or classes), their properties, and the relationship between objects in the management group. Relationships usually take the form of a dependence on another object, or of a container of another object.

Without management packs and the knowledge they deliver, any OpsMgr group is just a big empty brain. You can compare a management group to the brain, which is a physical structure; in contrast, management packs are analogous to the memories and ideas that live in that brain. Useful thoughts are crafted in the brain based on knowledge and experience. Useful workflow in a management group is made possible by management packs.

We can continue to use a biological metaphor to explain the way management packs convert human knowledge and experience into actionable machine workflows. In the medical profession, a very precise lexicon exists to describe objects in the body. If you think of the parts of your body, you realize the many classes, properties, and relationships that exist. Here are some examples to get you thinking this way:

  • You have a sensory organ "class" that include "objects" such as your eyes, ears, tongue, nose, and skin.
  • Many objects in your body need to be described along with a property or qualifier, such as "left" or "right," or "proximal" or "distal" to distinguish the particular body part (object).
  • Every object in the body has one or more relationships with other objects, such as the hand "depending" on the arm, or arteries that "contain" blood.

Classes, objects, and relationships are how OpsMgr recognizes an object, understands what the object is, and how to work with the object. Just as we more precisely describe a particular body part by adding the descriptor "left" to the object "hand," OpsMgr describes objects using a hierarchical system of descriptors that are increasingly specific.

Now you will see this SML layer concept in action as we describe a particular object, a website running on a managed Windows server. See the diagram in Figure 3.6, starting in the upper-left portion of the description, the Entity. This is another word for "object" in OpsMgr, and it's like a placeholder for the object's root.

Figure 3.6

Figure 3.6 Describing an object using the System Modeling Language.

Proceeding down and to the right in the hierarchy, or "tree," depicted in Figure 3.6, we add descriptors to successively narrow, or focus, the description of the particular managed object. As depicted, the Windows Computer Role is a subordinate descriptor to Computer Role. Likewise, the Internet Information Services (IIS) service is a particular Windows Computer Role in OpsMgr, and the monitored website is a particular feature of the IIS service.

Also illustrated in Figure 3.6 are relationships between objects, such as the Windows Operating System (OS) hosting the IIS service, and a particular disk drive hosting the monitored website—which is the object of interest in this description.

The ability of management packs to define relationships between objects, using such terms as "reference," "using," "hosting," and "containing," is critical to technological innovations found in OpsMgr over previous Microsoft management technologies. OpsMgr features such as monitoring distributed applications with containment relationships, diagrammatic cross-platform fault identification, and maintenance mode on individual computer components are possible via SML and its layered approach to describing objects.

Management pack authors include the ability to discern both objects and relationships between objects in the discovery process. Objects and relationships are discovered with probes that examine computer registries using Windows Management Instrumentation (WMI) queries, scripts, database queries (OLE DB), the Lightweight Directory Access Protocol (LDAP), and custom or "managed" code.

We're going to dive right into an advanced view of the Authoring space to highlight the importance of the process to discover both objects and their relationships in order to understand how OpsMgr works. In Figure 3.7, observe the OpsMgr Authoring space, focused on the Object Discoveries branch of the Management Pack Objects section. In the upper portion of the center pane, notice we have expanded three discovered type classes:

  • Windows Server 2003 Disk Partition
  • Windows Server 2003 Logical Disk
  • Windows Server 2003 Physical Disk
Figure 3.7

Figure 3.7 Probes such as WMI and Registry key queries discover object attributes.

Arrows on the left in Figure 3.7 point to object discovery rules (distributed in the Windows Server 2003 Base OS management pack) that discover disk partitions, logical disks, and physical disk attributes using WMI queries. In the lower (Details) portion of the center pane, we can see the actual WMI query strings used when discovering Windows logical disks (in this case looking for attributes such as what file system is in use and whether the volume is compressed).

Of course, disk partitions as well as logical and physical disks are highly interrelated object classes. Physical disks can contain multiple disk partitions, which in turn may contain multiple logical disks. Logical disks can span multiple disk partitions and physical disks.

Notice in Figure 3.7 that the target column of the discovery rules for a particular object type such as "Windows Server 2003 Disk Partition" identifies the object type that hosts the discovered type. For example, the Windows Server 2003 Operating System (OS) hosts Windows disk partitions; therefore, the Discover Windows Disk Partitions object discovery rule targets the Windows Server 2003 OS object type (or class).

Relationship discovery rules operate in addition to object discovery rules. Object discovery rules use WMI or other probes to locate managed objects and populate the Operations database with actionable object attributes. This enables relationship discovery rules to look at object properties for particular discovered attributes that indicate a dependence, hosting, or containing relationship.

After the Windows Server 2003 Base OS management pack discovers various disk objects, it also discovers the relationships between these classes of disk objects using separate relationship discovery rules. See the relationship discovery rules called out in Figure 3.8. The four relationship discovery rules in this example identify the relationships between physical disks and disk partitions, and between disk partitions and logical disks.

Figure 3.8

Figure 3.8 Both object and relationship discovery rules are associated with most object types, or "classes."

Health Models

After object and relationship discovery is complete, the Operations database is populated with the object's descriptive data (its attributes). Now OpsMgr can begin performing the primary work of the management pack, managing the state of the object's health model.

Every class, or object type, has a health model. The status, or health, of even the simplest managed object is represented by a health model. A model is a collection of monitors. We will be covering monitors in detail later in the "Monitors" section of this chapter. As we add monitors, we enrich the health model.

Monitors are arranged in a tree structure that is as deep or as shallow as required. The status of the health model represents the current state of the object. The Health Explorer shows a live view of an object's health model. The Health Explorer tool can be launched against any managed object from all views in the Monitoring pane of the console.

A key monitoring concept in OpsMgr 2007 is the rollup. We first heard this term from Microsoft early in OpsMgr development, used to describe the way health status "bubbles up" from lower levels in the health model hierarchy, or tree, to higher-level monitors. The top-level monitor in a health model, located at the root Entity object layer, is the rollup, which represents the overall health state of the object.

We return to the Service Modeling Language layer-based method of classifying objects introduced along with the concept of service modeling. Figure 3.9 diagrams (on the left side) the tree-like class hierarchy of the IIS service on a Windows computer. Notice the unit monitors located at the lower right of the diagram inline with the IIS service. Monitors in each of the four basic categories (Availability, Security, Performance, and Configuration) are represented by round "pearl" shapes.

Figure 3.9

Figure 3.9 The layers of SML allow for tactical placement of hierarchical monitors.

Lower-layer monitor status is propagated by the health model up to monitors in successively higher layers. For any given health model, monitors are not necessarily located at every layer, or within every monitor category. The management pack author determines what monitors are targeted against what object classes.

Finally, notice in Figure 3.9 the uppermost, triangular arrangement of four monitors rolling up into the health state for the managed object. The rollup occurs at the entity level; this is a universal feature of OpsMgr object health models. The second-level monitors that roll up into the top-level state monitor are called aggregate monitors.

State-based Management

Another key theme in OpsMgr is the employment of state-based management, in contrast to previous versions of Operations Manager that were alert-based. An alert-based management system watches for a condition, raises an alert, and optionally changes the state of the object due to the generation of the alert.

The point of the Health Explorer is to illustrate the state of a managed object's health, not to present a list of new or unacknowledged alerts that require operator evaluation. MOM 2005 administrators already know the difficulty in rapidly correlating and triaging a laundry list of alerts in order to answer the question, "What do we need to do to fix the problem?"

The OpsMgr implementation of state-based management applies the following workflow sequence:

  1. A unit monitor watches for a condition.
  2. When the unit monitor detects the condition, it changes the state of the unit monitor.
  3. Unit monitor states are rolled up as required to higher-level aggregate monitors in the object's health model.
  4. Rules optionally generate an alert or initiate a notification event.

Management Pack Schema

A management pack is an eXtensible Markup Language (XML) document that provides the structure to monitor specific software or hardware. A sealed managed pack is a read-only, encrypted version of the XML document. This XML document contains the definitions of the different components in software or hardware and the information needed by an administrator who must operate that application, device, or service efficiently.

We will take a quick look at the schema of the management pack so that you can appreciate how tightly management pack construction is aligned with the health model of an object. A high-level view of the management pack schema is diagrammed in Figure 3.10. From the management pack root, moving right, there are eight major sections: Manifest, TypeDefinitions, Monitoring, Templates, PresentationTypes, Presentation, Reporting, and LanguagePacks.

Figure 3.10

Figure 3.10 Management pack schema, with the Manifest and Monitoring sections expanded.

Only the Manifest section is mandatory, and that section is expanded in the upper-right portion of Figure 3.10. The Manifest section defines the identity and version of the management pack as well as all other management packs it is dependent on. The Identity, Name, and References sections are common and included in every management pack. Any management packs referenced must be sealed, and they must be imported to the OpsMgr management group before the management pack can be imported.

The other major schema section that is expanded in Figure 3.10, Monitoring, is where most of the action takes place in OpsMgr, and this chapter also is mainly about the sections you see contained there. The following list summarizes the purpose of each section of the Monitoring schema:

  • Discoveries—A discovery is a workflow that discovers one or more objects of a particular type. A discovery can discover objects of multiple types at one time. As introduced previously in the "Service Modeling" section of this chapter, there are both object discovery and relationship discovery rules.
  • Rules—A rule is a generic workflow that can do many different things. As an example, it could collect a data item, alert on a specific condition, or run a scheduled task at some specified frequency. Rules do not set state at all; they are primarily used to collect data to present in the console or in reports and to generate alerts.
  • Tasks—A task is a workflow that is executed on demand and is usually initiated by a user of the OpsMgr console. Tasks are not loaded by OpsMgr until required. There are also agent-initiated tasks, where the agent opens up a TCP/IP connection with the server, initiating the communication. After the connection is established, it is a two-way communication channel.
  • Monitors—A monitor is a state machine and ultimately contributes to the state of some type of object that is being monitored by OpsMgr. There are three monitor types: aggregate (internal rollup), dependency (external rollup), and unit monitors. The unit monitor is the simplest monitor, one that simply detects a condition, changes its state, and propagates that state to parent monitors in the health model that roll up the status as appropriate. We cover monitors in more detail in the next section of this chapter.
  • Diagnostics—A diagnostic is an on-demand workflow that is attached to a specific monitor. The diagnostic workflow is initiated automatically either when a monitor enters a particular state or upon demand by a user when the monitor is in a particular state. Multiple diagnostics can be attached to a monitor if required. A diagnostic does not change the application state.
  • Recoveries—A recovery is an on-demand workflow that is attached to a specific monitor or a specific diagnostic. The recovery workflow is initiated automatically when a monitor enters a particular state or when a diagnostic has run, or upon demand by an operator. Multiple recoveries can be attached to a monitor if required. A recovery changes the application state in some way; hopefully it fixes any problems the monitor detected!
  • Overrides—Overrides are used to change monitoring behavior in some way. Many types of overrides are available, including overrides of specific monitoring features such as discovery, diagnostics, and recoveries. Normally the OpsMgr administrator or operator sets overrides based on his specific, local environment. However, in some cases, a management pack vendor may recommend creating overrides in particular scenarios as a best practice.

Monitors

It all starts with monitors in Operations Manager 2007. We have mentioned that a health model is a collection of monitors. If you were to author a management pack, you would probably start with creating unit monitors. Unit monitors would detect conditions you determine are essential to assess some aspect of the health of the application, device, or service needing to be managed.

Monitors provide the basic function of monitoring in OpsMgr. You can think of each monitor as a state machine, a self-contained machine that sets the state of a component based on conditional changes. A monitor can be in only one state at any given time, and there are a finite number of operational states.

A monitor can check for a single event or a wide range of events that represent many different problems. The goal of monitor design is to ensure that each unhealthy state of a monitor indicates a well-defined problem that has known diagnostic and recovery steps.

Using a single monitor to cover a large number of separate problems is not recommended, because it provides less value. We mentioned in the lead-in to the "Health Models" section of this chapter that adding monitors to a health model increases the richness of an object's monitoring experience. The enhancement of an object's health model with many monitors adds fidelity to the health state of the object. More monitors in a health model also means more relationship connection points for other managed objects that host, contain, depend on, or reference that object.

We pointed out the "pearl" icon used to represent a monitor in health model diagrams. An empty pearl icon represents a generic or a non-operational monitor. Figure 3.11 is a chart showing the default monitor icon images and their corresponding operational state.

Figure 3.11

Figure 3.11 These state icons are encountered in the Operations console.

A functioning monitor displays exactly one of the primary state icons: green/success, yellow/warning, or red/critical. A newly created or nonfunctional monitor will show the blank pearl icon. The gray maintenance mode "wrench" icon appears in all monitoring views inline with the object that was placed in maintenance mode. The final type of state icon you will encounter is the grayed state icon, which indicates that the managed object is out of contact. For example, this could reference a managed notebook computer that is off the network at the moment.

To be clear, there are three kinds of monitors that management pack authors can create: aggregate rollup monitors, dependency rollup monitors, and unit monitors. In the next sections we will describe each of these monitor types.

Aggregate Rollup Monitors

Let's return to the Figure 3.9 view of the layers of the SML, which permits tactical placement of interrelated monitors. On the right, notice the monitors are classified in categories, essentially four vertical columns that are connected by a rollup to the top-level entity health status. Microsoft selected these four categories during OpsMgr development as a framework to aggregate the health of any managed object.

The four standard types of aggregate monitors in a state monitor are detailed in the following list:

  • Availability Health—Examples include checking that services are running, that modules within the OpsMgr health service are loaded, and basic node up/down tracking.
  • Performance Health—Examples include thresholds for available memory, processor utilization, and network response time.
  • Security Health—Monitors related to security that are not included in the other aggregate monitors.
  • Configuration Health—Examples include confirming the Windows activation state and that IIS logging is enabled and functioning.

Dependency Rollup Monitors

The second category of monitor is the dependency rollup. Such a monitor rolls up health states from targets linked to one another by either a hosting or a membership relationship. Dependency rollup monitors function similarly to aggregate rollup monitors, but are located at intermediate layers of the SML hierarchy.

In Figure 3.9, notice again the unit monitors for the IIS service located in the lower right. There are two unit monitors of the performance type at the IIS Service level that merge at the Windows Computer Role level. The merge point represents one or more dependency rollup monitor(s) targeted at the Windows Computer Role.

Earlier in the "Service Modeling" section of this chapter, we explored how objects such as disk partitions, logical disks, and physical disks have numerous relationships. Figure 3.12 shows a sample dependency rollup monitor involving disk systems created in the OpsMgr authoring space.

Figure 3.12

Figure 3.12 Creating a dependency rollup monitor when the target is a disk partition.

The monitor created in Figure 3.12 is targeted against the Windows Server 2003 Disk Partition class. OpsMgr knows that disk partitions contain logical disks, so when you create a new dependency rollup monitor targeting the Windows Server 2003 Disk Partition class, OpsMgr offers existing monitors to select from for the Windows Server 2003 Logical Disk class.

We can also expand the example of the "merged" IIS service performance unit monitors in Figure 3.9. If we were creating that dependency rollup monitor in the authoring space, we would have selected the Windows Computer Role as the target of our monitor. The Create a Dependency Monitor Wizard would provide us with a list of dependent objects to select from that includes those IIS service performance monitors.

Unit Monitors

A unit monitor allows management pack authors to define a list of states and how to detect those states. A simple unit monitor is a Basic Service Monitor. This monitor raises state changes when a Windows service stops running. More complex unit monitors run scripts, examine text logs, and perform Simple Network Management Protocol (SNMP) queries. A unit monitor is deployed, or targeted, at a class of objects when it is authored.

When creating monitors and envisioning operational states, Microsoft advises OpsMgr administrators and management pack authors to do so without initially regarding actual implementation of those monitors. The reasoning is that OpsMgr not only provides many monitor types by default for common scenarios, but makes it possible to build different workflows to meet any monitoring requirement. Basically, the management pack architect is encouraged to think "outside the box" and describe in plain ideas how an application's health can be assessed. After that, you can look to the many tools OpsMgr provides to instrument the application accordingly.

Figure 3.13 presents a montage screenshot that includes all possible types of unit monitors available in the authoring space of the OpsMgr console. These are the tools used to architect the instrumentation of the health model.

Figure 3.13

Figure 3.13 The complete menu of types of unit monitors that can be created.

Over 50 unit monitor types are available to place as software instrumentation in the SML framework. Remember that unit monitors roll up into the aggregate monitors (Availability, Performance, Security, and Configuration), sometimes via dependency rollup monitors. The goal of monitor design is to ensure that each unhealthy state of a monitor indicates a well-defined problem that has known diagnostic and recovery steps. Table 3.2 provides some explanation of the unit monitor types found in the menu in Figure 3.13.

Table 3.2. Unit Monitor Types

Monitor type

Description

Average Threshold

Average value over a number of samples.

Consecutive Samples over Threshold

Value that remains over or below a threshold for a consecutive number of minutes.

Delta Threshold

Change in value.

Simple Threshold

Single threshold.

Double Threshold

Two thresholds (monitors whether values are between a given pair of thresholds).

Event Reset

A clearing condition occurs and resets the state automatically.

Manual Reset

Event based; wait for operator to clear.

Timer Reset

Event based; automatically clear after certain time.

Basic Service Monitor

Uses WMI to check the state of the specified Windows service. The monitor will be unhealthy when the service is not running or has not been set to start automatically.

Two State Monitor

Monitor has two states: Healthy and Unhealthy.

Three State Monitor

Monitor has three states: Healthy, Warning, and Unhealthy.

To conclude this section on monitors, we're going to put it all together by overlaying the SML and the health model for a live service monitor. Figure 3.14 is a fully expanded view of the health model of the OpsMgr Health service itself running on a management server.

Figure 3.14

Figure 3.14 Expanded view of the health model for the OpsMgr Health Service.

Beginning at the lowest level of the object description tree, we see the MonitoringHost Private Bytes Threshold unit monitor on the computer Hurricane. Five unit monitors are shown in the lowest row that roll up into the Health Service Performance monitor. These unit monitors are labeled with the abbreviations Svc Handle, Svc Priv, Mon Handle, Mon Priv, and Send Queue in Figure 3.14. The MonitoringHost Private Bytes Threshold (abbreviated Mon Priv) unit monitor is in a critical state.

We can follow the propagation of this unit monitor state up the health model. The OpsMgr Health service is an application component of Windows Local Application Health Rollup. The Health Service is in a critical state due to the critical state of the MonitoringHost Private Bytes Threshold (abbreviated Mon Priv) unit monitor. Progressing upward, the application state is rolled up along with the hardware, OS, and computer states to the performance component of the object.

The critical state is propagated to the application component of the performance monitor. Finally at the top of the health model, an aggregate monitor rolls up the performance, availability, security, and configuration monitors. The root entity, which is the server Hurricane itself, indicates the aggregated health state, which is critical.

Figure 3.15 shows the Health Explorer for the computer in the state illustrated in Figure 3.14. If you noticed the critical state of the computer in the Monitoring pane of the Operations console, you would probably open the Health Explorer for the computer, which allows you to understand quickly what is wrong. By comparing the structure of the Health Explorer in Figure 3.15 with the SDK and health model layers presented in Figure 3.14, you can match up the same critical health icons in the health model and the Health Explorer.

Figure 3.15

Figure 3.15 Health Explorer screenshot of the health model detailed in Figure 3.14.

Workflow

It is accurate to describe Operations Manager 2007 at its core as being a giant workflow engine. In fact, monitoring in OpsMgr is based around the concept of workflows. An Operations Manager agent and server will run many workflows simultaneously in order to discover and monitor applications, devices, and services.

Module Types

Module types are the building blocks of Operations Manager workflows. Workflows are defined in management packs and then distributed to managed computers. Workflows can do many things, including collecting information and storing data in the Operations database or data warehouse, running timed scripts, creating alerts, and running on-demand tasks. Workflows are defined using modules, and modules are defined to be of a particular type known as a module type. Four different module types can be defined: data source, probe action, condition detection, and write action. Figure 3.16 illustrates these module types.

Figure 3.16

Figure 3.16 Workflow in OpsMgr is performed through four specific module types.

In the "Architectural Overview" section of this chapter, we compared the management group and management pack to macro and micro views that answer the question "How does OpsMgr do it"? In this section, we are going sub-micro! At the programmatic level, these are the terms and data flow structures used internally by the OpsMgr services:

  • Data Source—A data source module type generates data using some form of instrumentation or some timed trigger action. As an example, a data source may provide events from a specific Windows event log or it could poll Windows performance counters every 10 minutes for a specific performance counter. A data source takes no input data and provides one output data stream. Data sources do not change the state of any object.
  • Probe Action—A probe action module type uses some input data to provide some output data. A probe action will interrogate a monitored entity in some way, but it should not affect system state in any way. An example would be running a script that queries WMI to get some data. A probe action is often used in conjunction with a data source to run some action on a timed basis. The probe action module type may or may not use the input data item to affect the behavior. In other words, when triggered, a probe action generates output from external sources. Probe actions have one input stream and one output stream. Like data source modules, probe action modules do not change the state of objects.
  • Condition Detection—A condition detection module type filters the incoming data in some way. Examples of filter types include a simple filter on the input data, consolidation of like data items, correlation between multiple inputs, and averaging performance data. A condition detection module type can take one or more input data streams and provides a single output data steam. Condition detection modules do not use any external source and do not change object state.
  • Write Action—A write action module type takes a single input data stream and uses this in conjunction with some configuration to affect system state in some way. This change could be in the monitored system or in Operations Manager itself. As an example, the action may be to run a script that writes data into the Operations database, or one that generates an alert. A write action may or may not output data. This data cannot be passed to any other module because the write action is the last module in a workflow. However, the data may be sent to the Operations database. A sample action is running a command that outputs data, such as a command line that returns a report of success or errors. This data may be useful to the operator who executes the command, and it is returned to the Operations console and stored as task output.

"Cook Down"

Cook down is an important concept in management pack authoring. The Operations Manager agent or server is running many hundreds or even thousands of workflows at any given time. Each workflow that is loaded takes some system resource. Obviously the less system resources we take up for monitoring, the better.

The management pack author can do a lot to reduce the impact of monitoring on the system. One way is to ensure that workflows are not targeted too generically. We mentioned this already in this chapter, in the section on "Unit Monitors." For example, if you have a rule that is only applicable to servers running Microsoft ISA Server 2006, don't target the rule at all Windows servers; instead, you should target it at the appropriate ISA Server class.

Cook down is not about targeting; it is a principle whereby in most modules the Operations Manager Health service will try to minimize the number of instances in memory. This is accomplished by considering the configuration of modules. Usually, if the Health service sees two modules with the same configuration in different workflows that have the same configuration, it will only execute a single module and feed the output to all the workflows that defined the module. This is an efficiency you should be aware of, particularly if you will be authoring scripts for use by OpsMgr.

Here is a simple example of two rules that will "cook down":

  • Rule 1—Collect an event from the application log where Event ID =11724 and Event Source = MsiInstaller (application removal completed).
  • Rule 2—Collect an event from the application log where Event ID =1005 and Event Source = MsiInstaller (system requires a restart to complete or continue application configuration).

Operations Manager sees that the event log provider data source (application log events) is configured the same for both rules. Only one instance of the module will run. The two MsiInstaller event ID rules, or expression filters, will take input data from the output of the same module. A large number of expression filters can be handled by one condition detection module. In the case of the event log provider example, there will normally be only one module executing for each log being monitored (unless you are running the module under different credentials for different workflows).

Cook down becomes particularly important when writing scripts to be run by OpsMgr, especially when there are scripts running against multiple instances of an object type on the same Health service. If you do not think about cook down, you could end up running many scripts when you could actually run a single script by thinking about configuration and targeting.

Data Types

We have discussed module types and how they are used by OpsMgr internally to achieve workflow. Obviously, OpsMgr must pass data between modules. The format of this data varies depending on the module that output the data. As an example, a data source that reads from the event log will output a different type of data than a module that reads from a text-based log file. Some module types expect a certain type of data. A threshold module type expects performance data and the module type that writes data to the Operations Manager database expects event data. Therefore, it is necessary for Operations Manager to define and use different data types.

Data types are defined in management packs. However, this definition is merely a pointer to a code implementation of the data type. Operations Manager 2007 does not support extension of the data types provided out of the box.

Data types follow an inheritance model in a manner similar to class definitions, introduced in the "Service Modeling" section of this chapter. Whereas the class hierarchy starts with a base class called System.Entity, the data type hierarchy starts with a data type called System.BaseData. All data types eventually inherit from the base data type. Examples of data types in the System.BaseData class include Microsoft.Windows.RegistryData (for a probe action module that examines Registry values) and System.CommandOutput (for write action modules that return useful command-line output).

When a module type is defined it must, where applicable, specify the input and output data types that it accepts and provides. These must be valid data types defined in the same management pack or a referenced management pack. When a module is used in a workflow, the data types that the module type accepts and provides must be compatible with the other modules in the workflow.

  • + Share This
  • 🔖 Save To Your Account