Skip to content

Latest commit

 

History

History
457 lines (340 loc) · 19.2 KB

developer-guide.md

File metadata and controls

457 lines (340 loc) · 19.2 KB

SLA Core Developer Manual

Architecture

Architecture

The sla-core must be flexible enough to work with the different monitoring systems available:

  • Simple monitoring systems that must be polled in order to retrieve the metrics.
  • Monitoring systems that are able to push the metrics into the sla core once the are available.
  • Smart monitoring systems that are able to evaluate the constraints, and raise the appropriate violations

Conceptual class diagram

In the following diagram, it is shown the main entities and terms that are in the spec and the way they are related.

TODO Image

  • A Provider offers a Service: a software service, hardware resources, etc.
  • The service is described by ServiceDescriptionTerms with a Domain Specific Language. The ServiceDescriptionTerms are intended to define a service that has to be provisioned. This sla module needs external provision.
  • The service is represented by a Template, and the Template can be used to generate an Agreement.
  • An agreement is a "document" that associates a Service and a Consumer. When the relation is in negotiation-phase, it's called an AgreementOffer. Once the agreement is accepted, it's called a Contract (this name is not used by the spec).
  • A Template and an Agreement can describe some restrictions (DSLed) to be fulfilled by the Consumer or by the Provider.
  • A violation of any restriction generates a Violation.

The WS-Agreement xml specifies in the context element who is the provider and the consumer, with the elements:

  • AgreementInitiator (some kind of initiator identifier; OPTIONAL),
  • AgreementResponder (some kind of responder identifier; OPTIONAL),
  • ServiceProvider ( = AgreementInitiator | AgreementResponder)

Usually, the consumer is the initiator. Although ws-agreement specifies AgreementInitiator/Responder as optional, it is recommended to specify them. In case of a multi-provider configuration, the SP MUST be specified.

##Agreement class diagram##

TODO: Image

Every entity is identified internally (to the sla-core) with autoincremented ids. Every entity that need identification outside the sla-core will be identified by an uuid if the entity is created internally. If the entity is created externally, it is not mandatory to be an uuid, but unique.

###Providers###

The Providers API defines the ProvidersRepository and the CRUD operations to be called by the sla module user in the case of a multiprovider environment. If the sla is sticked to only one provider, this should be created once in module initialization.

Enforcement

The enforcement is the process by which it is evaluated that the provider complies with an agreement, i.e. the measured metrics for the variables in guarantee terms fulfill the constraints. The sla-core must be flexible enough to work with the different monitoring systems available:

  • Simple monitoring systems that must be polled in order to retrieve the metrics.
  • More complex monitoring systems that push new metrics to observers (in this case, the sla core).
  • Smart monitoring systems that are able to evaluate the constraints, and raise the appropriate violations.

The current enforcement implementation is able to work with the first two monitoring systems, while a bit of customization is needed to be utilized with the third one.

The enforcement process is depicted in the following sequence diagrams:

Periodic Enforcement

On Demand Enforcement

Agreement Evaluation

The job of each class is the following:

  • AgreementEnforcement: retrieves the metrics needed to enforce an agreement (if periodic enforcement execution), calls the AgreementEvaluator and saves the detected violations and compensations in repository.
  • AgreementEvaluator: calls the GuaranteeTermEvaluator for each guarantee term in the agreement.
  • GuaranteeTermEvaluator: calls the ServiceLevelEvaluator, obtaining the raised violations; calls the BusinessValuesEvaluator using the violations as input.
  • ServiceLevelEvaluator: calculates the violations. The implemented evaluator uses the concept of policy, where a number of violation metrics (a breach) must occur in a period of time (specified in the policy) to raise a violation. More details below.
  • ConstraintEvaluator. Parses the service level constraint, and evaluates if a metric fulfills the constraint.
  • MetricsRetriever. Actively queries for the last metrics of an agreement. Used in periodic execution.
  • MetricsReceiver. Passively receives the last metrics of an agreement. Used in on-demand execution. The frontend for the receiver can be a REST service, a Message Queue, etc.

###Metrics Retriever###

It's very likely that the monitoring system is independent from the SLA module. For this reason, the ServiceProvider has to implement an adapter in order to the SLA retrieve the metrics needed to do the enforcement.

The ServiceProvider has to implement one of the following interfaces in its adapter to the monitoring system:

public interface IMetricsRetriever {
	public List<IMonitoringMetric> getMetrics(
		String agreementId, 
		String serviceScope, 
		String variable, 
		Date begin, 
		Date end, 
		int maxResults);
}

public interface IMetricsRetrieverV2 {
	
	public interface RetrievalItem {
		IGuaranteeTerm getGuaranteeTerm();
		String getServiceScope();
		String getVariable();
		Date getBegin();
		Date getEnd();
	}

	public Map<IGuaranteeTerm, List<IMonitoringMetric>> getMetrics(
		String agreementId, 
		List<RetrievalItem> retrievalItems, 
		int maxResults);
}

Constraint Evaluation

A simple constraint definition and evaluation is implemented, where a constraint can be defined by the following grammar:

E -> id OP VALUELIST
OP -> "GT" | "GE" | "EQ" | "LT" | "LE" | "NE" | "BETWEEN" | "IN"
VALUELIST -> "(" value "," value ")"
VALUELIST -> value

Only float values are allowed.

For example, the following constraints are valid:

responsetime LT 200
availability EQ 1
voltage BETWEEN (4.5, 5.5)
status IN (200, 204)

A constraint is satisfied if a metric evaluate the condition to true. Otherwise, the metric is considered as a Breach.

If any sla provider wants to substitute the default constraint evaluation, the interface to implement is the following:

public interface IConstraintEvaluator {
	
	List<IMonitoringMetric> evaluate(
		String kpiName, String constraint, List<IMonitoringMetric> metrics);

	String getConstraintVariable(String constraint);	
}
  • The getConstraintVariable() returns the metric name that has to be passed as parameter to the monitoring adapter.
  • The evaluate() function return which metrics are breaches in the list passed as parameter.

Aggregated values

Two types of metrics can be considered:

  • Simple metrics. Each measure is a metric value.
  • Aggregated metrics. The metric value is an aggregation of values using an aggregation function. For example (using SQL syntax): SUM, AVG, COUNT... The aggregated values makes sense within a time interval or number of samples interval. As such, the interval must be implicit or explicitly stated in the metric name. For example, a metric called "avg-uptime" can be considered by the Monitoring System (or the adapter), the uptime per hour.

In this section it is explained how to evaluate aggregated values. Let's suppose that the constraint to satisfy is avg-availability-3600, being defined like this in the service level.

The most efficient and simple way to handle aggregated values it that the Monitoring System is ready for them. For example: the metric name avg-availability-3600 can be interpreted by the Monitoring System as the avg availability of the last hour, and return the value accordingly. The monitoring adapter can perform a little name change if needed (e.g.: from avg-availability-3600 to AVG(availability, 3600)).

The second place to handle aggregated values is in the monitoring adapter. The adapter have to take into account that the metric name starts with avg- (or whatever). Then, it queries the availability samples in the last hour, and calculates the average. It is less efficient than the first method, but it is still decoupled enough to be easily customizable and extendable.

The third place to handle aggregated values is in the ConstraintEvaluator itself. The reference implementation of ConstraintEvaluator calculates an average for variables that have the following syntax: avg_<variable>_<timeinterval>. The metric name to be passed to the monitoring adapter is <variable>.

Complex constraints

It is a restriction of the SLA core that only one variable can be retrieved from the monitoring system per constraint (it greatly simplifies the implementation).

So, if it is needed to evaluate constraint of the type: varA < value1 AND varB < value2, this must be accomplished with the following approaches:

  • Let the monitoring system evaluate the constraint, if it is able to do so.
  • Let the monitoring adapter evaluate the constraint. For example, for the previous constraint, IConstraintEvaluator.getConstraintVariable() can return varA_value1_AND_varB_value2. The adapter interprets the variable, and makes the appropriate evaluation, returning 1 or 0 for each tuple. The tuples must be carefully handled, because the monitoring platform may not return the values as tuples. It is advisable to use only aggregated metrics, as only one value per variable is returned.
  • Implement a ConstraintEvaluator that evaluates this kind of constraints. The getConstraintVariable() must return the variables (f.e., as varA,varB). As the monitoring adapter returns only a list of MonitoringMetric (and not a list of lists), the adapter can choose to:
    • "Zip" each tuple in a single MonitoringMetric, using, comma separated values in the metric name and in the metric values. E.g.: a metric can be (name="var_a,var_b",value="1,0",time="2014-06-09 16:39").
    • Use a MonitoringMetric for each variable. The ConstraintEvaluator has to take metrics in groups of two.

Policies

The policies used in C4S are an interesting feature not covered by the WS-Agreement.

A policy is compound by:

  • a date interval
  • a number of occurrences

The objective of a policy is specify when to raise a violation. Instead of raise a violation on every constraint breach, it's raised when a number of breaches are found within the specified interval.

WS-Agreement does not make room to define something like a policy in a ServiceLevelObjective, so we are going to define the policy in the CustomServiceLevel, along with the constraint definition. So, it's proposed to define the CustomServiceLevel like this:

{ 
  policies: [ { count : 2, interval: 120 }, { count: 2, interval: 3600 ],
  constraint: "responsetime LT 100"
}

So, if a service provider wants to offer policies in their SLA, they have to be compliant with this format. The constraint string is still totally domain defined.

###Business rules### A simple and generic implementation of business rules has been included in the core. Each guarantee term may have a BusinessValueList element where the penalties of not satisfying a guarantee term are defined.

The business structure defined in the wsag specification is not expressive enough for our purposes, so that structure is basically ignored, and the implementation makes use of CustomBusinessValues to define the business rules as a generalization of the standard penalties.

The recognized xml structure is:

<wsag:BusinessValueList>
  <wsag:Importance>xs:integer</wsag:Importance>?
  <wsag:CustomBusinessValue count="xs:integer" duration="xs:duration">
    <sla:Penalty 
      type="xs:string" 
      expression="xs:string"
      unit="xs:string"
      validity="xs:string"
    />*
  </wsag:CustomBusinessValue>*
</wsag:BusinessValueList>

Count and duration attributes are optional. If not specified, this CustomBusinessValue applies at each violation. Otherwise, it applies only if count violations occur in a duration interval of time.

The interpretation of every Penalty attribute is up to an external accounting module, but the intended meaning is:

  • type: kind of penalty (f.e: discount, service, terminate)
  • expression, unit: value of the penalty (f.e. discount of (50, euro), discount(100, %), service(sms))
  • validity: interval of time where the penalty is applied

Each time a violation is generated, the enforcement calculates if a business rule must be applied. If so, the corresponding Penalty is saved, and passed to the notification component.

The parsing of the business value list is performed in the IModelConverter. If a project wants to use a different xml structure, it can write the jaxb classes and a new BusinessValueListParser. The assignment of the parser to the model converter is done in the applicationContext.xml. The class to assign is defined in the configuration.properties file.

Project personalization

It is possible to program some plugins that can be injected in the code and that will personalize the behavior of the SLA software.

As it has been commented in the installation guide, several parameters can be configured through the configuration.properties. From these parameters we emphasize the following:

  • enforcement.constraintEvaluator.class: a class that must implement eu.atos.sla.evaluation.constraint.IConstraintEvaluator
  • enforcement.metricsRetriever.class: a class that must implement eu.atos.sla.monitoring.IMetricsRetriever
  • parser.json.agreement.class: a class that must implement eu.atos.sla.parser.IParser<eu.atos.sla.parser.data.wsag.Agreement>
  • parser.json.template.class: a class that must implement eu.atos.sla.parser.IParser<eu.atos.sla.parser.data.wsag.Template>
  • parser.xml.agreement.class: a class that must implement eu.atos.sla.parser.IParser<eu.atos.sla.parser.data.wsag.Agreement>
  • parser.xml.template.class: a class that must implement eu.atos.sla.parser.IParser<eu.atos.sla.parser.data.wsag.Template>

These classes are used for the enforcement and for parsing the serialized string into something wsag compliant.

The implementation of these classes should be done in the sla-personalization project. It is an empty code project that already imports the sla-enforcement, sla-repository and sla-tools, and is included in the war the has to be deployed. The pom.xml should be touched when a library is used that is not initially contemplated.

###Enforcement###

TODO roman

####Notifications It is possible to, once results from the enforcement task have been recorded, to notify a class.

The code must be included in the sla-personalization project. The new clas must implement the IAgreementEnforcementNotifier interface. It is defined as:

public interface IAgreementEnforcementNotifier {

	void onFinishEvaluation(IAgreement agreement, Map<IGuaranteeTerm, GuaranteeTermEvaluationResult>  guaranteeTermEvaluationMap);
}

The only existing method must be implemented by the class. It will receive the same data as it is recorded in the database. It will receive the agreement, from where the agreementId or any other data can be retrieve. An the guaranteeTermEvaluationMap that has a type of GuaranteeTermEvaluationResult which is implemented as:

public interface GuaranteeTermEvaluationResult {
	
	List<IViolation> getViolations();
	List<? extends ICompensation> getCompensations();
}

The information of the violations and compensations can be retrieved.

###Parsing###

To implement a new parser, it doesn't matter if it's for and Template, an Agreement, for JSON or XML format, the class must extend from the eu.atos.sla.parser.IParser.

public interface IParser <T> {
	/*
   	* getWsagObject receives in serializedData the object information in an xml, json or any other format 
   	* and must return the T object (eu.atos.sla.parser.data.wsag.Agreement or an 
	* eu.atos.sla.parser.data.wsag.Template)
	*/
	public T getWsagObject(String serializedData) throws ParserException;

	/*
	* getWsagAsSerializedData receives in serializedData the object information in an xml, json or any other 
	* format and  must return information following and xml in wsag standard.
	*/	
	public String getWsagAsSerializedData(String serializedData) throws ParserException;

   /*
    * getSerializedData receives in wsagSerialized the information in wsag standard as it was generated with 
	* the getWsagAsSerializedData method and returns it in a xml, json or any other format
	*/
	public String getSerializedData(String wsagSerialized) throws ParserException;
}

Depending if it is configured in parser.json or parser.xml the serializedData variable from the methods will be in json or in xml format.

###Build the code### To build the source code:

  • open a command window
  • go to the root directory (the same level where you have the configuration.properties.sample file)
  • type: mvn -Dmaven.test.skip clean install

The war file should be copied automatically to the tomcat.directory folder defined in the configuration.properties file.

Appendix 1: Sequence diagrams source

Periodic agreement enforcement

title Agreement enforcement (periodic execution)

Worker->EnforcementTask: new Thread()
EnforcementTask->+AgreementEnforcement: enforce()
AgreementEnforcement->MetricsRetriever: {m1..mn} = getMetrics({metrickey1..metrickeyn})
AgreementEnforcement->AgreementEvaluator: {v1p1..vnpn} = evaluate(mi)

AgreementEnforcement->EnforcementService: save(v1p1..vnpn)
AgreementEnforcement-->-EnforcementTask:

note right of AgreementEnforcement: v=violations;p=penalties

On demand agreement enforcement

title Agreement enforcement (on demand execution)

MetricsReceiver->+AgreementEnforcement: enforce({m1..mn})
AgreementEnforcement->AgreementEvaluator: {v1p1..vnpn} = evaluate(mi)

AgreementEnforcement->EnforcementService: save({v1p1..vnpn})
AgreementEnforcement-->-MetricsReceiver:

note right of AgreementEnforcement: v=violations;p=penalties

Agreement evaluation

title Agreement Evaluation

note left of AgreementEvaluator: input: agreement, mi
note right of AgreementEvaluator: v=violations;p=penalties
AgreementEvaluator->+GuaranteeTermEvaluator: vp += eval(mi)
GuaranteeTermEvaluator->+ServiceLevelEvaluator: vi=eval(mi)
ServiceLevelEvaluator->BreachRepository: b_old = getByTimeRange()
ServiceLevelEvaluator->ConstraintEvaluator: b_new = getBreaches(servicelevel, mi)
note right of ServiceLevelEvaluator: calculate violations
ServiceLevelEvaluator->BreachRepository: saveBreaches(b_new)
ServiceLevelEvaluator-->-GuaranteeTermEvaluator:
GuaranteeTermEvaluator->BusinessValuesEvaluator: pi=eval(vi)
GuaranteeTermEvaluator-->-AgreementEvaluator:

note left of AgreementEvaluator: output:{v1p1..vnpn}