Understanding the RED FLAGS engine

Previously we showed you the bird's eye view of the Red Flags architecture, now take a look at the engine itself, in details:

Control & data flow

Let's see how the engine works.

  1. Firstly, the main method is RedflagsEngineApp.main(). It handles special command line arguments that doesn't need the engine itself. If none of them were specified, it starts the Spring platform.
  2. Spring now takes the lead an initializes all components. The engine's first stage is RedflagsEngineBoot. It depends on RedflagsEngineConfig which will be filled with the configuration parameters.
  3. RedflagsEngineBoot calls ScopeProvider to select and instantiate the appropriate implementation of AbstractScope. Scopes are basically NoticeID iterators, you can find them in the hu.petabyte.redflags.engine.scope package.
  4. After that it creates and configures a TedInterface to be used by the engine later. TedInterface deals with the downloading from TED and also it manages the file cache.
  5. Then it calls GearLoader to load the appropriate gears (Notice processors).
  6. Lastly, RedflagsEngineBoot starts a new RedflagsEngineSession.
  7. RedflagsEngineSession receives the scope, the gears, and the thread count, and initializes the gears at first.
  8. Then it creates a thread pool, starts iterating over the notice IDs (using scope) and creates a GearTrain thread for each one of them.
  9. Lastly, it waits for all threads to finish.
  10. GearTrain objects are processor chains containing the gears. A chain is receiving an empty Notice object, gives it to the first gear, takes its result, gives it to the next gear and so on.
  11. Notices are not being collected, but there's an option for that in RedflagsEngineSession.

Gear trains

GearLoader reads the gear list from the redflags.engine.gears property. The default list is the following:


    # parse metadata
    - metadataParser
    #- cancelledNoticeFilter

    # skip non-Hungarian notices
    - countryFilter
    - originalLanguageFilter

    # download all tabs, and related notice tabs
    - archiver
    - docFamilyFetcher
    - docFamilyArchiver

    # skip further processing of old notices and tenders with non-classic dir.
    - publicationDateFilter
    - docFamilyDateFilter
    - directiveCorrector
    - directiveFilter

    # magic I. - parse document based on template
    - templateBasedDocumentParser
    - rawValueParser
    - frameworkAgreementParser
    - estimatedValueParser
    - renewableParser
    - countOfInvOpsParser
    - awardCriteriaParser
    - openingDateParser

    # magic II. - indicators
    - fwAgOneParticipantIndicator
    - fwAgFewParticipantsIndicator
    - fwAgLongDurationIndicator
    - fwAgHighEstimatedValueIndicator
    - contrDescCartellingIndicator
    - highEstimatedValueIndicator
    - totalQuantityHiDeltaIndicator
    - renewalOfContractIndicator
    - durationLongOrIndefiniteIndicator
    - persSitMissingCondIndicator
    - finAbMissingMinCondIndicator
    - finAbEquityCondIndicator
    - finAbRevenueCondExceedEstimValIndicator
    - finAbRevenueCondManyYearsIndicator
    - techCapMissingMinCondIndicator
    - techCapSingleContractRefCondIndicator
    - techCapExpertsExpCondManyYearsIndicator
    - techCapGeoCondIndicator
    - techCapRefCondExceedEstimValIndicator
    - techCapRefCondManyYearsIndicator
    - techCapEURefCondIndicator
    - procTypeAcceleratedIndicator
    - procTypeNegotiatedNoJustificationIndicator
    - countOfInvOpsLowIndicator
    - countOfInvOpsNoCondIndicator
    - awCritLacksIndicator
    - awCritPaymentDeadlineCondIndicator
    - awCritMethodMissingIndicator
    - deadlineIsTightIndicator
    - openingDateDiffersFromDeadlineIndicator
    - offerGuaranteeIsHighIndicator
    - finalValFarFromEstimValIndicator
    - procWithoutContractNoticeIndicator
    - numberOfOffersLowIndicator
    - unsuccessfulProcWithRiskIndicator
    - unsuccessfulProcWithoutInfo1Indicator
    - unsuccessfulProcWithoutInfo2Indicator
    - decisionDateDiffersFromOpeningDateIndicator
    - highFinalValueIndicator

    - contractingOrgInKMDBIndicator
    - winnerOrgInKMDBIndicator

    # output
    - H1FlagExporter
    - H2FlagExporter
    - mySQLExporter

    - stopGear

Gears are being passed to a GearTrain with a notice. GearTrain objects are Callable<Void> instances, their call method will go through the gears and pass them the notice object in their process methods. The first gear receives a Notice object which has only the notice ID in it.

Gear trains have a filter feature: when a gear returns null, the the cycle breaks, the processing stops at that point.

See the stopGear at the end. The StopGear returns null, so stops the processing. You should always have it at the end of your list, because if you specify a gear list which is shorter than the default list, it only overwrites the first part of the list, and the rest of the default list will be also in use.

When a gear throws an exception, it will be logged by GearTrain, but the cycle continues with the next gear in line.

GearTrains are called by RedflagsEngineSession on separate threads. The size of the thread pool can be configured by the threads property.

Gears in general

As I said before, gears are notice processors. They take an input notice and should return an output notice or null. Every gear is an implementation of AbstractGear. There is only one method that has to be implemented:

protected abstract Notice processImpl(Notice notice) throws Exception;

This method is called by the public final method process after a null-check of the input notice. The process method is called by the GearTrain.

Additionally there are two methods for gear initialization and finalization which can be overridden: beforeSession() and afterSession(). They are called by RedflagsEngineSession before and after calling the GearTrain. When a beforeSession method throws an exception, the session ends at that point.

It is recommended not to use @PostConstruct annotation or InitializingBean when you need initialization because in that ways, the gear is being initialized at the application startup. But when you use the beforeSession method, the initialization will only be performed when the gear is present in the gear list.

Gears must have the @Component annotation, this way GearLoader can instantiate them using Spring. All @Component in Spring are singletons. If you need separate instances of a gear for every notice (for example for thread-safety), you should add @Scope("prototype") annotation to your gear class.

By default your the bean name of your gear is the class name with the first letter transformed to lower case. You can override this by adding the default argument to the @Component annotation: @Component("CustomName"). The bean name is used in the gear list.

Property values can be injected into gears, this way they are able to be configured out of the box. There is two way to do that:

Also, you can use other gears in a gear, this way:

private MyOtherGear otherGear;

But make sure you call beforeSession and afterSession of the other gear if needed.

Here's an example gear:

@ConfigurationProperties(prefix = "mygear")
// @Scope("prototype")                               // in case you need separate instances
public class MyGear extends AbstractGear {           // "myGear" in the gear list

  private String  strOption = "default value";       // mygear.strOption
  private int     intOption = 42;                    // mygear.intOption
  private Pattern myPattern = Pattern.compile(".*"); // mygear.myPattern
  //private @Autowired OtherGear g;                  // in case you need another gear

  protected Notice processImpl(Notice notice) throws Exception {
    // do something with notice and fields
    return notice;

We can categorize gears by their main tasks:

Let's discuss their features.

Downloader gears

Their task is to fetch HTML documents for a notice by either downloading them from TED and placing in the cache, or by reading directly from the cache if they already downloaded. They should be somewhere at the beginning of the list.

The mechanism mentioned above is implemented by TedInterface.

List of Red Flags' download gears

Parser gears

Parser gears process the downloaded data and fill in the POJOs. Like downloader gears, parser gears also doesn't have an abstract superclass to implement.

Some built-in parsers use downloader gears to fetch HTML content, and some of them

List of Red Flags' parser gears

Filter gears

A filter gear examines the current notice and decides whether it can reach the next processor or not. Conventionally these gears should implement the AbstractFilter class and its only one method: boolean accept(Notice). Example:

public class MyFilter extends AbstractFilter {

  public boolean accept(Notice notice) throws Exception {
    return condition(notice);

List of Red Flags' filter gears

Indicator gears

A task of an indicators is to examine something in a notice and generate one or zero flag. Indicator classes should implement AbstractIndicator and its method IndicatorResult flag(Notice).

All indicators have properties called category and weight, they can be used later in other gears or in the webapp, for example to differentiate displaying of flags. These values are initialized in beforeSession from package.ClassName.category and package.ClassName.weight properties. The default category is null, the default weight is 1.0.

An IndicatorResult object contains the following things:

The flag type can be one of these enum value: MISSING_DATA, IRRELEVANT_DATA, NO_FLAG, FLAG. NO_FLAG and FLAG is trivial, I think. MISSING_DATA can be used when some information needed by the indicator is missing from the notice. IRRELEVANT_DATA result type can be used to indicate the notice is not from the type that the indicator wants to process. Only the results with FLAG type will be displayed on the website.

The description is handled specially by the webapp. First of all it must be a message label identifier, e.g. "myindicator.desc". But you can pass actual information as parameters to be displayed in the webapp. Here's how:

Fortunately AbstractIndicator has helper methods to generate description strings, and also complete IndicatorResult objects:

| ----------------------------------------------------------|--- label(String label, String... args) | generates a description string where the label identifier will be flag.IndicatorId.specifiedLabel and parameters will be the param=value pairs given in args irrelevantData() and missingData() | generates IndicatorResult objects with the corresponding flag types returnFlag() | generates a result with label flag.IndicatorId.info and type FLAG, category and weight will be copied from the current indicator class returnFlag(String label, String... args) | lets you replace info to another label and add parameters returnFlag(double weight, String lable, String... args) | lets you override the indicator's weight too

IndicatorId is generated from the name of the direct parent package and the simple class name, e.g. hu.petabyte.redflags.engine.gear.indicator.pl.AwCritLacksIndicator will have pl.AwCritLacksIndicator as its id. This way indicators of different languages/countries can be separated in language files too.

If you want to return NO_FLAG you can simply return null from your flag(Notice) method.

Let's see an example indicator:

package mypackage.lang;

import hu.petabyte.redflags.engine.gear.indicator.AbstractIndicator;
import hu.petabyte.redflags.engine.model.IndicatorResult;
import hu.petabyte.redflags.engine.model.Notice;
import java.util.regex.Pattern;
import org.springframework.stereotype.Component;

public class MyIndicator extends AbstractIndicator {

  protected IndicatorResult flagImpl(Notice notice) {
    if (!dataExistsInNotice(notice)) {
      return missingData();                            // MISSING_DATA
    } else if (condition1(notice)) {
      return returnFlag();                             // FLAG, "flag.lang.MyIndicator.info"
    } else if (condition2(notice)) {
      return returnFlag("info2");                      // FLAG, "flag.lang.MyIndicator.info2"
      // or with parameters:
      // return returnFlag("info2", "p1=v1", "p2=v2");
    return null;                                       // NO_FLAG

There are 2 additional abstract classes, AbstractTD3CIndicator and AbstractTD7Indicator. They override the flag method by testing the document type and calling the abstract flagImpl method (same signature as flag) if the notice is contract notice (TD-3 or TD-C) or contract award notice (TD-7).

List of Red Flags' indicator gears

Exporter gears

Exporter gears are usually on at the end of the list, their task is to produce an output from the parsed and generated data, such as CSV files or relational databases. Conventionally all exporter gears should implement AbstractExporter and its abstract method: void export(Notice). Example:

@Scope("prototype") // you probably need only one of this
public class MyExporter extends AbstractExporter {

  public void export(Notice notice) {
    // write out notice

  public void beforeSession() throws Exception {
    // initialize output

  public void afterSession() throws Exception {
    // finalize output

List of Red Flags' exporter gears