0
\$\begingroup\$

I have two consumers and each consumer has their own validation logic entirely different from one other so I am using two validator class for that, one for each other.

  • ConsumerA, for this consumer, I am using ValidatorA.
  • ConsumerA, for this consumer, I am using ValidatorB.

ConsumerA class

public class ConsumerA extends Consumer {
  .....

  @Override
  protected void run(String consumerName, Properties consumerProps) {
    consumer = new KafkaConsumer<>(consumerProps);
    consumer.subscribe(getTopicsBasisOnConsumerName());

    try {
      while (!closed.get()) {
        ConsumerRecords<byte[], byte[]> records = consumer.poll(2000);
        for (ConsumerRecord<byte[], byte[]> record : records) {
          GenericRecord genericRecord = decoder.decode(record.value());
          ValidatorA validation = new ValidatorA(consumerName, genericRecord);
          if (!validation.isValid()) {
            logger.logError("dropping records. payload= ", genericRecord);
            continue;
          }

          // setting variables to DataPacket, do I need this class at all?
          DataPacket packet = new DataPacket.Builder(genericRecord).setClientId(validation.getClientId())
              .setDeviceId(validation.getDeviceId()).setPayId(validation.getPayId()).setHolder(validation.getHolder())
              .setOldTimestamp(validation.getOldTimestamp()).setNewTimestamp(validation.getNewTimestamp()).build();   

          Processor.getInstance().execute(packet);

        }
      }
    } catch (Exception ex) {
      System.out.println("error= " + ex);
    }
  }
}

ConsumerB class

public class ConsumerB extends Consumer {
  .....

  @Override
  protected void run(String consumerName, Properties consumerProps) {
    consumer = new KafkaConsumer<>(consumerProps);
    consumer.subscribe(getTopicsBasisOnConsumerName());

    try {
      while (!closed.get()) {
        ConsumerRecords<byte[], byte[]> records = consumer.poll(2000);
        for (ConsumerRecord<byte[], byte[]> record : records) {
          GenericRecord genericRecord = decoder.decode(record.value());
          ValidatorB validation = new ValidatorB(consumerName, genericRecord);
          if (!validation.isValid()) {
            logger.logError("dropping records. payload= ", genericRecord);
            continue;
          }

          // setting variables to DataPacket, do I need this class at all?
          DataPacket packet = new DataPacket.Builder(genericRecord).setType(validation.getType())
              .setDatumId(validation.getDatumId()).setItemId(validation.getItemId()).setOldTimestamp(validation.getOldTimestamp())
              .setNewTimestamp(validation.getNewTimestamp()).build();     

          Processor.getInstance().execute(packet);
        }
      }
    } catch (Exception ex) {
      System.out.println("error= " + ex);
    }
  }
}

Below are my corresponding validator classes:

ValidatorA class This class is called by ConsumerA, basically it's isValid method for validation.

public class ValidatorA extends Validator {
  private static final Logger logger = Logger.getInstance(ValidatorA.class);    
  private final String consumerName;
  private final GenericRecord genericRecord;
  private final Long oldTimestamp;
  private final Long newTimestamp;
  private final String clientId;
  private final String deviceId;
  private final Integer payId;
  private final Map<String, String> holder;

  public ValidatorA(String consumerName, GenericRecord genericRecord) {
    this.consumerName = consumerName;
    this.genericRecord = genericRecord
    this.oldTimestamp = (Long) DataUtils.parse(genericRecord, "oldTimestamp");
    this.newTimestamp = (Long) DataUtils.parse(genericRecord, "newTimestamp");  
    this.clientId = (String) DataUtils.parse(genericRecord, "clientId");
    this.deviceId = (String) DataUtils.parse(genericRecord, "deviceId");
    this.payId = (Integer) DataUtils.parse(genericRecord, "payId");
    this.holder = (Map<String, String>) DataUtils.parse(genericRecord, "holder");
  }

  @Override
  public boolean isValid() {
    return isValidClientIdDeviceId() && isValidPayId() && isValidHolder();
  }

  private boolean isValidHolder() {
    if (MapUtils.isEmpty(holder)) {
      logger.logError("invalid holder.");
      return false;
    }
    return true;
  }

  private boolean isValidPayId() {
    if (payId == null) {
      logger.logError("invalid payId.");
      return false;
    }
    return true;
  }

  private boolean isValidClientIdDeviceId() {
    if (Strings.isNullOrEmpty(clientId) && Strings.isNullOrEmpty(deviceId)) {
      logger.logError("invalid clientId and deviceId.");
      return false;
    }
    return true;
  }

    // getter and toString method here
}

ValidatorB class This class is called by ConsumerB, basically it's isValid method for validation.

public class ValidatorB extends Validator {
  private static final Logger logger = Logger.getInstance(ValidatorB.class);    
  private final String consumerName;
  private final GenericRecord genericRecord;
  private final Long oldTimestamp;
  private final Long newTimestamp;    
  private final String type;
  private final String datumId;
  private final String itemId;

  public ValidatorB(String consumerName, GenericRecord genericRecord) {
    this.consumerName = consumerName;
    this.genericRecord = genericRecord
    this.oldTimestamp = (Long) DataUtils.parse(genericRecord, "oldTimestamp");
    this.newTimestamp = (Long) DataUtils.parse(genericRecord, "newTimestamp");      
    this.type = (String) DataUtils.parse(genericRecord, "type");
    this.datumId = (String) DataUtils.parse(genericRecord, "datumId");
    this.itemId = (Integer) DataUtils.parse(genericRecord, "itemId");
  }

  @Override
  public boolean isValid() {
    return isValidType() && isValidDatumId() && isValidItemId();
  }

  private boolean isValidType() {
    if (Strings.isNullOrEmpty(type)) {
      logger.logError("invalid type is coming.");
      return false;
    }
    return true;
  }  

  private boolean isValidDatumId() {
    if (Strings.isNullOrEmpty(datumId)) {
      logger.logError("invalid datumId is coming.");
      return false;
    }
    return true;
  }   

  private boolean isValidItemId() {
    if (Strings.isNullOrEmpty(itemId)) {
      logger.logError("invalid itemId is coming.");
      return false;
    }
    return true;
  }

    // getter and toString method here
}

Validator class

public abstract class Validator {
  public abstract boolean isValid();
}

Now in my Processor class in the execute method, I am using DataPacket passed from both the consumer to extract all these variables which have been set, do some processing by using those variables. And later on we will get some new variables which I am setting it by cloning the old packet builder.

private void execute(DataPacket packet) {

    // extract all the fields from this packet
    // do some processing using them
    // now clone the packet and set processed variables.
    String type = packet.getType();
    String clientId = packet.getClientId();

    // do some processing

    String schemaId = ....;
    Schema schema = ....;

    // cloning old packet and setting new variables
    DataPacket clonedPacket = new DataPacket.Builder(packet).setSchemaId(schemaId).setSchema(schema).build();

    // send clonedPacket to some other system which will send to database

}

My question is - Do I need this DataPacket at all when I already have ValidatorA, ValidatorB class? DataPacket class just contains all the variables combine from both the validator class and I am using DataPacket builder class just to set data in both the consumers and then use it in execute method of Processor class, and then clone the old builder to make a new builder object again by setting some new variables.

Is there any better way to do this thing? Looks like I can get rid of DataPacket but not sure how.

\$\endgroup\$
3
  • \$\begingroup\$ This question looks a bit foobarish. Do "A" and "B" mean something, and if so, what? \$\endgroup\$ Dec 21 '16 at 6:35
  • \$\begingroup\$ They are kafka consumers. I mentioned in the question they are consumers but forgot to mention kafka consumers. Will add that. \$\endgroup\$
    – david
    Dec 21 '16 at 6:36
  • \$\begingroup\$ "A" and "B" still sound like rather poor names, though, for whatever you are trying to do. \$\endgroup\$ Dec 21 '16 at 6:38
5
\$\begingroup\$

The question is rather vague, I'll assume you haven't taken the time to step back and look at your overall architecture.

Still, a lot can be said already, so here's my understanding (you should explain this in your Question, a code dump isnt going to provide much answers):

  • ConsumerA and ConsumerB run constantly, and in parallel
  • They produce some DataPacket
  • Each DataPacketis validated (ConsumerA uses ValidatorA etc.)
  • Each DataPacketis sent to a singleton Processor class 's process method (Processor class is not provided but process method is)
  • The process() method clones the DataPacket and modifies it.
  • The cloned DataPacket is sent to 'some other system'

You're bothered because ValidatorX class and DataPacket class share some internal representations (to which extent, that not obvious because you did not provide DataPacket class).

Read your own code carefully, it tells you what's wrong

ValidatorB validation = new ValidatorB(consumerName, genericRecord);

You define two Validator classes. However, you name their instances 'validation'. That is because the Validator class is wrongly named and you felt it when you had to use it: it does not validate some input you may throw at it, it validates one data when it is built (which is already frowned upon), then gives you the result upon request (the real job is in the DataUtils.parse calls in the Constructor).

I expect a Validator to be stateless (or lightly prameterized if it helps reuse) and be used like this:

public class ConsumerA {
    private static final ValidatorA = new Validator();
    (...)
    protected void run(String consumerName, Properties consumerProps) {
        for(...)
            (...)
            if(validator.isValid(consumerName, genericRecord)){
                 // Do something
            }
        }
    }
}

This way the instance is reused, no useless internal state is kept, etc. That already solves what was bothering you (Validator and DataPAcket sharing internal state), plus it saves a bunch of object creation/destroy.

Since your ValidatorX classes will still need to pass around the data in a coherent manner internally, you could even make such a Validator work on a DataPacket internally, and return null when validation fails or said DataPacket when validation passes. But since I don't have DataPacket class, and I don't know if you're allowed to modify it if need be, I can't be adamant here.

Consumers share duplicate code.

You could regroup them in an abstract class and two sub-classes with just the differing bits.

Or (much better) use Strategy pattern on a single GenericRecordConsumer, which you build by providing it with a specific Validator. Note, you can only do that if you implemented a reusable Validator class and not a disposable Validation class (as mentioned above).

public class GenericRecordConsumer extends Consumer {

  public GenericRecordConsumer(Validator validator){
    this.validator = validator;
  }
  .....

  @Override
  protected void run(String consumerName, Properties consumerProps) {
    consumer = new KafkaConsumer<>(consumerProps);
    consumer.subscribe(getTopicsBasisOnConsumerName());
    try {
      while (!closed.get()) {
        ConsumerRecords<byte[], byte[]> records = consumer.poll(2000);
        for (ConsumerRecord<byte[], byte[]> record : records) {
          GenericRecord genericRecord = decoder.decode(record.value());
          if (!this.validator.isValid(consumerName, genericRecord)) {
            logger.logError("dropping records. payload= ", genericRecord);
            continue;
          }
          // setting variables to DataPacket, do I need this class at all?
          DataPacket packet = new DataPacket.Builder(genericRecord).setClientId(validation.getClientId())
              .setDeviceId(validation.getDeviceId()).setPayId(validation.getPayId()).setHolder(validation.getHolder())
              .setOldTimestamp(validation.getOldTimestamp()).setNewTimestamp(validation.getNewTimestamp()).build();
          Processor.getInstance().execute(packet);
        }
      }
    } catch (Exception ex) {
      System.out.println("error= " + ex);
    }
  }
}

Then you can create them as easily as :

Consumer consumerA = new GenericRecordConsumer (new ValidatorA("consumerA"));
Consumer consumerB = new GenericRecordConsumer (new ValidatorB("consumerB"));

Much shorter, more extendable, zero code duplication. If you have other behaviour that change between the consumer versions (in your ellusive (...)), that code can have the same fate and be Strategy-fied.

Exception handling

Yup, don't do this:

try {
  while (!closed.get()) {
     (...)
  }
} catch (Exception ex) {
  System.out.println("error= " + ex);
}

First, System.out is a very bad exception handling mechanism (use a Logger).

Second, a consumer system running through a list of jobs continuously should have two sorts of exceptions:

  • Recoverable exceptions, which are caught, logged properly, and probably also stored in DB for the admins, etc. You may even do some processing to salvage the operation that failed. Ultimately, they do not interrupt the Consumer work loop.
  • Unrecoverable exceptions, which are of two kinds:
    • Unrecoverable exceptions of the Consumer (ConnectionToMyDatabaseIsLostException etc.), which are caught, logged, and trigger a special warning (email the admins, store in DB). Ultimately, they should interrupt the Consumer because its function is compromised somehow. Potentially, the rest of the system can still funtion, albeit without this particular consumer. You might have an admin panel with a 'revive Consumer' button for instance.
    • Unrecoverable exceptions which you know nothing about (OutOfMemoryException etc.). You can't tell why or when they appear, and you can do nothing about it. These should not be caught.

The for loop could look like:

try {
  while (!closed.get()) {
     try{
       (...)
     } catch(HarmlessException ex){
       log.WARN("blah", ex);
       continue; // Pursue the job
     }
  }
} catch (HarmfulExceptionThatKillsTheConsumer ex) {
  System.out.println("error= " + ex);
  OtherSystem.signalConsumerFailure(this);
  return; // Exit the Consumer work early
}
// Note I did not catch any other exception, which I wouldn't know what to do with

(of course I would actually extract the wole internal try/catch in its own method with a throws to clarify)

Data validation

This makes me cringe a bit:

this.oldTimestamp = (Long) DataUtils.parse(genericRecord, "oldTimestamp");
this.newTimestamp = (Long) DataUtils.parse(genericRecord, "newTimestamp");      
this.type = (String) DataUtils.parse(genericRecord, "type");
this.datumId = (String) DataUtils.parse(genericRecord, "datumId");
this.itemId = (Integer) DataUtils.parse(genericRecord, "itemId");

You have a single method, that can (apparently) return Objects of several classes. It's probably some kind of reflection applied on a bean-like Object. I'm not against reflection, it is elegant. But it shouldn't be an excuse to create a caryall method that forgets its general purpose.

Since this method seem to return the common types (Long, Integer, String) and not some esotheric custom objects list that would have to be extended as you go, it would be much more robust and elegant to have some parseInt, parseString and parseLong methods. This allows three things:

  • It ditches that cast (I hate casting. If there's a cast, there's a better way!)
  • It allows the parse method to be prepared for its expected type, and maybe do some more work on the input (Is integer positive? Is String empty? etc.)
  • It prevents some stupid mistakes like this.datumId = (Long) DataUtils.parse(genericRecord, "itemId"); where an Int is in fact (successfully) cast to Long and all hells come loose because it doesn't mean a thing business-wise.

Ties to external systems

Your comment :

// send clonedPacket to some other system which will send to database

Is rather vague. How is this sent? Is the DataPacket wholly sent, or just some fields? If this is not imposed by an external API, it might be improved but I can't tell.

Overall

Comments are good, sufficient and to the point

Variable naming is good and case is correct.

Log messages are a bit clumsy:

  logger.logError("invalid itemId is coming.");

Could be replaced by

logger.logError("Invalid ItemId provided. ItemId must not be null nor empty. " + this.consumerName + " cannot validate this record.");
\$\endgroup\$
0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.