tdlib-session-container/src/main/java/it/tdlight/reactiveapi/KafkaConsumer.java

132 lines
5.4 KiB
Java

package it.tdlight.reactiveapi;
import static java.lang.Math.toIntExact;
import it.tdlight.common.Init;
import it.tdlight.common.utils.CantLoadLibrary;
import java.time.Duration;
import java.util.HashMap;
import java.util.Map;
import java.util.logging.Level;
import java.util.regex.Pattern;
import org.apache.kafka.clients.consumer.CommitFailedException;
import org.apache.kafka.clients.consumer.ConsumerConfig;
import org.apache.kafka.common.errors.RebalanceInProgressException;
import org.apache.kafka.common.record.TimestampType;
import org.apache.kafka.common.serialization.IntegerDeserializer;
import org.apache.logging.log4j.LogManager;
import org.apache.logging.log4j.Logger;
import org.jetbrains.annotations.NotNull;
import org.jetbrains.annotations.Nullable;
import reactor.core.publisher.Flux;
import reactor.core.publisher.SignalType;
import reactor.kafka.receiver.KafkaReceiver;
import reactor.kafka.receiver.ReceiverOptions;
import reactor.util.retry.Retry;
public abstract class KafkaConsumer<K> {
private static final Logger LOG = LogManager.getLogger(KafkaConsumer.class);
private final KafkaParameters kafkaParameters;
public KafkaConsumer(KafkaParameters kafkaParameters) {
this.kafkaParameters = kafkaParameters;
}
public KafkaReceiver<Integer, K> createReceiver(@NotNull String groupId, @Nullable Long userId) {
try {
Init.start();
} catch (CantLoadLibrary e) {
LOG.error("Can't load TDLight library", e);
throw new RuntimeException(e);
}
Map<String, Object> props = new HashMap<>();
props.put(ConsumerConfig.BOOTSTRAP_SERVERS_CONFIG, kafkaParameters.bootstrapServers());
props.put(ConsumerConfig.CLIENT_ID_CONFIG, kafkaParameters.clientId() + (userId != null ? ("_" + userId) : ""));
props.put(ConsumerConfig.GROUP_ID_CONFIG, groupId);
props.put(ConsumerConfig.KEY_DESERIALIZER_CLASS_CONFIG, IntegerDeserializer.class);
props.put(ConsumerConfig.VALUE_DESERIALIZER_CLASS_CONFIG, getChannelName().getDeserializerClass());
props.put(ConsumerConfig.AUTO_OFFSET_RESET_CONFIG, "latest");
props.put(ConsumerConfig.MAX_POLL_INTERVAL_MS_CONFIG, toIntExact(Duration.ofMinutes(5).toMillis()));
if (isQuickResponse()) {
props.put(ConsumerConfig.HEARTBEAT_INTERVAL_MS_CONFIG, "5000");
props.put(ConsumerConfig.SESSION_TIMEOUT_MS_CONFIG, "10000");
} else {
props.put(ConsumerConfig.HEARTBEAT_INTERVAL_MS_CONFIG, "10000");
props.put(ConsumerConfig.SESSION_TIMEOUT_MS_CONFIG, "30000");
props.put(ConsumerConfig.FETCH_MIN_BYTES_CONFIG, "1048576");
props.put(ConsumerConfig.FETCH_MAX_WAIT_MS_CONFIG, "100");
}
ReceiverOptions<Integer, K> receiverOptions = ReceiverOptions.create(props);
Pattern pattern;
if (userId == null) {
pattern = Pattern.compile("tdlib\\." + getChannelName() + "\\.\\d+");
} else {
pattern = Pattern.compile("tdlib\\." + getChannelName() + "\\." + userId);
}
ReceiverOptions<Integer, K> options = receiverOptions
.subscription(pattern)
.addAssignListener(partitions -> LOG.debug("onPartitionsAssigned {}", partitions))
.addRevokeListener(partitions -> LOG.debug("onPartitionsRevoked {}", partitions));
return KafkaReceiver.create(options);
}
public abstract KafkaChannelName getChannelName();
public abstract boolean isQuickResponse();
protected Flux<Timestamped<K>> retryIfCleanup(Flux<Timestamped<K>> eventFlux) {
return eventFlux.retryWhen(Retry
.backoff(Long.MAX_VALUE, Duration.ofMillis(100))
.maxBackoff(Duration.ofSeconds(5))
.transientErrors(true)
.filter(ex -> ex instanceof RebalanceInProgressException)
.doBeforeRetry(s -> LOG.warn("Rebalancing in progress")));
}
protected Flux<Timestamped<K>> retryIfCommitFailed(Flux<Timestamped<K>> eventFlux) {
return eventFlux.retryWhen(Retry
.backoff(10, Duration.ofSeconds(1))
.maxBackoff(Duration.ofSeconds(5))
.transientErrors(true)
.filter(ex -> ex instanceof CommitFailedException)
.doBeforeRetry(s -> LOG.warn("Commit cannot be completed since the group has already rebalanced"
+ " and assigned the partitions to another member. This means that the time between subsequent"
+ " calls to poll() was longer than the configured max.poll.interval.ms, which typically implies"
+ " that the poll loop is spending too much time message processing. You can address this either"
+ " by increasing max.poll.interval.ms or by reducing the maximum size of batches returned in poll()"
+ " with max.poll.records.")));
}
public Flux<Timestamped<K>> consumeMessages(@NotNull String subGroupId, long userId) {
return consumeMessagesInternal(subGroupId, userId);
}
public Flux<Timestamped<K>> consumeMessages(@NotNull String subGroupId) {
return consumeMessagesInternal(subGroupId, null);
}
private Flux<Timestamped<K>> consumeMessagesInternal(@NotNull String subGroupId, @Nullable Long userId) {
return createReceiver(kafkaParameters.groupId() + "-" + subGroupId, userId)
.receive()
.log("consume-messages" + (userId != null ? "-" + userId : ""),
Level.FINEST,
SignalType.REQUEST,
SignalType.ON_NEXT,
SignalType.ON_ERROR,
SignalType.ON_COMPLETE
)
//.doOnNext(result -> result.receiverOffset().acknowledge())
.map(record -> {
if (record.timestampType() == TimestampType.CREATE_TIME) {
return new Timestamped<>(record.timestamp(), record.value());
} else {
return new Timestamped<>(1, record.value());
}
})
.transform(this::retryIfCleanup)
.transform(this::retryIfCommitFailed);
}
}