src/java/com/twitter/search/earlybird/stats/EarlybirdRPCStats.java - main - githubmirror/the-algorithm

src/java/com/twitter/search/earlybird/stats/EarlybirdRPCStats.java

55 строк2 KB

twitter-team

Twitter Recommendation Algorithm

01 апр 2023, 01:36

01 апр 2023, 01:36ef4c5eb

package com.twitter.search.earlybird.stats;

import java.util.concurrent.TimeUnit;

import com.twitter.search.common.metrics.SearchCounter;
import com.twitter.search.common.metrics.SearchRateCounter;
import com.twitter.search.common.metrics.SearchRequestStats;

/**
 * SearchRequestStats with earlybird-specific additional stats.
 */
public final class EarlybirdRPCStats {
  private final SearchRequestStats requestStats;
  // Number of queries that were terminated early.
  private final SearchCounter earlyTerminatedRequests;

  // We do not count client error in the response error rate, but track it separately.
  private final SearchRateCounter responseClientErrors;

  public EarlybirdRPCStats(String name) {
    requestStats = SearchRequestStats.export(name, TimeUnit.MICROSECONDS, true, true);
    earlyTerminatedRequests = SearchCounter.export(name + "_early_terminated");
    responseClientErrors = SearchRateCounter.export(name + "_client_error");
  }

  public long getRequestRate() {
    return (long) (double) requestStats.getRequestRate().read();
  }

  public long getAverageLatency() {
    return (long) (double) requestStats.getTimerStats().read();
  }

  /**
   * Records a completed earlybird request.
   * @param latencyUs how long the request took to complete, in microseconds.
   * @param resultsCount how many results were returned.
   * @param success whether the request was successful or not.
   * @param earlyTerminated whether the request terminated early or not.
   * @param clientError whether the request failure is caused by client errors
   */
  public void requestComplete(long latencyUs, long resultsCount, boolean success,
                              boolean earlyTerminated, boolean clientError) {
    // We treat client errors as successes for top-line metrics to prevent bad client requests (like
    // malformed queries) from dropping our success rate and generating alerts.
    requestStats.requestComplete(latencyUs, resultsCount, success || clientError);

    if (earlyTerminated) {
      earlyTerminatedRequests.increment();
    }
    if (clientError) {
      responseClientErrors.increment();
    }
  }
}