i.e. when remote target is configured with --sync mode in `mc admin bucket remote add` does not apply to `DELETE` operations. The version being deleted on the source cluster needs to maintain state and ensure that the operation is mirrored to the target cluster prior to completing on the source object version. Since this needs to account for the target cluster availability and the need to serialize concurrent DELETE operations on different versions of the same object during multi DELETE operations,...

    public T get() {
      // Another variant of Double Checked Locking.
      //
      // We use two volatile reads. We could reduce this to one by
      // putting our fields into a holder class, but (at least on x86)
      // the extra memory consumption and indirection are more
      // expensive than the extra volatile reads.
      long nanos = expirationNanos;
      long now = System.nanoTime();
      if (nanos == 0 || now - nanos >= 0) {

				r.Close()
				if err == nil {
					response := time.Since(t)
					ttfb := time.Since(*fbr.t)
					// Only capture success criteria - do not
					// have to capture failed reads, truncated
					// reads etc.
					atomic.AddUint64(&totalBytesRead, uint64(n))
					mu.Lock()
					downloadTimes = append(downloadTimes, response)
					downloadTTFB = append(downloadTTFB, ttfb)
					mu.Unlock()
				}

      }

      else -> break@loop
    }
  }
  return commaFound
}

private fun Buffer.startsWith(prefix: Byte): Boolean = !exhausted() && this[0] == prefix

/**
 * Reads a double-quoted string, unescaping quoted pairs like `\"` to the 2nd character in each
 * sequence. Returns the unescaped string, or null if the buffer isn't prefixed with a
 * double-quoted string.
 */

		header: []byte{0x42, 0x5a, 'h'},
		f:      formatBZ2,
	},
}

type untarOptions struct {
	ignoreDirs bool
	ignoreErrs bool
	prefixAll  string
}

// disconnectReader will ensure that no reads can take place on
// the upstream reader after close has been called.
type disconnectReader struct {
	r  io.Reader
	mu sync.Mutex
}

func (d *disconnectReader) Read(p []byte) (n int, err error) {

- On Kubernetes setups, the statefulset specification needs to be modified by changing the command line input for the MinIO container. Once the relevant changes are done, proceed to execute `kubectl apply -f statefulset.yaml`.

_2019-10-06_

 *  Fix: When closing a canceled HTTP/2 stream, don't send the `END_STREAM` flag. This could cause
    the server to incorrectly interpret the stream as having completed normally. This is most useful
    when a request body needs to cancel its own call.


## Version 4.2.1

_2019-10-02_

 *  Fix: In 4.1.0 we introduced a performance regression that prevented connections from being

// List of some generic middlewares which are applied for all incoming requests.
var globalMiddlewares = []mux.MiddlewareFunc{
	// set x-amz-request-id header and others
	addCustomHeadersMiddleware,
	// The generic tracer needs to be the first middleware to catch all requests
	// returned early by any other middleware (but after the middleware that
	// sets the amz request id).
	httpTracerMiddleware,

		nsLk.lockMapMutex.Unlock()

		// To trigger the race:
		// 1) lk3 or lk4 need to advance and increment the ref on the existing resource,
		//    successfully acquiring the lock.
		// 2) lk2 then needs to advance and remove the resource from lockMap.
		// 3) lk3 or lk4 (whichever didn't execute in step 1) then executes and creates
		//    a new entry in lockMap and acquires a lock for the same resource.

		<-lk2ch

        // Use single channel for small reads
        normalRead(buffer, offset, length);
        return;
    }
    
    // Split large read across multiple channels
    ChannelManager channelManager = session.getChannelManager();
    List<ChannelInfo> channels = channelManager.getHealthyChannels();
    int channelCount = Math.min(channels.size(), 4);  // Max 4 parallel reads
    
    int chunkSize = length / channelCount;