*       allocating garbage).
 * </ul>
 *
 * <p>Disadvantages compared to {@code int[]}:
 *
 * <ul>
 *   <li>Memory footprint has a fixed overhead (about 24 bytes per instance).
 *   <li><i>Some</i> construction use cases force the data to be copied (though several construction
 *       APIs are offered that don't).

    test to timeout but it could be useful when you need to execute a single
    test or more in general your filtered/selected tests have a very low
    execution time and the sharding
    [could create an overhead on the test execution](https://github.com/bazelbuild/bazel/issues/2113#issuecomment-264054799).

2.  Using [Docker](https://www.docker.com) and TensorFlow's CI scripts.

    ```bash

  ) : Sink {
    /**
     * Buffer of outgoing data. This batches writes of small writes into this sink as larges frames
     * written to the outgoing connection. Batching saves the (small) framing overhead.
     */
    private val sendBuffer = Buffer()

    /** Trailers to send at the end of the stream. */
    var trailers: Headers? = null

    var closed: Boolean = false

  // There are a few design constraints to consider
  // * We want to be responsive to small timeouts, unpark() has non trivial latency overheads (I
  //   have observed 12 micros on 64-bit linux systems to wake up a parked thread). So if the
  //   timeout is small we shouldn't park(). This needs to be traded off with the cpu overhead of
  //   spinning, so we use SPIN_THRESHOLD_NANOS which is what AbstractQueuedSynchronizer uses for
  //   similar purposes.

   * non-static, and is initialized/removed frequently - this causes churn in the Thread specific
   * hashmaps. Using a static ThreadLocal to avoid that overhead would mean that different
   * ExecutionSequencer objects interfere with each other, which would be undesirable, in addition
   * to increasing the memory footprint of every thread that interacted with it. In order to release

### Major Themes

  * <strong>Significant scale improvements</strong>. Increased cluster scale by 400% to 1000 nodes with 30,000 pods per cluster.
Kubelet supports 100 pods per node with 4x reduced system overhead.
  * <strong>Simplified application deployment and management. </strong>
     * Dynamic Configuration (ConfigMap API in the core API group) enables application

  // There are a few design constraints to consider
  // * We want to be responsive to small timeouts, unpark() has non trivial latency overheads (I
  //   have observed 12 micros on 64-bit linux systems to wake up a parked thread). So if the
  //   timeout is small we shouldn't park(). This needs to be traded off with the cpu overhead of
  //   spinning, so we use SPIN_THRESHOLD_NANOS which is what AbstractQueuedSynchronizer uses for
  //   similar purposes.

    /**
     * The smallest message that will be compressed. We use 1024 because smaller messages already
     * fit comfortably within a single ethernet packet (1500 bytes) even with framing overhead.
     *
     * For tests this must be big enough to realize real compression on test messages like
     * 'aaaaaaaaaa...'. Our tests check if compression was applied just by looking at the size if
     * the inbound buffer.

	paths = make([]string, len(elem))
	for i, e := range elem {
		paths[i] = pathJoin(prefix, e)
	}
	return paths
}

// string concat alternative to s1 + s2 with low overhead.
func concat(ss ...string) string {
	length := len(ss)
	if length == 0 {
		return ""
	}
	// create & allocate the memory in advance.
	n := 0
	for i := range length {
		n += len(ss[i])
	}

	if r.ContentLength <= 0 {
		writeErrorResponse(ctx, w, errorCodes.ToAPIErr(ErrMissingContentLength), r.URL)
		return
	}

	// The max. XML contains 100000 object names (each at most 1024 bytes long) + XML overhead
	const maxBodySize = 2 * 100000 * 1024

	if r.ContentLength > maxBodySize {
		writeErrorResponse(ctx, w, errorCodes.ToAPIErr(ErrEntityTooLarge), r.URL)
		return
	}