### Memória do servidor

Por exemplo, se seu código carrega um modelo de Machine Learning com **1 GB de tamanho**, quando você executa um processo com sua API, ele consumirá pelo menos 1 GB de RAM. E se você iniciar **4 processos** (4 trabalhadores), cada um consumirá 1 GB de RAM. Então, no total, sua API consumirá **4 GB de RAM**.

- Since golang 1.17 both net.ParseIP and net.ParseCIDR rejects leading zeros in the dot-decimal notation of IPv4 addresses,
  Kubernetes will keep allowing leading zeros on IPv4 address to not break the compatibility.
  IMPORTANT: Kubernetes interprets leading zeros on IPv4 addresses as decimal, users must not rely on parser alignment to not being impacted by the associated security advisory:

## Version 2.5.0

_2015-08-25_

 *  **Timeouts now default to 10 seconds.** Previously we defaulted to never
    timing out, and that was a lousy policy. If establishing a connection,
    reading the next byte from a connection, or writing the next byte to a
    connection takes more than 10 seconds to complete, you’ll need to adjust
    the timeouts manually.

		opts.VersionID = oi.VersionID
	}
	tags := newLifecycleAuditEvent(src, lcEvent).Tags()
	if lcEvent.Action.DeleteRestored() {
		// delete locally restored copy of object or object version
		// from the source, while leaving metadata behind. The data on
		// transitioned tier lies untouched and still accessible
		opts.Transition.ExpireRestored = true
		_, err := objectAPI.DeleteObject(ctx, oi.Bucket, oi.Name, opts)
		if err == nil {

   * <p>Unlike {@link Double#parseDouble(String)}, this method returns {@code null} instead of
   * throwing an exception if parsing fails. Valid inputs are exactly those accepted by {@link
   * Double#valueOf(String)}, except that leading and trailing whitespace is not permitted.
   *
   * <p>This implementation is likely to be faster than {@code Double.parseDouble} if many failures
   * are expected.
   *

    testRotate(new float[] {1}, 0, 0, 1, new float[] {1});
    testRotate(new float[] {1}, 1, 0, 1, new float[] {1});
    testRotate(new float[] {1}, 1, 1, 1, new float[] {1});

    // Rotate the central 5 elements, leaving the ends as-is
    testRotate(new float[] {0, 1, 2, 3, 4, 5, 6}, -6, 1, 6, new float[] {0, 2, 3, 4, 5, 1, 6});
    testRotate(new float[] {0, 1, 2, 3, 4, 5, 6}, -1, 1, 6, new float[] {0, 2, 3, 4, 5, 1, 6});

    testRotate(new int[] {1}, 0, 0, 1, new int[] {1});
    testRotate(new int[] {1}, 1, 0, 1, new int[] {1});
    testRotate(new int[] {1}, 1, 1, 1, new int[] {1});

    // Rotate the central 5 elements, leaving the ends as-is
    testRotate(new int[] {0, 1, 2, 3, 4, 5, 6}, -6, 1, 6, new int[] {0, 2, 3, 4, 5, 1, 6});
    testRotate(new int[] {0, 1, 2, 3, 4, 5, 6}, -1, 1, 6, new int[] {0, 2, 3, 4, 5, 1, 6});

    testRotate(new float[] {1}, 0, 0, 1, new float[] {1});
    testRotate(new float[] {1}, 1, 0, 1, new float[] {1});
    testRotate(new float[] {1}, 1, 1, 1, new float[] {1});

    // Rotate the central 5 elements, leaving the ends as-is
    testRotate(new float[] {0, 1, 2, 3, 4, 5, 6}, -6, 1, 6, new float[] {0, 2, 3, 4, 5, 1, 6});
    testRotate(new float[] {0, 1, 2, 3, 4, 5, 6}, -1, 1, 6, new float[] {0, 2, 3, 4, 5, 1, 6});

 * possible over their JDK equivalents whenever you are expecting to handle only IP address string
 * literals -- there is no blocking DNS penalty for a malformed string.
 *
 * <p>When dealing with {@link Inet4Address} and {@link Inet6Address} objects as byte arrays (vis.
 * {@code InetAddress.getAddress()}) they are 4 and 16 bytes in length, respectively, and represent
 * the address in network byte order.
 *

Si tu aplicación es **simple**, probablemente esto **no será un problema**, y puede que no necesites especificar límites de memoria estrictos. Pero si estás **usando mucha memoria** (por ejemplo, con modelos de **Machine Learning**), deberías verificar cuánta memoria estás consumiendo y ajustar el **número de contenedores** que se ejecutan en **cada máquina** (y tal vez agregar más máquinas a tu cluster).