}

      else -> {
        toEvict = null
        toEvictIdleAtNs = -1L
      }
    }

    when {
      toEvict != null -> {
        // We've chosen a connection to evict. Confirm it's still okay to be evicted, then close it.
        toEvict.withLock {
          if (toEvict.calls.isNotEmpty()) return 0L // No longer idle.

The obvious solution to this is to put the list of selected workloads into the policy itself.
However, this means anytime a workload changes (often), we need to update the policy.

Instead, the opposite was chosen: each workload will list the policies that select it.
This works out to be more efficient in common cases where policies change much less often than workloads.

Refer to sizing guide for more understanding on default values chosen depending on your erasure stripe size [here](https://github.com/minio/minio/blob/master/docs/distributed/SIZING.md). Parity settings can be changed using [storage classes](https://github.com/minio/minio/tree/master/docs/erasure/storage-class).

// They can prevent flapping by specifying the stabilization window, so that the
// number of replicas is not set instantly, instead, the safest value from the stabilization
// window is chosen.
message HPAScalingRules {
  // stabilizationWindowSeconds is the number of seconds for which past recommendations should be
  // considered while scaling up or scaling down.

  // Required.
  optional PriorityLevelConfigurationReference priorityLevelConfiguration = 1;

  // `matchingPrecedence` is used to choose among the FlowSchemas that match a given request. The chosen
  // FlowSchema is among those with the numerically lowest (which we take to be logically highest)
  // MatchingPrecedence.  Each MatchingPrecedence value must be ranged in [1,10000].

  // Required.
  optional PriorityLevelConfigurationReference priorityLevelConfiguration = 1;

  // `matchingPrecedence` is used to choose among the FlowSchemas that match a given request. The chosen
  // FlowSchema is among those with the numerically lowest (which we take to be logically highest)
  // MatchingPrecedence.  Each MatchingPrecedence value must be ranged in [1,10000].

            Errors.DEFINITELY_NON_NULLABLE_AS_REIFIED,
            Errors.REIFIED_TYPE_FORBIDDEN_SUBSTITUTION,
            Errors.REIFIED_TYPE_UNSAFE_SUBSTITUTION,
            Errors.CANDIDATE_CHOSEN_USING_OVERLOAD_RESOLUTION_BY_LAMBDA_ANNOTATION,
            Errors.RESOLUTION_TO_CLASSIFIER,
            Errors.RESERVED_SYNTAX_IN_CALLABLE_REFERENCE_LHS,
            Errors.PARENTHESIZED_COMPANION_LHS_DEPRECATION,

   * <p>For the case of {@link RoundingMode#HALF_EVEN}, this implementation uses the IEEE 754
   * default rounding mode: if the two nearest representable values are equally near, the one with
   * the least significant bit zero is chosen. (In such cases, both of the nearest representable
   * values are even integers; this method returns the one that is a multiple of a greater power of
   * two.)
   *

// To signify no more results are on an input channel, close it.
// The output channel will be closed when all inputs are emptied.
// If file names are equal, compareMeta is called to select which one to choose.
// The entry not chosen will be discarded.
// If the context is canceled the function will return the error,
// otherwise the function will return nil.

   *       storedPermits, we serve from those first
   *   <li>When _unused_, we go right at a constant rate! The rate at which we move to the right is
   *       chosen as maxPermits / warmupPeriod. This ensures that the time it takes to go from 0 to
   *       maxPermits is equal to warmupPeriod.
   *   <li>When _used_, the time it takes, as explained in the introductory class note, is equal to