Constructor Detail

• QuorumCheck.QuorumChecks

  public QuorumCheck.QuorumChecks()

Method Detail

• all

  public static QuorumCheck all()

  Checks that all members that should be up are up (i.e. ignores stopped nodes).

• allAndAtLeastOne

  public static QuorumCheck allAndAtLeastOne()

  Checks all members that should be up are up, and that there is at least one such member.

• atLeastOne

  public static QuorumCheck atLeastOne()

  Requires at least one member that should be up is up.

• atLeastOneUnlessEmpty

  public static QuorumCheck atLeastOneUnlessEmpty()

  Requires at least one member to be up if the total size is non-zero. i.e. okay if empty, or if non-empty and something is healthy, but not okay if not-empty and nothing is healthy. "Empty" means that no members are supposed to be up (e.g. there may be stopped members).

• alwaysTrue

  public static QuorumCheck alwaysTrue()

  Always "healthy"

• newInstance

  public static QuorumCheck newInstance(int minRequiredSize,
                        double minRequiredRatio,
                        boolean allowEmpty)

• newLinearRange

  public static QuorumCheck newLinearRange(java.lang.String range)

  See newLinearRange(String,String)

• newLinearRange

  public static QuorumCheck newLinearRange(java.lang.String range,
                           java.lang.String name)

  Given a JSON representation of a list of points (where a point is a list of 2 numbers), with the points in increasing x-coordinate value, this constructs a quorum check which does linear interpolation on those coordinates, with extensions to either side. The x-coordinate is taken as the total size, and the y-coordinate as the minimum required size.

  It sounds complicated but it gives a very easy and powerful way to define quorum checks. For instance:

  [[0,0],[1,1]] says that if 0 items are expected, at least 0 is required; if 1 is expected, 1 is required; and by extension if 10 are expected, 10 are required. In other words, this is the same as all().

  [[0,1],[1,1],[2,2]] is the same as the previous for x (number expected) greater-than or equal to 1; but if 0 is expected, 1 is required, and so it fails when 0 are present. In other words, allAndAtLeastOne().

  [[5,5],[10,10],[100,70],[200,140]] has all() behavior up to 10 expected (line extended to the left, for less than 5); but then gently tapers off to requiring only 70% at 100 (with 30 of 40 = 75% required at that intermediate point along the line [[10,10],[100,70]]); and then from 100 onwards it is a straight 70%.

  The type of linear regression described in the last example is quite useful in practise, to be stricter for smaller clusters (or possibly more lax for small values in some cases, such as when tolerating dangling references during rebind).

• of

  public static QuorumCheck of(java.lang.String nameOrRange)