When replicas specified in ml-config, fails wipe if number of hosts is one; allows replication with number of hosts less than 3.

[jamsilvia]

Two parts to this problem:

1. If the ml-config.xml specifies a replica, but there is only one host, bootstrap will succeed (with no warnings, and no replicas created). However, trying to wipe will fail with "ERROR: XDMP-DIVBYZERO". The problem being that the checks for the number of hosts to bypass "reassign-replicas" is only done on the create step, but not on the wipe step.
2. Replication is not going to be usable for failover unless the number of hosts > 3 (to ensure a quorum). In general, the host count should be "2n+1", where n is the number of replicas. A warning should be generated if requesting replication for < 3 hosts. A warning COULD be generated when the "2n+1" rule is not followed.

Reproduce:

1. Problem 1 is reproducible by:

• Edit a default ml-config.xml to setup replication for a content forest:

    <assignment>
      <forest-name>${content-db}</forest-name>
      <replica-names>
        <replica-name>${content-db}-rep1</replica-name>
      </replica-names>
      @ml.forest-data-dir-xml
    </assignment>
    <assignment>
      <forest-name nr-replicas="1">${content-db}-rep1</forest-name>
      @ml.forest-data-dir-xml
    </assignment>

• ./ml dev bootstrap
• ./ml dev wipe
• Should report the division by zero error.

2. Problem 2 is reproducible by:

• Edit a default ml-config.xml to setup replication for a content forest:

    <assignment>
      <forest-name>${content-db}</forest-name>
      <replica-names>
        <replica-name>${content-db}-rep1</replica-name>
      </replica-names>
      @ml.forest-data-dir-xml
    </assignment>
    <assignment>
      <forest-name nr-replicas="1">${content-db}-rep1</forest-name>
      @ml.forest-data-dir-xml
    </assignment>

• ./ml dev bootstrap
• Silently ignores replication.
• Create a 2-node cluster and do the same steps as above.
• Creates a replica, but failover would be unusable.

Which Operating System are you using? Linux and Mac OS-X

Which version of MarkLogic are you using? 9.0-2

Which version of Roxy are you using (see version.txt)? 1.7.6 and 1.7.7

I have a fix for this coded already, and we are currently testing it on our project.

[tdiepenbrock]

We have a fix already for this, Joe McGroarty has implemented it.

[grtjn]

@tdiepenbrock If you could share the changes applied, that would be grand. Offline, PR, or in a comment..

[jamsilvia]

Oh! I'll send that tonight....

[jamsilvia]

I submitted the pull request with the changes. Let me know if you need anything else.