Titan index update takes too long

Question

Even on an empty database, creating an index in Titan 1.0 takes several minutes. The time seems exact, which suggests there is an unnecessary delay.

My question is this: How to I shorten or eliminate the amount of time Titan takes to reindex? Conceptually, since no work is being done the time should be minimal, certainly not four minutes.

(N.B. I have previously been pointed to a solution that simply makes Titan wait the full delay without timing out. This is the wrong solution - I want to eliminate the delay entirely.)

The code I'm using to setup the database from scratch is:

graph = ... a local cassandra instance ...
graph.tx().rollback()

// 1. Check if the index already exists
mgmt = graph.openManagement()
i = mgmt.getGraphIndex('byIdent')
if(! i) {
  // 1a. If the index does not exist, add it
  idKey = mgmt.getPropertyKey('ident')
  idKey = idKey ? idKey : mgmt.makePropertyKey('ident').dataType(String.class).make()
  mgmt.buildIndex('byIdent', Vertex.class).addKey(idKey).buildCompositeIndex()
  mgmt.commit()
  graph.tx().commit()

  mgmt  = graph.openManagement()
  idKey = mgmt.getPropertyKey('ident')
  idx   = mgmt.getGraphIndex('byIdent')
  // 1b. Wait for index availability
  if ( idx.getIndexStatus(idKey).equals(SchemaStatus.INSTALLED) ) {
    mgmt.awaitGraphIndexStatus(graph, 'byIdent').status(SchemaStatus.REGISTERED).call()
  }
  // 1c. Now reindex, even though the DB is usually empty.
  mgmt.updateIndex(mgmt.getGraphIndex('byIdent'), SchemaAction.REINDEX).get()
  mgmt.commit()
  mgmt.awaitGraphIndexStatus(graph, 'byIdent').status(SchemaStatus.ENABLED).call()
} else { mgmt.commit() }

It appears to be the updateIndex...REINDEX call that blocks till timeout. Is this a known problem or worksformewon'tfix? Am I doing something wrong?

EDIT: Disabling the REINDEX, as discussed in comments is actually not a fix because the index does not seem to become active. I now see:

WARN  com.thinkaurelius.titan.graphdb.transaction.StandardTitanTx  - Query requires iterating over all vertices [(myindexedkey = somevalue)]. For better performance, use indexes

Possible duplicate of [Index state never change to ENABLED on Titan with Amazon DynamoDB backend](http://stackoverflow.com/questions/35088574/index-state-never-change-to-enabled-on-titan-with-amazon-dynamodb-backend) — Mohamed Taher Alrefaie, Jun 03 '16 at 14:43
Eliminate the call to `REINDEX` if there is no existing data, like when you create the property key and index for the first time. — Jason Plurad, Jun 03 '16 at 17:24
@JasonPlurad That is a good strategy for most my uses. What if the database is tiny at the time of index creation? Say, what if I have very few but non-zero verticies? Must I reindex and incur this seemingly meaningless delay (till I submit a pull request, at least)? — Thomas M. DuBuisson, Jun 03 '16 at 17:33
Yeah, if you have data in there, you'll need to `REINDEX` in that case. Best practice is to define your schema and indexes up front and keep it locked down. — Jason Plurad, Jun 03 '16 at 17:37

Thomas M. DuBuisson · Accepted Answer · 2016-06-03T21:43:58.333

The time delay is/was entirely unnecessary and due to my misuse of Titan (though the pattern does appear in Titan 1.0.0 documentation chapter 28).

Do not block in a transaction!

Instead of:

  mgmt  = graph.openManagement()
  idKey = mgmt.getPropertyKey('ident')
  idx   = mgmt.getGraphIndex('byIdent')
  // 1b. Wait for index availability
  if ( idx.getIndexStatus(idKey).equals(SchemaStatus.INSTALLED) ) {
    mgmt.awaitGraphIndexStatus(graph, 'byIdent').status(SchemaStatus.REGISTERED).call()
  }

Consider:

  mgmt  = graph.openManagement()
  idKey = mgmt.getPropertyKey('ident')
  idx   = mgmt.getGraphIndex('byIdent')
  // Wait for index availability
  if ( idx.getIndexStatus(idKey).equals(SchemaStatus.INSTALLED) ) {
    mgmt.commit()
    mgmt.awaitGraphIndexStatus(graph, 'byIdent').status(SchemaStatus.REGISTERED).call()
  } else { mgmt.commit() }

Use ENABLE_INDEX

Not: mgmt.updateIndex(mgmt.getGraphIndex('byIdent'), SchemaAction.REINDEX).get()

Rather: mgmt.updateIndex(mgmt.getGraphIndex('byIdent'),SchemaAction.ENABLE_INDEX).get()

Titan index update takes too long

1 Answers1