Routable query performance and troubleshooting

This topic details performance considerations and common issues you might encounter when using the routable query API.

This topic details performance considerations and common issues you might encounter when using the routable query API.

Using resource pools with routable queries

Individual routable queries are serviced quickly since they directly access a single node and return only one or a few rows of data. However, by default, Vertica resource pools use an AUTO setting for the execution parallelism parameter. When set to AUTO, the setting is determined by the number of CPU cores available and generally results in multi-threaded execution of queries in the resource pool. It is not efficient to create parallel threads on the server because routable query operations return data so quickly and routable query operations only use a single thread to find a row. To prevent the server from opening unneeded processing threads, you should create a specific resource pool for routable query clients. Consider the following settings for the resource pool you use for routable queries:

  • Set execution parallelism to 1 to force single-threaded queries. This setting improves routable query performance.

  • Use CPU affinity to limit the resource pool to a specific CPU or CPU set. The setting ensures that the routable queries have resources available to them, but it also prevents routable queries from significantly impacting performance on the system for other general queries.

  • If you do not set a CPU affinity for the resource pool, consider setting the maximum concurrency value of the resource pool to a setting that ensures good performance for routable queries, but does not negatively impact the performance of general queries.

Performance considerations for routable query connections

Because a VerticaRoutableConnection opens an internal pool of connections, it is important to configure MaxPooledConnections and MaxPooledConnectionsPerNode appropriately for your cluster size and the amount of simultaneous client connections. It is possible to impact normal database connections if you are overloading the cluster with VerticaRoutableConnections.

The initial connection to the initiator node discovers all other nodes in the cluster. The internal-pool connections are not opened until a VerticaRoutableExecutor or VGet query is sent. All VerticaRoutableExecutors/VGets in a connection object use connections from the internal pool and are limited by the MaxPooledConnections settings. Connections remain open until they are closed so a new connection can be opened elsewhere if the connection limit has been reached.

Troubleshooting routable queries

Routable query issues generally fall into two categories:

  • Not providing enough predicates.

  • Queries having to span multiple nodes.

Predicate Requirements

You must provide the same number of predicates that correspond to the columns of the table segmented by hash. To determine the segmented columns, call the Vertica function GET_PROJECTIONS. You must provide a predicate for each column displayed in the Seg Cols field.

For VGet, this means you must use addPredicate() to add each of the columns. For VerticaRoutableExecutor, this means you must provide all of the predicates and values in the map sent to execute().

Multi-node Failures

It is possible to define the correct number of predicates, but still have a failure because multiple nodes contain the data. This failure occurs because the projection's data is not segmented in such a way that the data being queried is contained on a single node. Enable logging for the connection and view the logs to verify the projection being used. If the client is not picking the correct projection, then try to query the projection directly by specifying the projection instead of the table in the create/prepare statement, for example:

  • Using VerticaRoutableExecutor:

    conn.createRoutableExecutor(schema, table/projection);
    
  • Using VGet:

    conn.prepareGet('schema','table/projection')
    

Additionally, you can use the EXPLAIN command in vsql to help determine if your query can run in single node. EXPLAIN can help you understand why the query is being run as single or multi-node.