USER MANUALS

Execution Trace of a Statement

After executing a statement, you can request the trace of its execution. It is also possible to obtain the execution plan before executing the query. In both cases, the information displayed is the same, except for certain parameters that are only known at run time (e.g. execution time).

The execution plan is displayed in a tree diagram, where each node represents one of these elements:

  • The tree root, which is called the “Execution node”. It contains information about the query.

  • An intermediate view involved in the execution of the statement

  • Or, a retrieval of data from a source.

The main attributes of these nodes are the following:

  • Type. If the node is a view, it indicates the type of the view (base view, union, join, selection, projection, etc.). If it is an access to a data source (wrapper), this indicates the type of source (JDBC, ODBC, Web Service, etc.).

    The main node of the trace is “Execution” (the node at the top of the trace), whose attributes represent the state of the entire query.

  • Execution time. Time (milliseconds) spent executing this node and its subnodes. In the “Execution” node, this is execution time of the query.

  • Start time. Instant at which the Server began executing this node. In the “Execution” node, is the instant at which the Server received the query.

  • End time. Instant at which the Server finished executing this node and its subnodes. In the “Execution” node, is the instant at which the Server finished the execution of the query.

  • Response time. Instant at which the node received the first row to process. In the “Execution” node, is the instant at which the Server obtained the first row of the query’s result.

    If the node did not process any row, the value of this attribute is unknown.

  • Effective time. Time spent to execute this node. It does not include the time spent waiting for other nodes of the execution tree to complete their work.

  • Data serialization time (only displayed in the “Execution Plan” node). Time used by the server to transform the result of the query to a byte stream to be sent to the client application.

  • Number of rows. Number of rows returned by the node. In the “Execution” node, it is the number of rows returned by the query.

  • State. It indicates whether the node was correctly executed or whether an error occurred during its execution of the execution of one of its subnodes.

    The table Values of the “State” attribute of the nodes of a query’s execution trace of the appendix Execution Trace Information lists the possible values for this attribute.

  • Completed. If yes, the node finished correctly. If no, there was an error during its execution or the execution of one of its subnodes.

  • Waiting time for execution (only in the “Execution” node). Time (milliseconds) the query was waiting in the queue of queries before the Server began executing it.

    A query is hold in the queue of queries when the limit of concurrent requests is reached. This limit is configured on the tab “Concurrent requests” of the menu “Administration > Server configuration”. See more about this limit in the section Limiting the Number of Concurrent Requests.

    The Execution time includes the Waiting time for execution.

  • Static optimization (only in the “Execution” node). Parameters regarding the Automatic simplifications described in the section Automatic Simplification of Queries.

    • Optimization time. Time spent by the Execution Engine to analyze the query plan.

    • Static optimized is yes if the query plan was simplified after being analyzed.

  • Cost optimization (only in the “Execution” node). Parameters regarding the usage of the Cost-based optimizations described in the section Cost-Based Optimization. If the execution engine uses the cost optimizer to select an execution plan, the trace also contains some information about its decision:

    • Estimated number of rows, row size in bytes and data volume: displayed when the execution plan was obtained without executing the query. It shows how many rows the optimizer expected this node to return. For join nodes, the right branch always assumes the strategy hash or merge.

    • Indexes: indexes that the optimizer expects to be used by the underlying data source for retrieving data.

    • Total, Left Branch and Right Branch cost (displayed only in join nodes): total cost of the join operation and the costs for its branches.

  • Optimizations applied (only in the “Execution” node). List of strategies use by the engine to improve the performance of the query.

  • Accelerated with summaries (only in the “Execution” node). List of summaries used when the “Summary acceleration” optimization has been applied.

  • Advanced parameters. These provide further details on each node type. For example:

    • In the case of “route” nodes, the queries executed on the data source and the connection URI to the source.

    • Swapping and number of swapped rows: for each node of the query, whether the data has been swapped to disk and the number of rows swapped to disk during the execution of the node. The top execution node also displays the maximum amount of disk used by the query for swapping and the maximum number of swap files that the query has opened at the same time.

    • “No delegation cause” explains why a query was not delegated to the source database.

      When a query uses two or more views that rely on the same JDBC or ODBC data source, Virtual DataPort will try to delegate the entire query to the source database. This way, it can obtain the whole result executing a single query. When Virtual DataPort cannot delegate the entire process of the query to a source database, it will show the icon image0 beside the view that could not be delegated and the “No delegation cause” parameter will show the reason why the delegation was not possible. For example, the query may use a function that is not supported by the source database, which will force Virtual DataPort to post-process the results obtained.

      The “No delegation cause” message will help us to rewrite the view so it can be delegated.

      The reasons for which a query involving views of a single JDBC data source might not be delegated are:

      • The query uses a function or operator not supported by the database.

      • The query uses a WHERE subquery that returns multiple values (for example, using an IN condition with several fields) and the database does not support this kind of subqueries. VDP delegates this kind of subqueries for some databases like Oracle, DB2, PostgreSQL, MySQL and redshift, but it can exist some database for which, even supporting this subqueries, the delegation is not enabled by default.

      • The query contains a subquery with ORDER BY fields.

      • You have modified certain settings of the “Source configuration” of the data source. Some settings affect the delegation of queries. For example, if you set “Delegate Joins” to “No”, the data source will not delegate any JOIN to this database. We advice against modifying these advanced settings unless advised by Denodo Support.

The most relevant parameters are displayed for each node on the query execution tree. You can copy the value of each parameter by right-clicking it and then, clicking Copy or copy the entire trace by clicking Copy trace to clipboard (image1)

To save the execution trace and the diagram, click Save (image2).


The figure below displays the execution trace of this query:

SELECT *
FROM internet_inc
WHERE summary REGEXP_LIKE '.*(Incident)|(Error).*'

If the database does not support the operator REGEXP_LIKE, the query cannot be delegated to the database and the selection node is marked with this icon image0, meaning that the selection condition will be executed by Virtual DataPort.

The main usage of the trace is debugging. When there is an error during the execution of a query, the nodes that caused the error are marked in red. By clicking on them, you can see more information about the error. The appendix Execution Trace Information explains the meaning of the icons and colors of the execution trace of a query.

If the Automatic simplification of queries (see section Automatic Simplification of Queries) is enabled, the Execution trace may be very different to the hierarchy defined by the queried view.

internet_inc view execution trace

internet_inc view execution trace

The figure below shows the execution trace of the incidents_sales join view in the case where the Web Service data source is not accessible.

Join view execution trace with errors

Join view execution trace with errors

Besides displaying the execution trace of the query, you can see the “query plan” of a query before executing it, to know how the Server will execute the query, without executing it. This is very useful when dealing with a complex hierarchy of views because you can see the join strategy that the Server will use in each join operation (merge, hash, nested, etc.), how a query will be delegated to a database, etc.

To display this, click Query Plan in the “Execute view” dialog. The process of calculating a query plan is very similar to the actual execution of the query but without actually “hitting” the actual data sources. For example, to simulate the execution of the right branch of a nested join, the Server has to generate a value for the attributes of the right side of the join condition. That is because these values are mandatory and at runtime, the Server uses the values obtained from the left side of the join. However, when obtaining the query plan, it does not have the actual values so it uses the values listed in this table:

Values used to generate query plans

Data type

Values used to generate the query plan

boolean

false

date

The current date

decimal

0.0

double

0.0

float

0.0

int

0

long

0

text

'' (empty string)

xml

'' (empty string)

array

{ } An empty array

register

A register with all its elements set to NULL

Storing the Execution Trace of Queries

You can configure the logging engine of Virtual DataPort to store the execution trace of the queries. This is useful to troubleshoot queries that are executed from an external application, not the administration tool. You can choose to store the execution traces in files that can be opened from the Trace Viewer or in a plain text file.

To enable this feature, execute the following command:

SET 'com.denodo.vdb.interpreter.execution.saveTrace' = '<save trace configuration>';
Possible values of the parameter “<save trace configuration>”

Value

Meaning of the Parameter

ALL_LOG

Store the trace of all the queries in plain text.

ALL_FILE

Store the trace of all the queries on a file for the Trace Viewer.

ALL_FULL

Store the trace of all the queries in plain text and on a file for the Trace Viewer.

ERROR_LOG

Only store the trace of the queries that fail. Store the trace in plain text.

ERROR_FILE

Only store the trace of the queries that fail. Store the trace on a file for the Trace Viewer.

ERROR_FULL

Only store the trace of the queries that fail. Store the trace in plain text and on a file for the Trace Viewer.

DISABLED

Disable this feature (default value).

Note

If you store the execution trace of all the queries, not just the ones that fail, consider the amount of disk space required. The execution trace of one simple query in plain text uses about 4 kilobytes of disk space.

Storing the execution traces in plain text

When the execution traces are stored in plain text (this is with the options ALL_LOG, ALL_FULL, ERROR_LOG or ERROR_FULL), all the traces are stored on the same file. The file where the traces are stored can be:

  1. The vdp.log file (<DENODO_HOME>/logs/vdp/). This option is appropriate for a quick test because it is easy to set up and you do not have to restart Virtual DataPort to enable it. However, if you plan on storing the execution traces all the time, we recommend option b). That is because with option a), the execution traces will be mixed with the error messages of Virtual DataPort, which will make the detection of future issues more difficult.

    To enable this option, also execute the following command (in addition to the command SET 'com.denodo.vdb.interpreter.execution.saveTrace'...)

    CALL LOGCONTROLLER('com.denodo.vdp.traces', 'DEBUG');
    

    If you restart Virtual DataPort, you have to execute this command again. If not, the execution traces will not be stored.

  2. Store the execution traces on another file, not in vdp.log (recommended option). To do this, modify <DENODO_HOME>/conf/vdp/log4j2.xml with the configuration below.

    You have to add the following fragments (RollingFile and Logger) inside the sections Appenders and Loggers respectively so the file ends up looking like this:

    <Appenders>
    
    [...]
    
        <RollingFile name="TRACESOUT" fileName="../logs/vdp/vdp-traces${env:vdp.instance.log}.log" filePattern="../logs/vdp/trace/vdp-traces${env:vdp.instance.log}.log.%i">
            <Policies>
                <SizeBasedTriggeringPolicy size="10 MB" />
            </Policies>
            <DefaultRolloverStrategy max="7" />
            <PatternLayout pattern="%-4r [%t] %d{yyyy-MM-dd'T'HH:mm:ss.SSS} %x -\t%m  %n" />
        </RollingFile>
    
    [...]
    
    </Appenders>
    
    <Loggers>
    
    [...]
    
        <Logger name="com.denodo.vdp.traces" level="debug" additivity="false">
            <AppenderRef ref="TRACESOUT" />
        </Logger>
    
    [...]
    
    </Loggers>
    

    After saving this file, restart Virtual DataPort to apply the changes.

    Explanation of this configuration:

    • The execution traces are stored on the file <DENODO_HOME>/logs/vdp/vdp-traces.log. If needed, you can change this path.

    • When the size of the file reaches 10 megabytes, the file is renamed to “vdp-traces.log.1” and a new file “vdp-traces.log” is created. This process is repeated until there are seven files. After that, the oldest file is deleted. The goal of this setting is to avoid filling the disk with execution traces.

Storing the execution traces on a zip file

Instead of storing the execution traces in plain text, you can store them on files that can be opened with the Trace Viewer of the administration tool (menu Tools). This is with the options ALL_FILE, ALL_FULL, ERROR_FILE or ERROR_FULL. From the Trace Viewer, you can graphically see the trace, click each node to see its execution details, etc. The zip file also contains the execution trace in plain text.

In this mode, the trace of each query is stored in a different zip file.

If you choose this option, you do not need to restart Virtual DataPort; just execute the command SET 'com.denodo.vdb.interpreter.execution.saveTrace'....

By default, these files are stored in the directory <DENODO_HOME>/logs/vdp/trace. To change this, execute the following command:

SET 'com.denodo.vdb.interpreter.execution.saveTraceDir' = '<directory to store the traces>';

If the directory does not exist, Denodo will create it. You do not need to restart to apply the change.

Add feedback