EXPLAIN ANALYZE

On this page Carat arrow pointing down

The EXPLAIN ANALYZE statement executes a SQL query and generates a statement plan with execution statistics. Statement plans provide information around SQL execution, which can be used to troubleshoot slow queries by figuring out where time is being spent, how long a processor (i.e., a component that takes streams of input rows and processes them according to a specification) is not doing work, etc. The (DISTSQL) option returns the statement plan and performance statistics as well as a generated link to a graphical distributed SQL physical statement plan tree. For more information about distributed SQL queries, see the DistSQL section of our SQL layer architecture docs. The (DEBUG) option generates a URL to download a bundle with more details about the statement plan for advanced debugging.

Note:

The generated physical statement plan is encoded into a byte string after the fragment identifier (#) in the generated URL. The fragment is not sent to the web server; instead, the browser waits for the web server to return a decode.html resource, and then JavaScript on the web page decodes the fragment into a physical statement plan diagram. The statement plan is, therefore, not logged by a server external to the CockroachDB cluster and not exposed to the public internet.

Aliases

EXPLAIN ANALYSE is an alias for EXPLAIN ANALYZE.

Synopsis

EXPLAIN ANALYZE ANALYSE ( PLAN DISTSQL DEBUG , ) explainable_stmt

Parameters

Parameter Description
PLAN (Default) Execute the statement and return a statement plan with planning and execution time for an explainable statement. See PLAN option.
DISTSQL Execute the statement and return a statement plan and performance statistics as well as a generated link to a graphical distributed SQL physical statement plan tree. See DISTSQL option.
DEBUG Execute the statement and generate a ZIP file containing files with detailed information about the query and the database objects referenced in the query. See DEBUG option.
preparable_stmt The statement you want to execute and analyze. All preparable statements are explainable.

Required privileges

The user requires the appropriate privileges for the statement being explained.

Success responses

A successful EXPLAIN ANALYZE statement returns a table with the following details in the info column:

Detail Description
Global properties The properties and statistics that apply to the entire statement plan.
Statement plan tree properties A tree representation of the hierarchy of the statement plan.
Node details The properties, columns, and ordering details for the current statement plan node in the tree.
Time The time details for the statement. The total time is the planning and execution time of the statement. The execution time is the time it took for the final statement plan to complete. The network time is the amount of time it took to distribute the statement across the relevant nodes in the cluster. Some statements do not need to be distributed, so the network time is 0ms.

If you use the DISTSQL option, the statement will also return a URL generated for a physical statement plan that provides high level information about how a statement will be executed. The generated physical statement plan is encoded into a byte string after the fragment identifier (#) in the generated URL. The fragment is not sent to the web server; instead, the browser waits for the web server to return a decode.html resource, and then JavaScript on the web page decodes the fragment into a physical statement plan diagram. The statement plan is, therefore, not logged by a server external to the CockroachDB cluster and not exposed to the public internet. For details about reading the physical statement plan, see DistSQL plan diagram.

If you use the DEBUG option, the statement will return only a URL and instructions to download the DEBUG bundle, which includes the physical statement plan.

Global properties

Property Description
planning time The total time the planner took to create a statement plan.
execution time The time it took for the final statement plan to complete.
distribution Whether the statement was distributed or local. If distribution is full, execution of the statement is performed by multiple nodes in parallel, then the results are returned by the gateway node. If local, the execution plan is performed only on the gateway node. Even if the execution plan is local, row data may be fetched from remote nodes, but the processing of the data is performed by the local node.
vectorized Indicates whether the vectorized execution engine was used in this statement.
rows read from KV The number of rows read from the storage layer.
cumulative time spent in KV The total amount of time spent in the storage layer.
maximum memory usage The maximum amount of memory used by this statement anytime during its execution.
network usage The amount of data transferred over the network while the statement was executed. If the value is 0 B, the statement was executed on a single node and didn't use the network.
regions The regions where the affected nodes were located.
max sql temp disk usage (DISTSQL option only) How much disk spilling occurs when executing a query. This property is displayed only when the disk usage is greater than zero.

Statement plan tree properties

Statement plan tree properties Description
processor Each processor in the statement plan hierarchy has a node with details about that phase of the statement. For example, a statement with a GROUP BY clause has a group processor with details about the cluster nodes, rows, and operations related to the GROUP BY operation.
nodes The names of the CockroachDB cluster nodes affected by this phase of the statement.
regions The regions where the affected nodes were located.
actual row count The actual number of rows affected by this processor during execution.
KV time The total time this phase of the statement was in the storage layer.
KV contention time The time the storage layer was in contention during this phase of the statement.
KV rows read During scans, the number of rows in the storage layer read by this phase of the statement.
KV bytes read During scans, the amount of data read from the storage layer during this phase of the statement.
estimated max memory allocated The estimated maximum allocated memory for a statement.
estimated max sql temp disk usage The estimated maximum temporary disk usage for a statement.
MVCC step count (ext/int) The number of times that the underlying storage iterator stepped forward during the work to serve the operator's reads, including stepping over MVCC keys that could not be used in the scan.
MVCC seek count (ext/int) The number of times that the underlying storage iterator jumped (seeked) to a different data location.
estimated row count The estimated number of rows affected by this processor according to the statement planner, the percentage of the table the query spans, and when the statistics for the table were last collected.
table The table and index used in a scan operation in a statement, in the form {table name}@{index name}.
spans The interval of the key space read by the processor. If spans is FULL SCAN, the table is scanned on all key ranges of the index. If spans is [/1 - /1], only the key with value 1 is read by the processor.

PLAN option

By default, EXPLAIN ANALYZE uses the PLAN option. EXPLAIN ANALYZE and EXPLAIN ANALYZE (PLAN) produce the same output.

PLAN options

The PLAN options VERBOSE and TYPES described in EXPLAIN options are also supported. For an example, see EXPLAIN ANALYZE (VERBOSE).

DISTSQL option

EXPLAIN ANALYZE (DISTSQL) generates a physical statement in the plan diagram. The DistSQL plan diagram displays the physical statement plan, as well as execution statistics. The statistics listed depend on the query type and the execution engine used. If the query contains subqueries or post-queries there will be multiple diagrams.

Note:

You can use EXPLAIN ANALYZE (DISTSQL) only as the top-level statement in a query.

DistSQL plan diagram

The graphical plan diagram displays the processors and operations that make up the statement plan. While the text output from the PLAN option shows the statement plan across the cluster, the DISTSQL option shows details on each node involved in the query.

Field Description Execution engine
<Processor>/<id> The processor and processor ID used to read data into the SQL execution engine.

A processor is a component that takes streams of input rows, processes them according to a specification, and outputs one stream of rows. For example, a TableReaderprocessor reads in data, and an Aggregator aggregates input rows.
Both
<table>@<index> The index used by the processor. Both
Spans The interval of the key space read by the processor. For example, [/1 - /1] indicates that only the key with value 1 is read by the processor. Both
Out The output columns. Both
KV time The total time this phase of the query was in the storage layer. Both
KV contention time The time the storage layer was in contention during this phase of the query. Both
KV rows read During scans, the number of rows in the storage layer read by this phase of the query. Both
KV bytes read During scans, the amount of data read from the storage layer during this phase of the query. Both
cluster nodes The names of the CockroachDB cluster nodes involved in the execution of this processor. Both
batches output The number of batches of columnar data output. Vectorized engine only
rows output The number of rows output. Vectorized engine only
IO time How long the TableReader processor spent reading data from disk. Vectorized engine only
stall time How long the processor spent not doing work. This is aggregated into the stall time numbers as the query progresses down the tree (i.e., stall time is added up and overlaps with previous time). Row-oriented engine only
bytes read The size of the data read by the processor. Both
rows read The number of rows read by the processor. Both
@<n> The index of the column relative to the input. Both
max memory used How much memory (if any) is used to buffer rows. Row-oriented engine only
max disk used How much disk (if any) is used to buffer data. Routers and processors will spill to disk buffering if there is not enough memory to buffer the data. Row-oriented engine only
execution time How long the engine spent executing the processor. Vectorized engine only
max vectorized memory allocated How much memory is allocated to the processor to buffer batches of columnar data. Vectorized engine only
max vectorized disk used How much disk (if any) is used to buffer columnar data. Processors will spill to disk buffering if there is not enough memory to buffer the data. Vectorized engine only
left(@<n>)=right(@<n>) The equality columns used in the join. Both
stored side The smaller table that was stored as an in-memory hash table. Both
rows routed How many rows were sent by routers, which can be used to understand network usage. Row-oriented engine only
network latency The latency time in nanoseconds between nodes in a stream. Vectorized engine only
bytes sent The number of actual bytes sent (i.e., encoding of the rows). This is only relevant when doing network communication. Both
Render The stage that renders the output. Both
by hash (Orange box) The router, which is a component that takes one stream of input rows and sends them to a node according to a routing algorithm.

For example, a hash router hashes columns of a row and sends the results to the node that is aggregating the result rows.
Both
unordered / ordered (Blue box) A synchronizer that takes one or more output streams and merges them to be consumable by a processor. An ordered synchronizer is used to merge ordered streams and keeps the rows in sorted order. Both
<data type> If you specify EXPLAIN (DISTSQL, TYPES), lists the data types of the input columns. Both
Response The response back to the client. Both

DEBUG option

EXPLAIN ANALYZE (DEBUG) executes a query and generates a link to a ZIP file that contains the physical statement plan, execution statistics, statement tracing, and other information about the query.

File Description
stats-{table}.sql Contains statistics for a table in the query.
schema.sql Contains CREATE statements for objects in the query.
env.sql Contains information about the CockroachDB environment.
trace.txt Contains statement traces in plaintext format.
trace.json Contains statement traces in JSON format.
trace-jaeger.json Contains statement traces in JSON format that can be imported to Jaeger.
distsql.html The query's physical statement plan. This diagram is identical to the one generated by EXPLAIN (DISTSQL).
plan.txt The query execution plan. This is identical to the output of EXPLAIN (VERBOSE).
opt.txt The statement plan tree generated by the cost-based optimizer. This is identical to the output of EXPLAIN (OPT).
opt-v.txt The statement plan tree generated by the cost-based optimizer, with cost details. This is identical to the output of EXPLAIN (OPT, VERBOSE).
opt-vv.txt The statement plan tree generated by the cost-based optimizer, with cost details and input column data types. This is identical to the output of EXPLAIN (OPT, TYPES).
vec.txt The statement plan tree generated by the vectorized execution engine. This is identical to the output of EXPLAIN (VEC).
vec-v.txt The statement plan tree generated by the vectorized execution engine. This is identical to the output of EXPLAIN (VEC, VERBOSE).
statement.txt The SQL statement for the query.

You can obtain this ZIP file by following the link provided in the EXPLAIN ANALYZE (DEBUG) output, or by activating statement diagnostics in the DB Console.

Warning:
Statement bundles can contain unredacted user data including histograms and placeholders that contain real data samples, statements that have real data which can contain PII information, and database schema that could be sensitive. Be careful when generating and sharing statement bundles.

To allow or disallow a role from seeing statements diagnostics bundles, set the VIEWACTIVITYREDACTED role option.

Examples

The following examples use the movr example dataset.

Start the MovR database on a 3-node CockroachDB demo cluster with a larger data set.

icon/buttons/copy
cockroach demo movr --num-histories 250000 --num-promo-codes 250000 --num-rides 125000 --num-users 12500 --num-vehicles 3750 --nodes 3

EXPLAIN ANALYZE

Use EXPLAIN ANALYZE without an option, or equivalently with the PLAN option, to execute a query and display the physical statement plan with execution statistics.

For example, the following EXPLAIN ANALYZE statement executes a simple query against the MovR database and then displays the physical statement plan with execution statistics:

icon/buttons/copy
> EXPLAIN ANALYZE SELECT city, AVG(revenue) FROM rides GROUP BY city;
  planning time: 604µs
  execution time: 51ms
  distribution: full
  vectorized: true
  rows read from KV: 125,000 (21 MiB)
  cumulative time spent in KV: 106ms
  maximum memory usage: 5.0 MiB
  network usage: 2.6 KiB (24 messages)
  regions: us-east1

  • group (streaming)
  │ nodes: n1, n2, n3
  │ regions: us-east1
  │ actual row count: 9
  │ estimated row count: 9
  │ group by: city
  │ ordered: +city
  │
  └── • scan
        nodes: n1, n2, n3
        regions: us-east1
        actual row count: 125,000
        KV time: 106ms
        KV contention time: 0µs
        KV rows read: 125,000
        KV bytes read: 21 MiB
        estimated max memory allocated: 21 MiB
        estimated row count: 125,000 (100% of the table; stats collected 1 hour ago)
        table: rides@rides_pkey
        spans: FULL SCAN
(30 rows)

If you perform a join, the estimated max memory allocation is also reported for the join. For example:

icon/buttons/copy
EXPLAIN ANALYZE SELECT * FROM vehicles JOIN rides ON rides.vehicle_id = vehicles.id and rides.city = vehicles.city limit 100;
                        info
-----------------------------------------------------
  planning time: 1ms
  execution time: 18ms
  distribution: full
  vectorized: true
  rows read from KV: 3,173 (543 KiB)
  cumulative time spent in KV: 37ms
  maximum memory usage: 820 KiB
  network usage: 3.3 KiB (2 messages)
  regions: us-east1

  • limit
  │ nodes: n1
  │ regions: us-east1
  │ actual row count: 100
  │ estimated row count: 100
  │ count: 100
  │
  └── • lookup join
      │ nodes: n1, n2, n3
      │ regions: us-east1
      │ actual row count: 194
      │ KV time: 31ms
      │ KV contention time: 0µs
      │ KV rows read: 173
      │ KV bytes read: 25 KiB
      │ estimated max memory allocated: 300 KiB
      │ estimated row count: 13,837
      │ table: vehicles@vehicles_pkey
      │ equality: (city, vehicle_id) = (city,id)
      │ equality cols are key
      │
      └── • scan
  ...
(41 rows)

EXPLAIN ANALYZE (VERBOSE)

The VERBOSE option displays the physical statement plan with additional execution statistics.

icon/buttons/copy
EXPLAIN ANALYZE (VERBOSE) SELECT city, AVG(revenue) FROM rides GROUP BY city;
                                         info
--------------------------------------------------------------------------------------
  planning time: 5ms
  execution time: 65ms
  distribution: full
  vectorized: true
  rows read from KV: 125,000 (21 MiB)
  cumulative time spent in KV: 114ms
  maximum memory usage: 5.0 MiB
  network usage: 2.6 KiB (24 messages)
  regions: us-east1

  • group (streaming)
  │ columns: (city, avg)
  │ nodes: n1, n2, n3
  │ regions: us-east1
  │ actual row count: 9
  │ vectorized batch count: 4
  │ estimated row count: 9
  │ aggregate 0: avg(revenue)
  │ group by: city
  │ ordered: +city
  │
  └── • scan
        columns: (city, revenue)
        ordering: +city
        nodes: n1, n2, n3
        regions: us-east1
        actual row count: 125,000
        vectorized batch count: 124
        KV time: 114ms
        KV contention time: 0µs
        KV rows read: 125,000
        KV bytes read: 21 MiB
        estimated max memory allocated: 21 MiB
        MVCC step count (ext/int): 125,000/125,000
        MVCC seek count (ext/int): 18/18
        estimated row count: 125,000 (100% of the table; stats collected 1 hour ago)
        table: rides@rides_pkey
        spans: FULL SCAN
(38 rows)

EXPLAIN ANALYZE (DISTSQL)

Use EXPLAIN ANALYZE (DISTSQL) to execute a query, display the physical statement plan with execution statistics, and generate a link to a graphical DistSQL statement plan.

icon/buttons/copy
EXPLAIN ANALYZE (DISTSQL) SELECT city, AVG(revenue) FROM rides GROUP BY city;
                                          info
----------------------------------------------------------------------------------------------------
  planning time: 638µs
  execution time: 132ms
  distribution: full
  vectorized: true
  rows read from KV: 125,000 (21 MiB)
  cumulative time spent in KV: 228ms
  maximum memory usage: 7.5 MiB
  network usage: 2.5 KiB (24 messages)
  regions: us-east1

  • group (streaming)
  │ nodes: n1, n2, n3
  │ regions: us-east1
  │ actual row count: 9
  │ estimated row count: 9
  │ group by: city
  │ ordered: +city
  │
  └── • scan
        nodes: n1, n2, n3
        regions: us-east1
        actual row count: 125,000
        KV time: 228ms
        KV contention time: 0µs
        KV rows read: 125,000
        KV bytes read: 21 MiB
        estimated max memory allocated: 20 MiB
        estimated row count: 125,000 (100% of the table; stats collected 1 second ago)
        table: rides@rides_pkey
        spans: FULL SCAN

  Diagram: https://cockroachdb.github.io/distsqlplan/decode.html#eJzUmF9u47YTx99_pyD4lMVPuxIpWZb8...
(32 rows)

To view the DistSQL plan diagram, open the URL following Diagram. For an example, see DISTSQL option.

EXPLAIN ANALYZE (DEBUG)

Use the DEBUG option to generate a ZIP file containing files with information about the query and the database objects referenced in the query. For example:

icon/buttons/copy
> EXPLAIN ANALYZE (DEBUG) SELECT city, AVG(revenue) FROM rides GROUP BY city;
                                      info
--------------------------------------------------------------------------------
  Statement diagnostics bundle generated. Download from the DB Console (Advanced
  Debug -> Statement Diagnostics History), via the direct link below, or using
  the SQL shell or command line.
  Admin UI: http://127.0.0.1:8080
  Direct link: http://127.0.0.1:8080/_admin/v1/stmtbundle/765493679630483457 (Not available for CockroachDB serverless clusters.)
  SQL shell: \statement-diag download 765493679630483457
  Command line: cockroach statement-diag download 765493679630483457
(7 rows)

To download the ZIP file containing the statement diagnostics, open the URL after Direct link, run the \statement-diag download command, or run cockroach statement-diag download. You can also obtain the bundle by activating statement diagnostics in the DB Console.

See also


Yes No
On this page

Yes No