Difference between revisions of "Profiling Module"

From BaseX Documentation
Jump to navigation Jump to search
(23 intermediate revisions by 2 users not shown)
Line 9: Line 9:
 
==prof:track==
 
==prof:track==
  
{{Mark|Introduced with Version 9.0:}}
+
{{Mark|Updated with Version 9.2:}} Decimal timing results; by default no memory profiling.
  
 
{| width='100%'
 
{| width='100%'
 
|-
 
|-
 
| width='120' | '''Signatures'''
 
| width='120' | '''Signatures'''
|{{Func|prof:track|$expr as item()|item()*}}<br/>{{Func|prof:track|$expr as item(), $options as map(*)|item()*}}
+
|{{Func|prof:track|$expression as item()|item()*}}<br/>{{Func|prof:track|$expression as item(), $options as map(*)?|item()*}}
 
|-
 
|-
 
| '''Summary'''
 
| '''Summary'''
 
|Measures the execution time and memory consumption required for evaluating the specified {{Code|$expression}} and returns a map with the results. The following {{Code|$options}} are available:
 
|Measures the execution time and memory consumption required for evaluating the specified {{Code|$expression}} and returns a map with the results. The following {{Code|$options}} are available:
* {{Code|memory}}: Enables time profiling (default: true).
+
* {{Code|time}}: Include execution time in result as {{Code|xs:decimal}} (unit: milliseconds; default: true).
* {{Code|time}}: Enables memory profiling (default: true).
+
* {{Code|memory}}: Include memory consumption in result as {{Code|xs:integer}} (unit: bytes; default: false).
* {{Code|value}}: Returns the value (default: true).
+
* {{Code|value}}: Include value in result (default: true).
 
Helpful notes:
 
Helpful notes:
 +
* If you are not interested in some of the returned results, you should disable them to save time and memory.
 
* Profiling might change the execution behavior of your code: An expression that might be executed iteratively will be cached by the profiling function.
 
* Profiling might change the execution behavior of your code: An expression that might be executed iteratively will be cached by the profiling function.
 
* If a value has a compact internal representation, memory consumption will be very low, even if the serialized result may consume much more memory.
 
* If a value has a compact internal representation, memory consumption will be very low, even if the serialized result may consume much more memory.
* In Java, memory profiling can only be approximative: To improve the measurement, and if the memory option is enabled, main-memory will be garbage-collected before and after evaluation. If you are not interested in the memory consumption, it is advisable to disable this option.
+
* Please note that memory profiling is only approximative, so it can be quite misleading. If the memory option is enabled, main-memory will be garbage-collected before and after evaluation to improve the quality of the measurement.
 
|-
 
|-
 
| '''Properties'''
 
| '''Properties'''
Line 30: Line 31:
 
|-
 
|-
 
| '''Examples'''
 
| '''Examples'''
|The function call <code>prof:track((1 to 1000000)[. mod 2 = 0])</code> will return something similar to:
+
|
 +
* Return a human-readable representation of the memory consumption caused by fetching an XML document (<code>fetch:xml</code> is used, as <code>fn:doc</code> may already be evaluated at compilation time):
 +
<pre class="brush:xquery">
 +
prof:track(fetch:xml('factbook.xml'))?memory
 +
=> prof:human()
 +
</pre>
 +
* The function call <code>prof:track((1 to 1000000)[. mod 2 = 0], map { 'time': false() })</code> will return something similar to:
 
<pre class="brush:xquery">
 
<pre class="brush:xquery">
 
map {
 
map {
   "memory": 41548400,
+
   "memory": 21548400,
  "time": 34.99,
 
 
   "value": (2, 4, 6, 8, 10, ...)
 
   "value": (2, 4, 6, 8, 10, ...)
 
}
 
}
Line 41: Line 47:
  
 
==prof:time==
 
==prof:time==
 
{{Mark|Updated with Version 9.0:}} {{Code|$cache}} argument removed.
 
  
 
{| width='100%'
 
{| width='100%'
Line 50: Line 54:
 
|-
 
|-
 
| '''Summary'''
 
| '''Summary'''
|Measures the time needed to evaluate {{Code|$expr}} and outputs a string to standard error or, if the GUI is used, to the Info View.<br/>An optional {{Code|$label}} may be specified to tag the profiling result. Helpful notes:
+
|Measures the time needed to evaluate {{Code|$expr}} and outputs a string to standard error or, if the GUI is used, to the Info View. An optional {{Code|$label}} may be specified to tag the profiling result. See {{Function|Profiling|prof:track}} for further notes.
* Profiling might change the execution behavior of your code: An expression that might be executed iteratively will be cached by the profiling function.
 
* If a value has a compact internal representation, evaluation time will be very low, even if it may take a long time to iterate through all results.
 
 
|-
 
|-
 
| '''Properties'''
 
| '''Properties'''
Line 63: Line 65:
  
 
==prof:memory==
 
==prof:memory==
 
{{Mark|Updated with Version 9.0:}} Renamed (old name: <code>prof:mem</code>); {{Code|$cache}} argument removed.
 
  
 
{| width='100%'
 
{| width='100%'
Line 72: Line 72:
 
|-
 
|-
 
| '''Summary'''
 
| '''Summary'''
|Measures the memory allocated by evaluating {{Code|$expr}} and outputs a string to standard error or, if the GUI is used, to the Info View. An optional {{Code|$label}} may be specified to tag the profiling result. Helpful notes:
+
|Measures the memory allocated by evaluating {{Code|$expr}} and outputs a string to standard error or, if the GUI is used, to the Info View. An optional {{Code|$label}} may be specified to tag the profiling result. See {{Function|Profiling|prof:track}} for further notes.
* Profiling might change the execution behavior of your code: An expression that might be executed iteratively will be cached by the profiling function.
 
* If a value has a compact internal representation, memory consumption will be very low, even if the serialized result may consume much more memory.
 
* In Java, memory profiling can only be approximative: To improve the measurement, main-memory will be garbage-collected before and after evaluation.
 
 
|-
 
|-
 
| '''Properties'''
 
| '''Properties'''
Line 137: Line 134:
 
|-
 
|-
 
| width='120' | '''Signatures'''
 
| width='120' | '''Signatures'''
|{{Func|prof:dump|$expr as item()|empty-sequence()}}<br />{{Func|prof:dump|$expr as item(), $label as xs:string|empty-sequence()}}<br />
+
|{{Func|prof:dump|$expr as item()*|empty-sequence()}}<br />{{Func|prof:dump|$expr as item()*, $label as xs:string|empty-sequence()}}<br />
 
|-
 
|-
 
| '''Summary'''
 
| '''Summary'''
Line 154: Line 151:
 
|-
 
|-
 
| '''Summary'''
 
| '''Summary'''
|Prints a list of all current local and global variable assignments to standard error or, if the GUI is used, to the Info View.<br />As every query is optimized before being evaluated, not all of the original variables may be visible in the output. Moreover, many variables of function calls will disappear because functions are inlined. Function inlining can be turned off by setting the [[Options#INLINELIMIT|INLINELIMIT]] option to <code>0</code>.
+
|Prints a list of all current local and global variable assignments to standard error or, if the GUI is used, to the Info View.<br />As every query is optimized before being evaluated, not all of the original variables may be visible in the output. Moreover, many variables of function calls will disappear because functions are inlined. Function inlining can be turned off by setting {{Option|INLINELIMIT}} to <code>0</code>.
 
|-
 
|-
 
| '''Properties'''
 
| '''Properties'''
Line 161: Line 158:
 
| '''Examples'''
 
| '''Examples'''
 
|
 
|
* {{Code|for $x in 1 to 2 return prof:variables()}} will dump the values of <code>$x</code> to standard error.
+
* {{Code|for $x in 1 to 2 return ($x, prof:variables())}} will dump the values of <code>$x</code> to standard error.
 
|}
 
|}
  
Line 173: Line 170:
 
| '''Summary'''
 
| '''Summary'''
 
|Similar to {{Code|fn:trace($expr, $msg)}}, but instead of a user-defined message, it emits the compile-time type and estimated result size of its argument.
 
|Similar to {{Code|fn:trace($expr, $msg)}}, but instead of a user-defined message, it emits the compile-time type and estimated result size of its argument.
 +
|}
 +
 +
==prof:gc==
 +
 +
{{Mark|Introduced with Version 9.2.}}
 +
 +
{| width='100%'
 +
|-
 +
| width='120' | '''Signatures'''
 +
|{{Func|prof:gc||empty-sequence()}}<br/>{{Func|prof:gc|$count as xs:integer|empty-sequence()}}
 +
|-
 +
| '''Summary'''
 +
|Enforces Java garbage collection. If no {{Code|$count}} is supplied, garbage will be collected once. Please note that this function should only be used for debugging purposes; in productive code, it is best to trust the garbage collecting strategies of Java.
 +
|}
 +
 +
==prof:runtime==
 +
 +
{{Mark|Introduced with Version 9.2.}}
 +
 +
{| width='100%'
 +
|-
 +
| width='120' | '''Signatures'''
 +
|{{Func|prof:runtime|$name of xs:string|xs:integer}}
 +
|-
 +
| '''Summary'''
 +
|Returns the value of the specified runtime {{Code|$option}}. The following options exist:
 +
* {{Code|max}}: Maximum memory that the Java virtual machine will attempt to use.
 +
* {{Code|total}}: Total memory in the Java virtual machine (varies over time).
 +
* {{Code|used}}: Currently used memory (varies over time, will shrink after garbage collection).
 +
* {{Code|processors}}: number of processors available to the Java virtual machine.
 +
|-
 +
|{{Code|option}}
 +
|The specified option is unknown.
 +
|-
 +
| '''Examples'''
 +
|
 +
* <code>prof:gc(3), prof:human(prof:runtime('used'))</code> performs some garbage collection and returns the currently used amount of memory in a user-friendly format.
 
|}
 
|}
  
Line 222: Line 256:
 
|
 
|
 
* {{Code|prof:human(16384)}} returns {{Code|16K}}.
 
* {{Code|prof:human(16384)}} returns {{Code|16K}}.
 +
|}
 +
 +
=Errors=
 +
 +
{| class="wikitable" width="100%"
 +
! width="110"|Code
 +
|Description
 +
|-
 +
|{{Code|option}}
 +
|The specified option is unknown.
 
|}
 
|}
  
 
=Changelog=
 
=Changelog=
 +
 +
;Version 9.2
 +
 +
* Added: [[#prof:gc|prof:gc]], [[#prof:runtime|prof:runtime]]
 +
* Updated: [[#prof:track|prof:track]]: decimal timing results; by default no memory profiling
  
 
;Version 9.0
 
;Version 9.0
  
 
* Added: [[#prof:track|prof:track]]
 
* Added: [[#prof:track|prof:track]]
* Updated: renamed prof:mem to [[#prof:memory|prof:memory]]
+
* Updated: renamed prof:mem to [[#prof:memory|prof:memory]], [[#prof:time|prof:time]]: <code>$cache</code> argument removed
  
 
;Version 8.5
 
;Version 8.5

Revision as of 18:10, 7 November 2019

This XQuery Module contains various functions to test and profile code, and to dump information to standard output.

Conventions

All functions and errors in this module are assigned to the http://basex.org/modules/prof namespace, which is statically bound to the prof prefix.

Performance Functions

prof:track

Template:Mark Decimal timing results; by default no memory profiling.

Signatures prof:track($expression as item()) as item()*
prof:track($expression as item(), $options as map(*)?) as item()*
Summary Measures the execution time and memory consumption required for evaluating the specified $expression and returns a map with the results. The following $options are available:
  • time: Include execution time in result as xs:decimal (unit: milliseconds; default: true).
  • memory: Include memory consumption in result as xs:integer (unit: bytes; default: false).
  • value: Include value in result (default: true).

Helpful notes:

  • If you are not interested in some of the returned results, you should disable them to save time and memory.
  • Profiling might change the execution behavior of your code: An expression that might be executed iteratively will be cached by the profiling function.
  • If a value has a compact internal representation, memory consumption will be very low, even if the serialized result may consume much more memory.
  • Please note that memory profiling is only approximative, so it can be quite misleading. If the memory option is enabled, main-memory will be garbage-collected before and after evaluation to improve the quality of the measurement.
Properties The function is non-deterministic: evaluation order will be preserved by the compiler.
Examples
  • Return a human-readable representation of the memory consumption caused by fetching an XML document (fetch:xml is used, as fn:doc may already be evaluated at compilation time):
prof:track(fetch:xml('factbook.xml'))?memory
=> prof:human()
  • The function call prof:track((1 to 1000000)[. mod 2 = 0], map { 'time': false() }) will return something similar to:
map {
  "memory": 21548400,
  "value": (2, 4, 6, 8, 10, ...)
}

prof:time

Signatures prof:time($expr as item()) as item()*
prof:time($expr as item(), $label as xs:string) as item()*
Summary Measures the time needed to evaluate $expr and outputs a string to standard error or, if the GUI is used, to the Info View. An optional $label may be specified to tag the profiling result. See prof:track for further notes.
Properties The function is non-deterministic: evaluation order will be preserved by the compiler.
Examples
  • prof:time(prof:sleep(1000)) outputs something similar to 1000.99 ms.

prof:memory

Signatures prof:memory($expr as item()) as item()*
prof:memory($expr as item(), $label as xs:string) as item()*
Summary Measures the memory allocated by evaluating $expr and outputs a string to standard error or, if the GUI is used, to the Info View. An optional $label may be specified to tag the profiling result. See prof:track for further notes.
Properties The function is non-deterministic: evaluation order will be preserved by the compiler.
Examples
  • prof:memory((1 to 100000) ! <a/>) will output something similar to 5620 kB.

prof:current-ms

Signatures prof:current-ms() as xs:integer
Summary Returns the number of milliseconds passed since 1970/01/01 UTC. The granularity of the value depends on the underlying operating system and may be larger. For example, many operating systems measure time in units of tens of milliseconds.
Properties In contrast to fn:current-time(), the function is non-deterministic and returns different values every time it is called. Its evaluation order will be preserved by the compiler.
Examples
  • convert:integer-to-dateTime(prof:current-ms()) returns the current miliseconds in the xs:dateTime format.

prof:current-ns

Signatures prof:current-ns() as xs:integer
Summary Returns the current value of the most precise available system timer in nanoseconds.
Properties In contrast to fn:current-time(), the function is non-deterministic and returns different values every time it is called. Its evaluation order will be preserved by the compiler.
Examples Measures the time of an expression:
let $ns1 := prof:current-ns()
return (
  (: process to measure :)
  (1 to 1000000)[. = 0],
  let $ns2 := prof:current-ns()
  let $ms := ((($ns2 - $ns1) idiv 10000) div 100)
  return $ms || ' ms'
)

Debugging Functions

prof:dump

Signatures prof:dump($expr as item()*) as empty-sequence()
prof:dump($expr as item()*, $label as xs:string) as empty-sequence()
Summary Dumps a serialized representation of $expr to STDERR, optionally prefixed with $label, and returns an empty sequence. If the GUI is used, the dumped result is shown in the Info View.
Properties In contrast to fn:trace(), the consumed expression will not be passed on.

prof:variables

Signatures prof:variables() as empty-sequence()
Summary Prints a list of all current local and global variable assignments to standard error or, if the GUI is used, to the Info View.
As every query is optimized before being evaluated, not all of the original variables may be visible in the output. Moreover, many variables of function calls will disappear because functions are inlined. Function inlining can be turned off by setting INLINELIMIT to 0.
Properties The function is non-deterministic: evaluation order will be preserved by the compiler.
Examples
  • for $x in 1 to 2 return ($x, prof:variables()) will dump the values of $x to standard error.

prof:type

Signatures prof:type($expr as item()*) as item()*
Summary Similar to fn:trace($expr, $msg), but instead of a user-defined message, it emits the compile-time type and estimated result size of its argument.

prof:gc

Template:Mark

Signatures prof:gc() as empty-sequence()
prof:gc($count as xs:integer) as empty-sequence()
Summary Enforces Java garbage collection. If no $count is supplied, garbage will be collected once. Please note that this function should only be used for debugging purposes; in productive code, it is best to trust the garbage collecting strategies of Java.

prof:runtime

Template:Mark

Signatures prof:runtime($name of xs:string) as xs:integer
Summary Returns the value of the specified runtime $option. The following options exist:
  • max: Maximum memory that the Java virtual machine will attempt to use.
  • total: Total memory in the Java virtual machine (varies over time).
  • used: Currently used memory (varies over time, will shrink after garbage collection).
  • processors: number of processors available to the Java virtual machine.
option The specified option is unknown.
Examples
  • prof:gc(3), prof:human(prof:runtime('used')) performs some garbage collection and returns the currently used amount of memory in a user-friendly format.

Helper Functions

prof:void

Signatures prof:void($value as item()*) as empty-sequence()
Summary Swallows all items of the specified $value and returns an empty sequence. This function is helpful if some code needs to be evaluated and if the actual result is irrelevant.
Properties The function is non-deterministic: evaluation order will be preserved by the compiler.
Examples

prof:sleep

Signatures prof:sleep($ms as xs:integer) as empty-sequence()
Summary Sleeps for the specified number of milliseconds.
Properties The function is non-deterministic: evaluation order will be preserved by the compiler.

prof:human

Signatures prof:human($number as xs:integer) as xs:string
Summary Returns a human-readable representation of the specified $number.
Example
  • prof:human(16384) returns 16K.

Errors

Code Description
option The specified option is unknown.

Changelog

Version 9.2
Version 9.0
Version 8.5
Version 8.1
Version 7.7
Version 7.6
Version 7.5

This module was introduced with Version 7.3.