OpenSearch/docs/reference/transform/painless-examples.asciidoc

640 lines
21 KiB
Plaintext

[role="xpack"]
[testenv="basic"]
[[transform-painless-examples]]
= Painless examples for {transforms}
++++
<titleabbrev>Painless examples</titleabbrev>
++++
These examples demonstrate how to use Painless in {transforms}. You can learn
more about the Painless scripting language in the
{painless}/painless-guide.html[Painless guide].
* <<painless-top-hits>>
* <<painless-time-features>>
* <<painless-group-by>>
* <<painless-bucket-script>>
* <<painless-count-http>>
* <<painless-compare>>
* <<painless-web-session>>
NOTE: While the context of the following examples is the {transform} use case,
the Painless scripts in the snippets below can be used in other {es} search
aggregations, too.
[[painless-top-hits]]
== Getting top hits by using scripted metric aggregation
This snippet shows how to find the latest document, in other words the document
with the earliest timestamp. From a technical perspective, it helps to achieve
the function of a <<search-aggregations-metrics-top-hits-aggregation>> by using
scripted metric aggregation in a {transform}, which provides a metric output.
[source,js]
--------------------------------------------------
"aggregations": {
"latest_doc": {
"scripted_metric": {
"init_script": "state.timestamp_latest = 0L; state.last_doc = ''", <1>
"map_script": """ <2>
def current_date = doc['@timestamp'].getValue().toInstant().toEpochMilli();
if (current_date > state.timestamp_latest)
{state.timestamp_latest = current_date;
state.last_doc = new HashMap(params['_source']);}
""",
"combine_script": "return state", <3>
"reduce_script": """ <4>
def last_doc = '';
def timestamp_latest = 0L;
for (s in states) {if (s.timestamp_latest > (timestamp_latest))
{timestamp_latest = s.timestamp_latest; last_doc = s.last_doc;}}
return last_doc
"""
}
}
}
--------------------------------------------------
// NOTCONSOLE
<1> The `init_script` creates a long type `timestamp_latest` and a string type
`last_doc` in the `state` object.
<2> The `map_script` defines `current_date` based on the timestamp of the
document, then compares `current_date` with `state.timestamp_latest`, finally
returns `state.last_doc` from the shard. By using `new HashMap(...)` you copy
the source document, this is important whenever you want to pass the full source
object from one phase to the next.
<3> The `combine_script` returns `state` from each shard.
<4> The `reduce_script` iterates through the value of `s.timestamp_latest`
returned by each shard and returns the document with the latest timestamp
(`last_doc`). In the response, the top hit (in other words, the `latest_doc`) is
nested below the `latest_doc` field.
Check the
<<scripted-metric-aggregation-scope,scope of scripts>>
for detailed explanation on the respective scripts.
You can retrieve the last value in a similar way:
[source,js]
--------------------------------------------------
"aggregations": {
"latest_value": {
"scripted_metric": {
"init_script": "state.timestamp_latest = 0L; state.last_value = ''",
"map_script": """
def current_date = doc['date'].getValue().toInstant().toEpochMilli();
if (current_date > state.timestamp_latest)
{state.timestamp_latest = current_date;
state.last_value = params['_source']['value'];}
""",
"combine_script": "return state",
"reduce_script": """
def last_value = '';
def timestamp_latest = 0L;
for (s in states) {if (s.timestamp_latest > (timestamp_latest))
{timestamp_latest = s.timestamp_latest; last_value = s.last_value;}}
return last_value
"""
}
}
}
--------------------------------------------------
// NOTCONSOLE
[[painless-time-features]]
== Getting time features by using aggregations
This snippet shows how to extract time based features by using Painless in a
{transform}. The snippet uses an index where `@timestamp` is defined as a `date`
type field.
[source,js]
--------------------------------------------------
"aggregations": {
"avg_hour_of_day": { <1>
"avg":{
"script": { <2>
"source": """
ZonedDateTime date = doc['@timestamp'].value; <3>
return date.getHour(); <4>
"""
}
}
},
"avg_month_of_year": { <5>
"avg":{
"script": { <6>
"source": """
ZonedDateTime date = doc['@timestamp'].value; <7>
return date.getMonthValue(); <8>
"""
}
}
},
...
}
--------------------------------------------------
// NOTCONSOLE
<1> Name of the aggregation.
<2> Contains the Painless script that returns the hour of the day.
<3> Sets `date` based on the timestamp of the document.
<4> Returns the hour value from `date`.
<5> Name of the aggregation.
<6> Contains the Painless script that returns the month of the year.
<7> Sets `date` based on the timestamp of the document.
<8> Returns the month value from `date`.
[[painless-group-by]]
== Using Painless in `group_by`
It is possible to base the `group_by` property of a {transform} on the output of
a script. The following example uses the {kib} sample web logs dataset. The goal
here is to make the {transform} output easier to understand through normalizing
the value of the fields that the data is grouped by.
[source,console]
--------------------------------------------------
POST _transform/_preview
{
"source": {
"index": [ <1>
"kibana_sample_data_logs"
]
},
"pivot": {
"group_by": {
"agent": {
"terms": {
"script": { <2>
"source": """String agent = doc['agent.keyword'].value;
if (agent.contains("MSIE")) {
return "internet explorer";
} else if (agent.contains("AppleWebKit")) {
return "safari";
} else if (agent.contains('Firefox')) {
return "firefox";
} else { return agent }""",
"lang": "painless"
}
}
}
},
"aggregations": { <3>
"200": {
"filter": {
"term": {
"response": "200"
}
}
},
"404": {
"filter": {
"term": {
"response": "404"
}
}
},
"503": {
"filter": {
"term": {
"response": "503"
}
}
}
}
},
"dest": { <4>
"index": "pivot_logs"
}
}
--------------------------------------------------
// TEST[skip:setup kibana sample data]
<1> Specifies the source index or indices.
<2> The script defines an `agent` string based on the `agent` field of the
documents, then iterates through the values. If an `agent` field contains
"MSIE", than the script returns "Internet Explorer". If it contains
`AppleWebKit`, it returns "safari". It returns "firefox" if the field value
contains "Firefox". Finally, in every other case, the value of the field is
returned.
<3> The aggregations object contains filters that narrow down the results to
documents that contains `200`, `404`, or `503` values in the `response` field.
<4> Specifies the destination index of the {transform}.
The API returns the following result:
[source,js]
--------------------------------------------------
{
"preview" : [
{
"agent" : "firefox",
"200" : 4931,
"404" : 259,
"503" : 172
},
{
"agent" : "internet explorer",
"200" : 3674,
"404" : 210,
"503" : 126
},
{
"agent" : "safari",
"200" : 4227,
"404" : 332,
"503" : 143
}
],
"mappings" : {
"properties" : {
"200" : {
"type" : "long"
},
"agent" : {
"type" : "keyword"
},
"404" : {
"type" : "long"
},
"503" : {
"type" : "long"
}
}
}
}
--------------------------------------------------
// NOTCONSOLE
You can see that the `agent` values are simplified so it is easier to interpret
them. The table below shows how normalization modifies the output of the
{transform} in our example compared to the non-normalized values.
[width="50%"]
|===
| Non-normalized `agent` value | Normalized `agent` value
| "Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.1; SV1; .NET CLR 1.1.4322)" | "internet explorer"
| "Mozilla/5.0 (X11; Linux i686) AppleWebKit/534.24 (KHTML, like Gecko) Chrome/11.0.696.50 Safari/534.24" | "safari"
| "Mozilla/5.0 (X11; Linux x86_64; rv:6.0a1) Gecko/20110421 Firefox/6.0a1" | "firefox"
|===
[[painless-bucket-script]]
== Getting duration by using bucket script
This example shows you how to get the duration of a session by client IP from a
data log by using
<<search-aggregations-pipeline-bucket-script-aggregation,bucket script>>.
The example uses the {kib} sample web logs dataset.
[source,console]
--------------------------------------------------
PUT _transform/data_log
{
"source": {
"index": "kibana_sample_data_logs"
},
"dest": {
"index": "data-logs-by-client"
},
"pivot": {
"group_by": {
"machine.os": {"terms": {"field": "machine.os.keyword"}},
"machine.ip": {"terms": {"field": "clientip"}}
},
"aggregations": {
"time_frame.lte": {
"max": {
"field": "timestamp"
}
},
"time_frame.gte": {
"min": {
"field": "timestamp"
}
},
"time_length": { <1>
"bucket_script": {
"buckets_path": { <2>
"min": "time_frame.gte.value",
"max": "time_frame.lte.value"
},
"script": "params.max - params.min" <3>
}
}
}
}
}
--------------------------------------------------
// TEST[skip:setup kibana sample data]
<1> To define the length of the sessions, we use a bucket script.
<2> The bucket path is a map of script variables and their associated path to
the buckets you want to use for the variable. In this particular case, `min` and
`max` are variables mapped to `time_frame.gte.value` and `time_frame.lte.value`.
<3> Finally, the script substracts the start date of the session from the end
date which results in the duration of the session.
[[painless-count-http]]
== Counting HTTP responses by using scripted metric aggregation
You can count the different HTTP response types in a web log data set by using
scripted metric aggregation as part of the {transform}. You can achieve a similar
function with filter aggregations, check the
{ref}/transform-examples.html#example-clientips[Finding suspicious client IPs]
example for details.
The example below assumes that the HTTP response codes are stored as keywords in
the `response` field of the documents.
[source,js]
--------------------------------------------------
"aggregations": { <1>
"responses.counts": { <2>
"scripted_metric": { <3>
"init_script": "state.responses = ['error':0L,'success':0L,'other':0L]", <4>
"map_script": """ <5>
def code = doc['response.keyword'].value;
if (code.startsWith('5') || code.startsWith('4')) {
state.responses.error += 1 ;
} else if(code.startsWith('2')) {
state.responses.success += 1;
} else {
state.responses.other += 1;
}
""",
"combine_script": "state.responses", <6>
"reduce_script": """ <7>
def counts = ['error': 0L, 'success': 0L, 'other': 0L];
for (responses in states) {
counts.error += responses['error'];
counts.success += responses['success'];
counts.other += responses['other'];
}
return counts;
"""
}
},
...
}
--------------------------------------------------
// NOTCONSOLE
<1> The `aggregations` object of the {transform} that contains all aggregations.
<2> Object of the `scripted_metric` aggregation.
<3> This `scripted_metric` performs a distributed operation on the web log data
to count specific types of HTTP responses (error, success, and other).
<4> The `init_script` creates a `responses` array in the `state` object with
three properties (`error`, `success`, `other`) with long data type.
<5> The `map_script` defines `code` based on the `response.keyword` value of the
document, then it counts the errors, successes, and other responses based on the
first digit of the responses.
<6> The `combine_script` returns `state.responses` from each shard.
<7> The `reduce_script` creates a `counts` array with the `error`, `success`,
and `other` properties, then iterates through the value of `responses` returned
by each shard and assigns the different response types to the appropriate
properties of the `counts` object; error responses to the error counts, success
responses to the success counts, and other responses to the other counts.
Finally, returns the `counts` array with the response counts.
[[painless-compare]]
== Comparing indices by using scripted metric aggregations
This example shows how to compare the content of two indices by a {transform}
that uses a scripted metric aggregation.
[source,console]
--------------------------------------------------
POST _transform/_preview
{
"id" : "index_compare",
"source" : { <1>
"index" : [
"index1",
"index2"
],
"query" : {
"match_all" : { }
}
},
"dest" : { <2>
"index" : "compare"
},
"pivot" : {
"group_by" : {
"unique-id" : {
"terms" : {
"field" : "<unique-id-field>" <3>
}
}
},
"aggregations" : {
"compare" : { <4>
"scripted_metric" : {
"init_script" : "",
"map_script" : "state.doc = new HashMap(params['_source'])", <5>
"combine_script" : "return state", <6>
"reduce_script" : """ <7>
if (states.size() != 2) {
return "count_mismatch"
}
if (states.get(0).equals(states.get(1))) {
return "match"
} else {
return "mismatch"
}
"""
}
}
}
}
}
--------------------------------------------------
// TEST[skip:setup kibana sample data]
<1> The indices referenced in the `source` object are compared to each other.
<2> The `dest` index contains the results of the comparison.
<3> The `group_by` field needs to be a unique identifier for each document.
<4> Object of the `scripted_metric` aggregation.
<5> The `map_script` defines `doc` in the state object. By using
`new HashMap(...)` you copy the source document, this is important whenever you
want to pass the full source object from one phase to the next.
<6> The `combine_script` returns `state` from each shard.
<7> The `reduce_script` checks if the size of the indices are equal. If they are
not equal, than it reports back a `count_mismatch`. Then it iterates through all
the values of the two indices and compare them. If the values are equal, then it
returns a `match`, otherwise returns a `mismatch`.
[[painless-web-session]]
== Getting web session details by using scripted metric aggregation
This example shows how to derive multiple features from a single transaction.
Let's take a look on the example source document from the data:
.Source document
[%collapsible%open]
=====
[source,js]
--------------------------------------------------
{
"_index":"apache-sessions",
"_type":"_doc",
"_id":"KvzSeGoB4bgw0KGbE3wP",
"_score":1.0,
"_source":{
"@timestamp":1484053499256,
"apache":{
"access":{
"sessionid":"571604f2b2b0c7b346dc685eeb0e2306774a63c2",
"url":"http://www.leroymerlin.fr/v3/search/search.do?keyword=Carrelage%20salle%20de%20bain",
"path":"/v3/search/search.do",
"query":"keyword=Carrelage%20salle%20de%20bain",
"referrer":"http://www.leroymerlin.fr/v3/p/produits/carrelage-parquet-sol-souple/carrelage-sol-et-mur/decor-listel-et-accessoires-carrelage-mural-l1308217717?resultOffset=0&resultLimit=51&resultListShape=MOSAIC&priceStyle=SALEUNIT_PRICE",
"user_agent":{
"original":"Mobile Safari 10.0 Mac OS X (iPad) Apple Inc.",
"os_name":"Mac OS X (iPad)"
},
"remote_ip":"0337b1fa-5ed4-af81-9ef4-0ec53be0f45d",
"geoip":{
"country_iso_code":"FR",
"location":{
"lat":48.86,
"lon":2.35
}
},
"response_code":200,
"method":"GET"
}
}
}
}
...
--------------------------------------------------
// NOTCONSOLE
=====
By using the `sessionid` as a group-by field, you are able to enumerate events
through the session and get more details of the session by using scripted metric
aggregation.
[source,js]
--------------------------------------------------
POST _transform/_preview
{
"source": {
"index": "apache-sessions"
},
"pivot": {
"group_by": {
"sessionid": { <1>
"terms": {
"field": "apache.access.sessionid"
}
}
},
"aggregations": { <2>
"distinct_paths": {
"cardinality": {
"field": "apache.access.path"
}
},
"num_pages_viewed": {
"value_count": {
"field": "apache.access.url"
}
},
"session_details": {
"scripted_metric": {
"init_script": "state.docs = []", <3>
"map_script": """ <4>
Map span = [
'@timestamp':doc['@timestamp'].value,
'url':doc['apache.access.url'].value,
'referrer':doc['apache.access.referrer'].value
];
state.docs.add(span)
""",
"combine_script": "return state.docs;", <5>
"reduce_script": """ <6>
def all_docs = [];
for (s in states) {
for (span in s) {
all_docs.add(span);
}
}
all_docs.sort((HashMap o1, HashMap o2)->o1['@timestamp'].millis.compareTo(o2['@timestamp'].millis));
def size = all_docs.size();
def min_time = all_docs[0]['@timestamp'];
def max_time = all_docs[size-1]['@timestamp'];
def duration = max_time.millis - min_time.millis;
def entry_page = all_docs[0]['url'];
def exit_path = all_docs[size-1]['url'];
def first_referrer = all_docs[0]['referrer'];
def ret = new HashMap();
ret['first_time'] = min_time;
ret['last_time'] = max_time;
ret['duration'] = duration;
ret['entry_page'] = entry_page;
ret['exit_path'] = exit_path;
ret['first_referrer'] = first_referrer;
return ret;
"""
}
}
}
}
}
--------------------------------------------------
// NOTCONSOLE
<1> The data is grouped by `sessionid`.
<2> The aggregations counts the number of paths and enumerate the viewed pages
during the session.
<3> The `init_script` creates an array type `doc` in the `state` object.
<4> The `map_script` defines a `span` array with a timestamp, a URL, and a
referrer value which are based on the corresponding values of the document, then
adds the value of the `span` array to the `doc` object.
<5> The `combine_script` returns `state.docs` from each shard.
<6> The `reduce_script` defines various objects like `min_time`, `max_time`, and
`duration` based on the document fields, then declares a `ret` object, and
copies the source document by using `new HashMap ()`. Next, the script defines
`first_time`, `last_time`, `duration` and other fields inside the `ret` object
based on the corresponding object defined earlier, finally returns `ret`.
The API call results in a similar response:
[source,js]
--------------------------------------------------
{
"num_pages_viewed" : 2.0,
"session_details" : {
"duration" : 131374,
"first_referrer" : "https://www.bing.com/",
"entry_page" : "http://www.leroymerlin.fr/v3/p/produits/materiaux-menuiserie/porte-coulissante-porte-interieure-escalier-et-rambarde/barriere-de-securite-l1308218463",
"first_time" : "2017-01-10T21:22:52.982Z",
"last_time" : "2017-01-10T21:25:04.356Z",
"exit_path" : "http://www.leroymerlin.fr/v3/p/produits/materiaux-menuiserie/porte-coulissante-porte-interieure-escalier-et-rambarde/barriere-de-securite-l1308218463?__result-wrapper?pageTemplate=Famille%2FMat%C3%A9riaux+et+menuiserie&resultOffset=0&resultLimit=50&resultListShape=PLAIN&nomenclatureId=17942&priceStyle=SALEUNIT_PRICE&fcr=1&*4294718806=4294718806&*14072=14072&*4294718593=4294718593&*17942=17942"
},
"distinct_paths" : 1.0,
"sessionid" : "000046f8154a80fd89849369c984b8cc9d795814"
},
{
"num_pages_viewed" : 10.0,
"session_details" : {
"duration" : 343112,
"first_referrer" : "https://www.google.fr/",
"entry_page" : "http://www.leroymerlin.fr/",
"first_time" : "2017-01-10T16:57:39.937Z",
"last_time" : "2017-01-10T17:03:23.049Z",
"exit_path" : "http://www.leroymerlin.fr/v3/p/produits/porte-de-douche-coulissante-adena-e168578"
},
"distinct_paths" : 8.0,
"sessionid" : "000087e825da1d87a332b8f15fa76116c7467da6"
}
...
--------------------------------------------------
// NOTCONSOLE