We are attempting to create a schema to load a massive JSON structure into Hive. We are having a problem, however, in that some fields have leading underscores for names--at the root level, this is fine, but we have not found a way to make this work for nested fields.
Sample JSON:
{
"_id" : "319FFE15FF908EDD86B7FDEADBEEFBD8D7284128841B14AA6A966923C268DF39",
"SomeThing" :
{
"_SomeField" : 22,
"AnotherField" : 2112,
"YetAnotherField": 1
}
. . . etc . . . .
Using a schema as follows:
create table testSample
(
id string,
something struct
<
somefield:int,
anotherfield:bigint,
yetanotherfield:int
>
)
row format serde 'org.openx.data.jsonserde.JsonSerDe'
with serdeproperties
(
"mapping.id" = "_id",
"mapping.somefield" = "_somefield"
);
This schema builds OK--however, after loading the in above sample, the value of "somefield" (the nested + leading underscore one) is always null (all the other values exist and are correct).
We've been trying a lot of syntax combinations, but to no avail.
Does anyone know the trick to hap a nested field with a leading underscore in its name?
Cheers!
Answering my own question here: there is no trick because you can't.
However, there's an easy work-around: you can tell Hive to treat the names as literals upon creating the schema. If you do this, you will also need to query using the same literal syntax. In the above example, it would look like:
`_something` struct<rest_of_definitions>
without any special serde properties for it.
Then use again in query:
select stuff.`_something` from sometable;
e.g., schema:
create table testSample
(
id string,
something struct
<
`_somefield`:int,
anotherfield:bigint,
yetanotherfield:int
>
)
row format serde 'org.openx.data.jsonserde.JsonSerDe'
with serdeproperties("mapping.id" = "_id");
for an input JSON like:
{
"_id": "someuid",
"something":
{
"_somefield": 1,
"anotherfield": 2,
"yetanotherfield": 3
}
}
with a query like:
select something.`_somefield`
from testSample
where something.anotherfield = 2;
Related
Given the table:
C1 C2 C3
----------------
1 'v1' 1.1
2 'v2' 2.2
3 'v3' 3.3
Is there any "easy" way to return JSON in this format:
{
"columns": [ "C1", "C2", "C3" ],
"rows": [
[ 1, "v1", 1.1 ],
[ 2, "v2", 2.2 ],
[ 3, "v3", 3.3 ]
]
}
To generate an array with single values from a table there is a neat trick like this:
SELECT JSON_QUERY(REPLACE(REPLACE(
(
SELECT id
FROM table a
WHERE pk in (1,2)
FOR JSON PATH
), '{"id":',''),'}','')) 'ids'
Which generates
"ids": [1,2]
But to construct the nested array above the replacing gets really tedious, anyone know a good way to achieve this?
Well, you ask for an easy way but the following will not be easy :-)
The tricky part is to know which values need to be qouted and which can remain naked.
This needs generic type-analysis to find, which values are strings.
The only way I know to get on meta data (besides building dynamic sql using meta views like INFORMATIONSCHEMA.COLUMNS) is XML together with an AUTO-schema.
This XML is very near to your needs actually. There is a list of columns at the beginning, followed by a list of rows. But it is not JSON of course...
Try this out:
--This is a mockup table with the values you provided.
DECLARE #mockup TABLE(C1 INT,C2 VARCHAR(100),C3 DECIMAL(4,2));
INSERT INTO #mockup VALUES
(1,'v1',1.1)
,(2,'v2',2.2)
,(3,'v3',3.3);
--Now we create an XML out of this
DECLARE #xml XML =
(
SELECT *
FROM #mockup t
FOR XML RAW,XMLSCHEMA,TYPE
);
--Check the XML's content with SELECT #xml to see how it is looking internally
--Now the real query can start:
SELECT '{"columns":[' +
STUFF(#xml.query('declare namespace xsd="http://www.w3.org/2001/XMLSchema";
for $col in /xsd:schema/xsd:element//xsd:attribute
return
<x>,{concat("""",xs:string($col/#name),"""")}</x>
').value('.','nvarchar(max)'),1,1,'') +
'],"rows":[' +
STUFF(
(
SELECT
',[' + STUFF(b.query(' declare namespace xsd="http://www.w3.org/2001/XMLSchema";
for $attr in ./#*
return
<x>,{if(/xsd:schema/xsd:element//xsd:attribute[#name=local-name($attr)]//xsd:restriction/#base="sqltypes:varchar") then
concat("""",$attr,"""")
else
xs:string($attr)
}
</x>
').value('.','nvarchar(max)'),1,1,'') + ']'
FROM #xml.nodes('/*:row') B(b)
FOR XML PATH(''),TYPE
).value('.','nvarchar(max)'),1,1,'') +
']}';
The result
{"columns":["C1","C2","C3"],"rows":[[3,"v3",3.30],[1,"v1",1.10],[2,"v2",2.20]]}
Some explanation:
The first part will use XQuery to find all columns (xsd:attribute within XML-schema) and create the array of column names.
The second part will againt use XQuery in order to run through all rows and write their column values in a concatenated string. Each value can refer to its type within the schema. Whenever this type is sqltypes:varchar the value will be quoted. All other values remain naked.
This will not solve each and any case generically...
To be honest, this was more for my own curiosity :-) Wanted to find out, how one can solve this.
Quite probably the best answer is: Use another tool. SQL-Server is not the best choice here ;-)
I have a postgresql table called datasource with jsonb column called config. It has the following structure:
{
"url":"some_url",
"password":"some_password",
"username":"some_username",
"projectNames":[
"project_name_1",
...
"project_name_N"
]
}
I would like to transform nested json array projectNames into a map and add a default value for each element from the array, so it would look like:
{
"url":"some_url",
"password":"some_password",
"username":"some_username",
"projectNames":{
"project_name_1": "value",
...
"project_name_N": "value"
}
}
I have selected projectNames from the table using postgresql jsonb operator config#>'{projectNames}', but I have no idea how to perform transform operation.
I think, I should use something like jsonb_object_agg, but it converts all data into a single row.
I'm using PostgreSQL 9.6 version.
You need to first unnest the array, then build a new JSON document from that. Then you can put that back into the column.
update datasource
set config = jsonb_set(config, '{projectNames}', t.map)
from (
select id, jsonb_object_agg(pn.n, 'value') as map
from datasource, jsonb_array_elements_text(config -> 'projectNames') as pn (n)
group by id
) t
where t.id = datasource.id;
The above assumes that there is a primary (or at least unique) column named id. The inner select transforms the array into a map.
Online example: http://rextester.com/GPP85654
are you looking for smth like:
t=# with c(j) as (values('{
"url":"some_url",
"password":"some_password",
"username":"some_username",
"projectNames":[
"project_name_1",
"project_name_N"
]
}
'::jsonb))
, n as (select j,jsonb_array_elements_text(j->'projectNames') a from c)
select jsonb_pretty(jsonb_set(j,'{projectNames}',jsonb_object_agg(a,'value'))) from n group by j
;
jsonb_pretty
------------------------------------
{ +
"url": "some_url", +
"password": "some_password", +
"username": "some_username", +
"projectNames": { +
"project_name_1": "value",+
"project_name_N": "value" +
} +
}
(1 row)
Time: 19.756 ms
if so, look at:
https://www.postgresql.org/docs/current/static/functions-aggregate.html
https://www.postgresql.org/docs/current/static/functions-json.html
I have an external table that is built off of a json file.
All of the json keys are columns and are populated as expected except for one key that has a space.
Here is the DDL:
CREATE EXTERNAL TABLE foo.bar
( event ARRAY <STRUCT
value:STRING
,info:STRUCT
<id:STRING
,event_source:STRING>>
ROW FORMAT SERDE 'org.openx.data.jsonserde.JsonSerDe'
WITH SERDEPROPERTIES("mapping.event_source"="event source")
STORED AS INPUTFORMAT 'org.apache.hadoop.mapred.TextInputFormat'
OUTPUTFORMAT 'org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat'
LOCATION 'foo/bar'
All of the values show up as expected except for event_source, which shows up as NULL. The original form of event_source in the json file is 'event source' without the single quotes.
Is there something I need to do different with the WITH SERDEPROPERTIES setting in order to get the key to work properly?
Thanks
you mean that the json has data like
{ id: "myid", event source: "eventsource" }
If so, there's not much that can be done since it's simply broken JSON.
If not, can you post a sample of the JSON you're trying to read ?
I have encountered a similar problem as above but with a slight variation that the input data is correct json.
I have an external table that is built off of a json file. All of the json keys are populated except one msrp_currency
Here is the DDL:
CREATE EXTERNAL TABLE foo.bar
( id string,
variants array<struct<pid:string, msrp_currency:string>>
)
ROW FORMAT SERDE 'org.openx.data.jsonserde.JsonSerDe'
WITH SERDEPROPERTIES ( "ignore.malformed.json" = "true" ,
'mapping.variants.msrp_currency' = 'variants.msrpcurrency')
LOCATION 'foo/bar'
All of the values show up as expected except for msrp_currency, which shows up as NULL. The reason I need to introduce underscore is because later I need to extract the same field value as msrpCurrecny using brickhouse to_json UDF.
sample values:
{ "pid": "mypid", "msrpCurrency": "USD" }
I am trying to load a huge volume json data with nested structure to hive using a Json serde. some of the field names start with $ in nested structure. I am mapping hive filed names Using SerDeproperties, but how ever when i query the table, getting null in the field starting with $, tried with different syntax,but no luck.
Sample JSON:
{
"_id" : "319FFE15FF90",
"SomeThing" :
{
"$SomeField" : 22,
"AnotherField" : 2112,
"YetAnotherField": 1
}
. . . etc . . . .
Using a schema as follows:
create table testSample
(
`_id` string,
something struct
<
$somefield:int,
anotherfield:bigint,
yetanotherfield:int
>
)
row format serde 'org.openx.data.jsonserde.JsonSerDe'
with serdeproperties
(
"mapping.somefield" = "$somefield"
);
This schema builds OK, however, somefield(starting with $) in the above table is always returning null (all the other values exist and are correct).
We've been trying a lot of syntax combinations, but to no avail.
Does anyone know the trick to hap a nested field with a leading $ in its name?
You almost got it right. Try creating the table like this.
The mistake you're making is that when mapping in the serde properties (mapping.somefield ="$somefield") you're saying "when looking for the hive column named 'somefield', look for the json field '$somefield', but in hive you defined the column with the dollar sign, which if not outright illegal it's for sure not the best practice in hive.
create table testSample
(
`_id` string,
something struct
<
somefield:int,
anotherfield:bigint,
yetanotherfield:int
>
)
row format serde 'org.openx.data.jsonserde.JsonSerDe'
with serdeproperties
(
"mapping.somefield" = "$somefield"
);
I tested it with some test data:
{ "_id" : "123", "something": { "$somefield": 12, "anotherfield":13,"yetanotherfield":100}}
hive> select something.somefield from testSample;
OK
12
I am suddenly starting to see this problem as well but for normal column names as well (no special characters such as $)
I am populating an external table (Temp) from another internal table (Table2) and want the output of Temp table in JSON format. I want column names in camel case in the output JSON file and so am also using the Serdepoperties in the Temp table to specify correct names. However, I am seeing that when I do Select * from the Temp table, it gives NULL values for the columns whose names have been used in the mapping.
I am running Hive 0.13. Here are the commands:
Create table command:
CREATE EXTERNAL TABLE Temp (
data STRUCT<
customerId:BIGINT, region:STRING, marketplaceId:INT, asin:ARRAY<STRING>>
)
ROW FORMAT SERDE 'org.openx.data.jsonserde.JsonSerDe'
WITH SERDEPROPERTIES (
'mapping.customerid' = 'customerId',
'mapping.marketplaceid' = 'marketplaceId'
)
LOCATION '/output';
INSERT INTO TABLE Temp
SELECT
named_struct ('customerId',customerId, 'region', region, 'marketplaceId', marketplaceId, 'asin', asin)
FROM Table2;
Select * from Temp:
{"customerid":null,"region":"EU","marketplaceid":null,"asin":["B000FC1PZC"]}
{"customerid":null,"region":"EU","marketplaceid":null,"asin":["B000FC1C9G"]}
See how "customerid" and "marketplaceid" are null. Generated JSON file is:
{"data":{"region":"EU","asin":["B000FC1PZC"]}}
{"data":{"region":"EU","asin":["B000FC1C9G"]}}
Now, if I remove the with serdeproperties, the table starts getting all values:
{"customerid":1,"region":"EU","marketplaceid":4,"asin":["B000FC1PZC"]}
{"customerid":2,"region":"EU","marketplaceid":4,"asin":["B000FC1C9G"]}
And then the JSON file so generated is:
{"data":{"region":"EU","marketplaceid":4,"asin":["B000FC1PZC"],"customerid":1}}
{"data":{"region":"EU","marketplaceid":4,"asin":["B000FC1C9G"],"customerid":2}}
I am trying to get some values out of nested JSON for millions of rows (5 TB+ table). What is the most efficient way to do this?
Here is an example:
{"country":"US","page":227,"data":{"ad":{"impressions":{"s":10,"o":10}}}}
I need these values out of the above JSON:
Country Page impressions_s impressions_o
--------- ----- ------------- --------------
US 2 10 10
This is Hive's json_tuple function, I am not sure if this is the best function.
https://cwiki.apache.org/confluence/display/Hive/LanguageManual+UDF#LanguageManualUDF-getjsonobject
You can use get_json_object:
select get_json_object(fieldname, '$.country'),
get_json_object(fieldname, '$.data.ad.s') from ...
You will get better performance with json_tuple but I found a "how to" to get the values in json inside json;
To formating your table you can use something like this:
from table t lateral view
explode( split(regexp_replace(get_json_object(ln, ''$.data.ad.s'), '\\[|\\]', ''), ',' ) ) tb1 as s
this code above will transform you "Array" in a column.
form more: https://cwiki.apache.org/confluence/display/Hive/LanguageManual+UDF
I hope this help ...
Here is what you can quickly try , I would suggest to use Json-Ser-De.
nano /tmp/hive-parsing-json.json
{"country":"US","page":227,"data":{"ad":{"impressions":{"s":10,"o":10}}}}
Create base table :
hive > CREATE TABLE hive_parsing_json_table ( json string );
Load json file to Table :
hive > LOAD DATA LOCAL INPATH '/tmp/hive-parsing-json.json' INTO TABLE hive_parsing_json_table;
Query the table :
hive > select v1.Country, v1.Page, v4.impressions_s, v4.impressions_o
from hive_parsing_json_table hpjp
LATERAL VIEW json_tuple(hpjp.json, 'country', 'page', 'data') v1
as Country, Page, data
LATERAL VIEW json_tuple(v1.data, 'ad') v2
as Ad
LATERAL VIEW json_tuple(v2.Ad, 'impressions') v3
as Impressions
LATERAL VIEW json_tuple(v3.Impressions, 's' , 'o') v4
as impressions_s,impressions_o;
Output :
v1.country v1.page v4.impressions_s v4.impressions_o
US 227 10 10
Using hive native json-serde('org.apache.hive.hcatalog.data.JsonSerDe') you can do this.. here are the steps
ADD JAR /path/to/hive-hcatalog-core.jar;
create a table as below
CREATE TABLE json_serde_nestedjson (
country string,
page int,
data struct < ad: struct < impressions: struct < s:int, o:int > > >
)
ROW FORMAT SERDE 'org.apache.hive.hcatalog.data.JsonSerDe';
then load data(stored in file)
LOAD DATA LOCAL INPATH '/tmp/nested.json' INTO TABLE json_serde_nestedjson;
then get required data using
SELECT country, page, data.ad.impressions.s, data.ad.impressions.o
FROM json_serde_nestedjson;
Implementing a SerDe to parse your data in JSON is a better way for your case.
A tutorial on how to implement SerDe for parsing JSON can be found here
http://blog.cloudera.com/blog/2012/12/how-to-use-a-serde-in-apache-hive/
You can use the following sample SerDe implementation as well
https://github.com/rcongiu/Hive-JSON-Serde