Any ways to achieve sql features like stored procedure or functions in sparksql?
I'm aware about hpl sql and coprocessor in hbase. But want to know if anything similar like is available in spark or not.
You may consider of use User Defined Function and inbuilt function
A quick example
val dataset = Seq((0, "hello"), (1, "world")).toDF("id", "text")
val upper: String => String = _.toUpperCase
import org.apache.spark.sql.functions.udf
val upperUDF = udf(upper)
// Apply the UDF to change the source dataset
scala> dataset.withColumn("upper", upperUDF('text)).show
Result
| id| text|upper|
+---+-----+-----+
| 0|hello|HELLO|
| 1|world|WORLD|
We cannot create SP/Functions in SparkSql. However best way is to create a temp table just like CTE and used those tables for further usage. Or you can create a UDF Function in Spark.
Related
I have an Oracle database that contains multiple users/schemas and I would like to generate Slick Schemas automatically for a specific user. This is what I've tried so far :
import scala.concurrent.ExecutionContext.Implicits.global
val profileInstance: JdbcProfile =
Class.forName("slick.jdbc.OracleProfile$")
.getField("MODULE$")
.get(null).asInstanceOf[JdbcProfile]
val db = profileInstance.api.Database
.forURL("jdbc:oracle:thin:#//myhost:myport/servicename","user","pass")
val modelAction = OracleProfile.createModel(Some(OracleProfile.defaultTables))
val model = Await.result(db.run(modelAction), Duration.Inf)
model.tables.foreach(println)
This doesn't print anything, I guess I have to provide the current schema to use, but I don't know how to do this.
On the other hand, I am able to list all the schemas of the database, using the following code :
val resultSet = db.createSession().metaData.getSchemas.getStatement.getResultSet
while(resultSet.next()) {
println(resultSet.getString(1))
}
How can I specify which schema I want to use with Slick ?
I've found out how to do it. Instead of using OracleProfile.defaultTable I manually defined the tables and views I needed like this :
val modelAction = OracleProfile.createModel(
Some(MTable.getTables(None, Some("MYSCHEMA"), None, Some(Seq("TABLE", "VIEW"))))
)
I have a dataset in BigQuery. This dataset contains multiple tables.
I am doing the following steps programmatically using the BigQuery API:
Querying the tables in the dataset - Since my response is too large, I am enabling allowLargeResults parameter and diverting my response to a destination table.
I am then exporting the data from the destination table to a GCS bucket.
Requirements:
Suppose my process fails at Step 2, I would like to re-run this step.
But before I re-run, I would like to check/verify that the specific destination table named 'xyz' already exists in the dataset.
If it exists, I would like to re-run step 2.
If it does not exist, I would like to do foo.
How can I do this?
Thanks in advance.
Alex F's solution works on v0.27, but will not work on later versions. In order to migrate to v0.28+, the below solution will work.
from google.cloud import bigquery
project_nm = 'gc_project_nm'
dataset_nm = 'ds_nm'
table_nm = 'tbl_nm'
client = bigquery.Client(project_nm)
dataset = client.dataset(dataset_nm)
table_ref = dataset.table(table_nm)
def if_tbl_exists(client, table_ref):
from google.cloud.exceptions import NotFound
try:
client.get_table(table_ref)
return True
except NotFound:
return False
if_tbl_exists(client, table_ref)
Here is a python snippet that will tell whether a table exists (deleting it in the process--careful!):
def doesTableExist(project_id, dataset_id, table_id):
bq.tables().delete(
projectId=project_id,
datasetId=dataset_id,
tableId=table_id).execute()
return False
Alternately, if you'd prefer not deleting the table in the process, you could try:
def doesTableExist(project_id, dataset_id, table_id):
try:
bq.tables().get(
projectId=project_id,
datasetId=dataset_id,
tableId=table_id).execute()
return True
except HttpError, err
if err.resp.status <> 404:
raise
return False
If you want to know where bq came from, you can call build_bq_client from here: http://code.google.com/p/bigquery-e2e/source/browse/samples/ch12/auth.py
In general, if you're using this to test whether you should run a job that will modify the table, it can be a good idea to just do the job anyway, and use WRITE_TRUNCATE as a write disposition.
Another approach can be to create a predictable job id, and retry the job with that id. If the job already exists, the job already ran (you might want to double check to make sure the job didn't fail, however).
Enjoy:
def doesTableExist(bigquery, project_id, dataset_id, table_id):
try:
bigquery.tables().get(
projectId=project_id,
datasetId=dataset_id,
tableId=table_id).execute()
return True
except Exception as err:
if err.resp.status != 404:
raise
return False
There is an edit in exception.
you can use exists() now to check if dataset exists same with table
BigQuery exist documentation
recently big query introduced so called scripting statements that can be quite a game changer for some flows.
check them out here:
https://cloud.google.com/bigquery/docs/reference/standard-sql/scripting
Now for example to check if table exists you can use something like this:
sql = """
BEGIN
IF EXISTS(SELECT 1 from `YOUR_PROJECT.YOUR_DATASET.YOUR_TABLE) THEN
SELECT 'table_found';
END IF;
EXCEPTION WHEN ERROR THEN
# you can print your own message like above or return error message
# however google says not to rely on error message structure as it may change
select ##error.message;
END;
"""
With my_bigquery being an instance of class google.cloud.bigquery.Client (already authentified and associated to a project):
my_bigquery.dataset(dataset_name).table(table_name).exists() # returns boolean
It does an API call to test for the existence of the table via a GET request
Source: https://googlecloudplatform.github.io/google-cloud-python/0.24.0/bigquery-table.html#google.cloud.bigquery.table.Table.exists
It works for me using 0.27 of the Google Bigquery Python module
Inline SQL Alternative
tarheel's answer is probably the most correct at this point in time
but I was considering the comment from Ivan above that "404 could also mean the resource is not there for a bunch of reasons", so here is a solution that should always successfully run a metadata query and return a result.
It's not the fastest, because it always has to run the query, bigquery has overhead for small queries
A trick I've seen previously is to query information_schema for a (table) object, and union that to a fake query that ensures a record is always returned even if the the object doesn't. There's also a LIMIT 1 and an ordering to ensure the single record returned represents the table, if it does exist. See the SQL in the code below.
In spite of doc claims that Bigquery standard SQL is ISO compliant, they don't support information_schema, but they do have __table_summary__
dataset is required because you can't query __table_summary__ without specifying dataset
dataset is not a parameter in the SQL because you can't parameterize object names without sql injection issues (apart from with the magical _TABLE_SUFFIX, see https://cloud.google.com/bigquery/docs/querying-wildcard-tables )
#!/usr/bin/env python
"""
Inline SQL way to check a table exists in Bigquery
e.g.
print(table_exists(dataset_name='<dataset_goes_here>', table_name='<real_table_name'))
True
print(table_exists(dataset_name='<dataset_goes_here>', table_name='imaginary_table_name'))
False
"""
from __future__ import print_function
from google.cloud import bigquery
def table_exists(dataset_name, table_name):
client = bigquery.Client()
query = """
SELECT table_exists FROM
(
SELECT true as table_exists, 1 as ordering
FROM __TABLES_SUMMARY__ WHERE table_id = #table_name
UNION ALL
SELECT false as table_exists, 2 as ordering
) ORDER by ordering LIMIT 1"""
query_params = [bigquery.ScalarQueryParameter('table_name', 'STRING', table_name)]
job_config = bigquery.QueryJobConfig()
job_config.query_parameters = query_params
if dataset_name is not None:
dataset_ref = client.dataset(dataset_name)
job_config.default_dataset = dataset_ref
query_job = client.query(
query,
job_config=job_config
)
results = query_job.result()
for row in results:
# There is only one row because LIMIT 1 in the SQL
return row.table_exists
I am trying to use PIG to read data from HDFS where the files contain rows that look like:
"key1"="value1", "key2"="value2", "key3"="value3"
"key1"="value10", "key3"="value30"
In a way the rows of the data are essentially dictionaries:
{"key1":"value1", "key2":"value2", "key3":"value3"}
{"key1":"value10", "key3":"value30"}
I can read and dump portion of this data easily enough with something like:
data = LOAD '/hdfslocation/weirdformat*' as PigStorage(',');
sampled = SAMPLE data 0.00001;
dump sampled;
My problem is that I can't parse it efficiently. I have tried to use
org.apache.pig.piggybank.storage.MyRegExLoader
but it seems extremely slow.
Could someone recommend a different approach?
Seems like one way is to use a python UDF.
This solution is heavily inspired from bag-to-tuple
In myudfs.py write:
#!/usr/bin/python
def FieldPairsGenerator(dataline):
for x in dataline.split(','):
k,v = x.split('=')
yield (k.strip().strip('"'),v.strip().strip('"'))
#outputSchema("foo:map[]")
def KVDataToDict(dataline):
return dict( kvp for kvp in FieldPairsGenerator(dataline) )
then write the following Pig script:
REGISTER 'myudfs.py' USING jython AS myfuncs;
data = LOAD 'whereyourdatais*.gz' AS (foo:chararray);
A = FOREACH data GENERATE myfuncs.KVDataToDict(foo);
A now has the data stored as a PigMap
I am using HDP 2.0 and running a simple Pig Script.
I have registered the below jars and I am then executing the below code (updated the schema) -
register /usr/lib/pig/piggybank.jar;
register /usr/lib/hive/lib/hive-common-0.11.0.2.0.5.0-67.jar;
register /usr/lib/hive/lib/hive-exec-0.11.0.2.0.5.0-67.jar;
A = LOAD '/apps/hive/warehouse/test.db/hivetables' USING
org.apache.pig.piggybank.storage.HiveColumnarLoader('id int, name string,age
int,create_dt string,timestamp string,accno int');
F = FILTER A BY (id == 85986249 );
STORE F INTO '/user/test/Pigout' USING PigStorage();
The problem is , Though the value for F is available in the Hive table, the result always writes 0 records into the output. But it is able to load all the records into A.
Basically the Filter function is not working. My Hive table is not partitioned. I beleive that the problem could be in HiveColumarLoade but not able to figure out what it is.
Please let me know if you are aware of a solution. I am struggling a lot with this.
Thanks a lot for the help!!!
Based on the pig 0.12 documentation HiveColumnarLoader appears to require an intermediate relation before you can filter on a non-partition value. Given that id is not a partition that appears to be your problem.
try this:
A = LOAD '/apps/hive/warehouse/test.db/hivetables' USING
org.apache.pig.piggybank.storage.HiveColumnarLoader('id int, name string,age
int,create_dt string,timestamp string,accno int');
B = FOREACH GENERATE A.id, A.name, A.age, A.create_dt, A.timestamp, A.accno;
F = FILTER A BY (id == 85986249 );
STORE F INTO '/user/test/Pigout' USING PigStorage();
The documentation all seems to say that for processing the actual values you need intermediate relation B.
I have an external table in hive
CREATE EXTERNAL TABLE FOO (
TS string,
customerId string,
products array< struct <productCategory:string, productId:string> >
)
PARTITIONED BY (ds string)
ROW FORMAT SERDE 'some.serde'
WITH SERDEPROPERTIES ('error.ignore'='true')
LOCATION 'some_locations'
;
A record of the table may hold data such as:
1340321132000, 'some_company', [{"productCategory":"footwear","productId":"nik3756"},{"productCategory":"eyewear","productId":"oak2449"}]
Do anyone know if there is a way to simply extract all the productCategory from this record and return it as an array of productCategories without using explode. Something like the following:
["footwear", "eyewear"]
Or do I need to write my own GenericUDF, if so, I do not know much Java (a Ruby person), can someone give me some hints? I read some instructions on UDF from Apache Hive. However, I do not know which collection type is best to handle array, and what collection type to handle structs?
===
I have somewhat answered this question by writing a GenericUDF, but I ran into 2 other problems. It is in this SO Question
You can use json serde or build-in functions get_json_object, json_tuple.
With rcongiu's Hive-JSON SerDe the usage will be:
define table:
CREATE TABLE complex_json (
DocId string,
Orders array<struct<ItemId:int, OrderDate:string>>)
load sample json into it (it is important for this data to be one-lined):
{"DocId":"ABC","Orders":[{"ItemId":1111,"OrderDate":"11/11/2012"},{"ItemId":2222,"OrderDate":"12/12/2012"}]}
Then fetching orders ids is as easy as:
SELECT Orders.ItemId FROM complex_json LIMIT 100;
It will return the list of ids for you:
itemid
[1111,2222]
Proven to return correct results on my environment. Full listing:
add jar hdfs:///tmp/json-serde-1.3.6.jar;
CREATE TABLE complex_json (
DocId string,
Orders array<struct<ItemId:int, OrderDate:string>>
)
ROW FORMAT SERDE 'org.openx.data.jsonserde.JsonSerDe';
LOAD DATA INPATH '/tmp/test.json' OVERWRITE INTO TABLE complex_json;
SELECT Orders.ItemId FROM complex_json LIMIT 100;
Read more here:
http://thornydev.blogspot.com/2013/07/querying-json-records-via-hive.html
One way would be to use either the inline or explode functions, like so:
SELECT
TS,
customerId,
pCat,
pId,
FROM FOO
LATERAL VIEW inline(products) p AS pCat, pId
Otherwise you can write UDF. Check out this post and this post for that. Along with the following resources:
Matthew Rathbone's guide to writing generic UDFs
Mark Grover's how to guide
the baynote blog post on generic UDFs
If size of array is fixed ( like 2 ). Please try:
products[0].productCategory,products[1].productCategory
But if not, UDF should be the right solution. I guess that you could do it in JRuby. GL!