spark-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Mich Talebzadeh <mich.talebza...@gmail.com>
Subject Re: How to generate unique incrementing identifier in a structured streaming dataframe
Date Thu, 15 Jul 2021 21:03:31 GMT
Yes that is true. UUID only introduces uniqueness to the record. Some NoSql
databases requires a primary key where UUID can be used.


import java.util.UUID

scala> var pk = UUID.randomUUID

pk: java.util.UUID = 0d91e11a-f5f6-4b4b-a120-8c46a31dad0bscala>

pk = UUID.randomUUID
pk: java.util.UUID = 137ab1ef-625a-4277-9d94-9f4a11d793fc


So they are totally random.


Now Kafka producer requires a key, value pair, We generate UUID key as the
unique identifier of Kafka record


 uuidUdf= F.udf(lambda : str(uuid.uuid4()),StringType())
 result = df.withColumn("uuid",uuidUdf()) \


So back to your question. What is the use case for identity() in your SSS
application? If you want a true value as close as accurate (even MSSQL can
have gaps in the identity column because of crash etc), you need to store
the last value in a persistent storage like Hive table etc and start from


val start = spark.sql("SELECT MAX(id) FROM
test.randomData").collect.apply(0).getInt(0) + 1


HTH



   view my Linkedin profile
<https://www.linkedin.com/in/mich-talebzadeh-ph-d-5205b2/>



*Disclaimer:* Use it at your own risk. Any and all responsibility for any
loss, damage or destruction of data or any other property which may arise
from relying on this email's technical content is explicitly disclaimed.
The author will in no case be liable for any monetary damages arising from
such loss, damage or destruction.




On Thu, 15 Jul 2021 at 20:40, Felix Kizhakkel Jose <
felixkizhakkeljose@gmail.com> wrote:

> Thank you so much for the insights.
> @Mich Talebzadeh <mich.talebzadeh@gmail.com> Really appreciate your
> detailed examples.
> @Jungtaek Lim I see your point. I am thinking of having a mapping table
> with UUID to incremental ID and leverage range pruning etc on a large
> dataset.
> @sebastian I have to check how to do something like snowflake id. Do you
> have any examples or directions?
>
> Let me ask you another way, how are you handling the non incrementing
> UUIDs? Because Parquet - range stats has min and max, but if your id is a
> UUID, this doesn't help to decide whether the value that you search is
> present in the files until you scan the entire file, because min-max on
> uuid doesn't work greatly.
>
> Please share your experiences or ideas on how you handled this situation.
>
> Regards,
> Felix K Jose
>
> On Tue, Jul 13, 2021 at 7:59 PM Jungtaek Lim <kabhwan.opensource@gmail.com>
> wrote:
>
>> Theoretically, the composed value of batchId +
>> monotonically_increasing_id() would achieve the goal. The major downside is
>> that you'll need to deal with "deduplication" of output based on batchID
>> as monotonically_increasing_id() is indeterministic. You need to ensure
>> there's NO overlap on output against multiple reattempts for the same batch
>> ID.
>>
>> Btw, even just assume you dealt with auto increasing ID on write, how do
>> you read files and apply range pruning by auto increasing ID? Is the
>> approach scalable and efficient? You probably couldn't avoid reading
>> unnecessary files unless you build an explicit metadata regarding files
>> like the map file name to the range of ID and also craft a custom reader to
>> leverage the information.
>>
>>
>> On Wed, Jul 14, 2021 at 6:00 AM Sebastian Piu <sebastian.piu@gmail.com>
>> wrote:
>>
>>> If you want them to survive across jobs you can use snowflake IDs or
>>> similar ideas depending on your use case
>>>
>>> On Tue, 13 Jul 2021, 9:33 pm Mich Talebzadeh, <mich.talebzadeh@gmail.com>
>>> wrote:
>>>
>>>> Meaning as a monolithically incrementing ID as in Oracle sequence for
>>>> each record read from Kafka. adding that to your dataframe?
>>>>
>>>> If you do Structured Structured Streaming in microbatch mode, you will
>>>> get what is known as BatchId
>>>>
>>>>            result = streamingDataFrame.select( \
>>>>                      col("parsed_value.rowkey").alias("rowkey") \
>>>>                    , col("parsed_value.ticker").alias("ticker") \
>>>>                    , col("parsed_value.timeissued").alias("timeissued")
>>>> \
>>>>                    , col("parsed_value.price").alias("price")). \
>>>>                      writeStream. \
>>>>                      outputMode('append'). \
>>>>                      option("truncate", "false"). \
>>>>                      *foreachBatch(sendToSink). \*
>>>>                      trigger(processingTime='30 seconds'). \
>>>>                      option('checkpointLocation', checkpoint_path). \
>>>>                      queryName(config['MDVariables']['topic']). \
>>>>
>>>> That function sendToSink will introduce two variables df and batchId
>>>>
>>>> def *sendToSink(df, batchId):*
>>>>     if(len(df.take(1))) > 0:
>>>>         print(f"""md batchId is {batchId}""")
>>>>         df.show(100,False)
>>>>         df. persist()
>>>>         # write to BigQuery batch table
>>>>         s.writeTableToBQ(df, "append",
>>>> config['MDVariables']['targetDataset'],config['MDVariables']['targetTable'])
>>>>         df.unpersist()
>>>>         print(f"""wrote to DB""")
>>>>     else:
>>>>         print("DataFrame md is empty")
>>>>
>>>> That value batchId can be used for each Batch.
>>>>
>>>>
>>>> Otherwise you can do this
>>>>
>>>>
>>>> startval = 1
>>>> df = df.withColumn('id', monotonicallyIncreasingId + startval)
>>>>
>>>> HTH
>>>>
>>>>
>>>>    view my Linkedin profile
>>>> <https://www.linkedin.com/in/mich-talebzadeh-ph-d-5205b2/>
>>>>
>>>>
>>>>
>>>> *Disclaimer:* Use it at your own risk. Any and all responsibility for
>>>> any loss, damage or destruction of data or any other property which may
>>>> arise from relying on this email's technical content is explicitly
>>>> disclaimed. The author will in no case be liable for any monetary damages
>>>> arising from such loss, damage or destruction.
>>>>
>>>>
>>>>
>>>>
>>>> On Tue, 13 Jul 2021 at 19:53, Felix Kizhakkel Jose <
>>>> felixkizhakkeljose@gmail.com> wrote:
>>>>
>>>>> Hello,
>>>>>
>>>>> I am using Spark Structured Streaming to sink data from Kafka to AWS
>>>>> S3. I am wondering if its possible for me to introduce a uniquely
>>>>> incrementing identifier for each record as we do in RDBMS (incrementing
>>>>> long id)?
>>>>> This would greatly benefit to range prune while reading based on this
>>>>> ID.
>>>>>
>>>>> Any thoughts? I have looked at monotonically_incrementing_id but seems
>>>>> like its not deterministic and it wont ensure new records gets next id
from
>>>>> the latest id what  is already present in the storage (S3)
>>>>>
>>>>> Regards,
>>>>> Felix K Jose
>>>>>
>>>>

Mime
View raw message