RE: [spark-streaming] kafka source and flow control
I was hoping I could make the system behave as a blocking queue : if the outputs is too slow, buffers (storing space for RDDs) fills up, then blocks instead of dropping existing rdds, until the input itself blocks (slows down it’s consumption). On a side note I was wondering: is there the same issue with file (hdfs) inputs ? how can I be sure the input won’t “overflow” the process chain ? From: Tobias Pfeiffer [mailto:t...@preferred.jp] Sent: mardi 12 août 2014 02:58 To: Gwenhael Pasquiers Cc: u...@spark.incubator.apache.org Subject: Re: [spark-streaming] kafka source and flow control Hi, On Mon, Aug 11, 2014 at 9:41 PM, Gwenhael Pasquiers gwenhael.pasqui...@ericsson.commailto:gwenhael.pasqui...@ericsson.com wrote: We intend to apply other operations on the data later in the same spark context, but our first step is to archive it. Our goal is somth like this Step 1 : consume kafka Step 2 : archive to hdfs AND send to step 3 Step 3 : transform data Step 4 : save transformed data to HDFS as input for M/R I see. Well I think Spark Streaming may be well suited for that purpose. To us it looks like a great flaw if, in streaming mode, spark-streaming cannot slow down it’s consumption depending on the available resources. On Mon, Aug 11, 2014 at 10:10 PM, Gwenhael Pasquiers gwenhael.pasqui...@ericsson.commailto:gwenhael.pasqui...@ericsson.com wrote: I think the kind of self-regulating system you describe would be too difficult to implement and probably unreliable (even more with the fact that we have multiple slaves). Isn't slow down its consumption depending on the available resources a self-regulating system? I don't see how you can adapt to available resources without measuring your execution time and then change how much you consume. Did you have any particular form of adaption in mind? Tobias
RE: [spark-streaming] kafka source and flow control
Hi, We intend to apply other operations on the data later in the same spark context, but our first step is to archive it. Our goal is somth like this Step 1 : consume kafka Step 2 : archive to hdfs AND send to step 3 Step 3 : transform data Step 4 : save transformed data to HDFS as input for M/R Yes, maybe spark-streaming isn’t the tool adapted to our needs, but it looked like it so I wonder if I didn’t miss something. To us it looks like a great flaw if, in streaming mode, spark-streaming cannot slow down it’s consumption depending on the available resources. From: Tobias Pfeiffer [mailto:t...@preferred.jp] Sent: lundi 11 août 2014 11:44 To: Gwenhael Pasquiers Subject: Re: [spark-streaming] kafka source and flow control Hi, On Mon, Aug 11, 2014 at 6:19 PM, gpasquiers gwenhael.pasqui...@ericsson.commailto:gwenhael.pasqui...@ericsson.com wrote: I’m using spark-streaming in a cloudera environment to consume a kafka source and store all data into hdfs. I assume you are doing something else in between? If not, maybe a software such as Apache Flume may be better suited? I have complete control over the kafka consumer since I developed a custom Receiver as a workaround to : https://issues.apache.org/jira/browse/SPARK-2103 but i’d like to make a flow control more intelligent than a simple rate limite (x messages or bytes per second). I’m interested in all ideas or suggestions. I think what I would try to do is measuring how much data I can process within one time window (i.e., keep track of processing speed) and then (continuously?) adapt the data rate to something that I am capable of processing. In that case you would have to make sure that the data doesn't instead get lost within Kafka. After all, the problem seems to be that your HDFS is too slow and you'll have to buffer that data *somewhere*, right? Tobias
RE: [spark-streaming] kafka source and flow control
I didn’t reply to the last part of your message: My source is Kafka, kafka already acts as a buffer with a lot of space. So when I start my spark job, there is a lot of data to catch up (and it is critical not to lose any), but the kafka consumer goes as fast as it can (and it’s faster than my hdfs’s maximum). I think the kind of self-regulating system you describe would be too difficult to implement and probably unreliable (even more with the fact that we have multiple slaves). From: Tobias Pfeiffer [mailto:t...@preferred.jp] Sent: lundi 11 août 2014 11:44 To: Gwenhael Pasquiers Subject: Re: [spark-streaming] kafka source and flow control Hi, On Mon, Aug 11, 2014 at 6:19 PM, gpasquiers gwenhael.pasqui...@ericsson.commailto:gwenhael.pasqui...@ericsson.com wrote: I’m using spark-streaming in a cloudera environment to consume a kafka source and store all data into hdfs. I assume you are doing something else in between? If not, maybe a software such as Apache Flume may be better suited? I have complete control over the kafka consumer since I developed a custom Receiver as a workaround to : https://issues.apache.org/jira/browse/SPARK-2103 but i’d like to make a flow control more intelligent than a simple rate limite (x messages or bytes per second). I’m interested in all ideas or suggestions. I think what I would try to do is measuring how much data I can process within one time window (i.e., keep track of processing speed) and then (continuously?) adapt the data rate to something that I am capable of processing. In that case you would have to make sure that the data doesn't instead get lost within Kafka. After all, the problem seems to be that your HDFS is too slow and you'll have to buffer that data *somewhere*, right? Tobias
Re: [spark-streaming] kafka source and flow control
Hi, On Mon, Aug 11, 2014 at 9:41 PM, Gwenhael Pasquiers gwenhael.pasqui...@ericsson.com wrote: We intend to apply other operations on the data later in the same spark context, but our first step is to archive it. Our goal is somth like this Step 1 : consume kafka Step 2 : archive to hdfs AND send to step 3 Step 3 : transform data Step 4 : save transformed data to HDFS as input for M/R I see. Well I think Spark Streaming may be well suited for that purpose. To us it looks like a great flaw if, in streaming mode, spark-streaming cannot slow down it’s consumption depending on the available resources. On Mon, Aug 11, 2014 at 10:10 PM, Gwenhael Pasquiers gwenhael.pasqui...@ericsson.com wrote: I think the kind of self-regulating system you describe would be too difficult to implement and probably unreliable (even more with the fact that we have multiple slaves). Isn't slow down its consumption depending on the available resources a self-regulating system? I don't see how you can adapt to available resources without measuring your execution time and then change how much you consume. Did you have any particular form of adaption in mind? Tobias
Re: [spark-streaming] kafka source and flow control
In general, (and I am prototyping), I have a better idea :) - Consume kafka in Spark from topic-A - transform data in Spark (normalize, enrich etc etc) - Feed it back to Kafka (in a different topic-B) - Have flume-HDFS (for M/R, Impala, Spark batch) or Spark-streaming or any other compute framework subscribe to B On Mon, Aug 11, 2014 at 5:57 PM, Tobias Pfeiffer t...@preferred.jp wrote: Hi, On Mon, Aug 11, 2014 at 9:41 PM, Gwenhael Pasquiers gwenhael.pasqui...@ericsson.com wrote: We intend to apply other operations on the data later in the same spark context, but our first step is to archive it. Our goal is somth like this Step 1 : consume kafka Step 2 : archive to hdfs AND send to step 3 Step 3 : transform data Step 4 : save transformed data to HDFS as input for M/R I see. Well I think Spark Streaming may be well suited for that purpose. To us it looks like a great flaw if, in streaming mode, spark-streaming cannot slow down it’s consumption depending on the available resources. On Mon, Aug 11, 2014 at 10:10 PM, Gwenhael Pasquiers gwenhael.pasqui...@ericsson.com wrote: I think the kind of self-regulating system you describe would be too difficult to implement and probably unreliable (even more with the fact that we have multiple slaves). Isn't slow down its consumption depending on the available resources a self-regulating system? I don't see how you can adapt to available resources without measuring your execution time and then change how much you consume. Did you have any particular form of adaption in mind? Tobias - To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org