I am using a Python script to compute users and items latent factors using Spark ML's ALS routine as described here.
After computing latent factors, I am trying to save those to S3 using the following:
model = als.fit(ratings)
# save items latent factors
model.itemFactors.rdd.saveAsTextFile(s3path_items)
# save users latent factors
model.userFactors.rdd.saveAsTextFile(s3path_users)
There are around 150 million users. LFA is computed quickly (~15 min) but writing out the latent factors to S3 takes almost 5 hours. So clearly, something is not right. Could you please help identify the problem?
I am using 100 users blocks and 100 items blocks in computing LFA using ALS - in case this info might be relevant.
Using 100 r3.8xlarge machines for the job.
Is this EMR, the official ASF Spark version, or something else?
One issue here is that the S3 clients have tended to buffer everything locally onto disk, then only start the upload afterwards.
If it's ASF code, you could make sure you are using Hadoop 2.7.x, use s3a:// as the output schema, and play with the fast output stream options, which can do incremental writes as things get generated. It's a bit brittle in 2.7, will be way better in 2.7.
If you are on EMR, you are on your own there.
Another possible cause is that S3 throttles clients generating lots of HTTPS requests to a particular shard of S3, which means: specific bits of an S3 bucket, with the first 5-8 characters apparently determining the shard. If you can use very unique names there, then you may get throttled less.
Related
We are using Spark history 3.2.1 to monitor our Spark applications.
We have thousands of daily jobs (running on Kubernetes) that writes event logs to S3 bucket (in a dedicated folder).
We are using history-server to analyze and compare completed jobs (incomplete running jobs never appeared in the UI but it's not a requirement now).
Recently I've noticed increase in our ListBucket API Operation in AWS billing cost explorer. This cost is higher than the cost of the StandardStorage (the price we pay for storing the data itself). It's up to few hundreds per month!
Running history-server with DEBUG log level exposed the "problem": every 10s the the history-server list the bucket to get all logs and then it iterate over each folder to get it's content. So if I want to keep the last 10,000 jobs, I'll have to pay for 10,101 ListBucket requests every 10s!
Here is one example (out of the 10k) reproduced locally with minio as S3:
22/02/20 06:44:31 DEBUG wire: http-outgoing-57 << "<ListBucketResult xmlns="http://s3.amazonaws.com/doc/2006-03-01/"><Name>local-audience</Name><Prefix>history-logs/eventlog_v2_spark-ffffdf5903c841259f28b53981746b76/</Prefix><KeyCount>2</KeyCount><MaxKeys>5000</MaxKeys><Delimiter>/</Delimiter><IsTruncated>false</IsTruncated><Contents><Key>history-logs/eventlog_v2_spark-ffffdf5903c841259f28b53981746b76/appstatus_spark-ffffdf5903c841259f28b53981746b76</Key><LastModified>2022-02-12T17:00:15.304Z</LastModified><ETag>"d41d8cd98f00b204e9800998ecf8427e"</ETag><Size>0</Size><Owner><ID></ID><DisplayName></DisplayName></Owner><StorageClass>STANDARD</StorageClass></Contents><Contents><Key>history-logs/eventlog_v2_spark-ffffdf5903c841259f28b53981746b76/events_1_spark-ffffdf5903c841259f28b53981746b76</Key><LastModified>2022-02-12T17:00:15.136Z</LastModified><ETag>"f91cc774d92c6f6c2ca4d0e1a1e76e13"</ETag><Size>868837</Size><Owner><ID></ID><DisplayName></DisplayName></Owner><StorageClass>STANDARD</StorageClass></Contents></ListBucketResult>"
To ensure that the cost comes from history-server I turned it off for a day and there was no charge per ListBucket since then:
To mitigate the problem (because we still need the history-server), I can set the spark.history.fs.update.interval to higher number (such as 3600s or so). As we are checking the history-server once a day it is overkill and doesn't worth it (cost wise).
Why does it scan the completed jobs every time (over and over again) and not only new jobs? is there a way to configure such behavior to avoid those ListBucket operations?
If I care only for completed jobs, and assuming I can wait few minutes to see the list, is there a mode that can load the list only when I login to the UI? (rather than periodically doing it for nothing).
P.S - I'm using AWS lifecycle rules to clean this folder every few few days (and not the server cleaning feature), by expiration objects after few days.
treewalking in s3 is (a) expensive and (b) horribly slow, especially given that a deep tree scan exists. If you want to fix this and can write scala code, see if you can fix the server to switch to a deep listing by moving to FileSystem.listFiles(path, true). Yes that involves coding, but the OSS community depends on everyone fixing their own personal issues and sharing the outcome
After digging into this issue, I decided to stop using the "rolling" feature for now - as my application jobs are relatively small.
I removed the:
spark.eventLog.rolling.enabled: true
spark.eventLog.rolling.maxFileSize: 16m
from the spark-submit command and the cost is now back to normal...
I also wrote about it here.
#stevel thanks for your answer - I will try to contribute and fix that! :)
At a high / theoretical level I know exactly the type of architecture I want to build and how it would work, but I'm attempting to construct this as cheaply as possible using AWS services and my lack of familiarity with the offerings of AWS has me running in circles.
The Data
We run a video streaming platform. On busy nights we have about 100 simultaneous live streams going with upwards of 30,000 viewers. We expect this number to rise to 100,000 in the next few years. A live stream lasts, on average, 2 hours.
We send a heartbeat from our player every 10 seconds with information about the viewer -- how much data they've viewed, how much data they've buffered, what quality they're streaming, etc.
These heartbeats are sent directly to an AWS Kinesis endpoint.
Finally, we want to retain all past messages for at least 5 years (hopefully longer) so that we can look at historic analytics.
Some back of the envelope calculations suggest we will have 0.1 * 60 * 60 * 2 * 100000 * 365 * 5 = 131 billion heartbeat messages five years from now.
Our Old Pipeline
Our old system had a single Kinesis consumer. Aggregate data was stored in DynamoDB. Whenever a message arrived we would read the record from DynamoDB, update the record, then write the new record back. This read-update-write loop limited the speed at which we could process messages and made it so that each message coming in was dependent on the messages before it, so they could not be processed in parallel.
Part of the reason for this setup is that our message schema was not well designed from the outset. We send the timestamp at which the message was sent, but we do not send "amount of video watched since last heartbeat". As a result in order to compute the total viewer time we need to look up the last heartbeat message sent by this player, subtract the timestamps, and add that value. Similar issues exist with many other metrics.
Our New Pipeline
We've begun to run into scaling issues. During our peak hours analytics can be delayed by as much as four hours while waiting for a backlog of messages to be processed. If this backlog reaches 24 hours Kinesis will start deleting data. So we need to fix our pipeline to remove this dependency on past messages so we can process them in parallel.
The first part of this was updating the messages sent by our players. Our new specification includes only metrics that can be trivially sum'd with no subtraction. So we can just keep adding to the "time viewed" metric, for instance, without any regard to past messages.
The second part of this was ensuring that Kinesis never backs up. We dump the raw messages to S3 as quickly as they arrive with no processing (Kinesis Data Fire Hose) so that we can crunch analytics on them at our leisure.
Finally, we now want to actually extract information from these analytics as quickly as possible. This is where I've hit a snag.
The Questions We Want to Answer
As this is an analytics pipeline, our questions mostly revolve around filtering these messages and then aggregating fields for the remaining messages (possibly, in fact likely, with grouping). For instance:
How many Android users watched last night's stream in HD? (FILTER by stream and OS)
What's the average bandwidth usage among all users? (SUM and COUNT, with later division of the final aggregates which could be done on the dashboard side)
What percent of users last year were on any Apple device (iOS, tvOS, etc)? (COUNT, grouped by OS)
What's the average time spent buffering among Android users for streams in the past year? (a mix of all of the above)
Options
AWS Athena would allow us to query the data in S3 directly as if it were an ANSI SQL table. However reading up on Athena, unless the data is properly formatted it can be incredibly slow. Some benchmarks I've seen show that processing 1.1 billion rows of CSV data can take up to 2 minutes. I'm looking at processing 100x that much data
AWS EMR and AWS Redshift sound like they are built for this purpose, but are complicated to set up and have a high base cost to run (requiring an EC2 cluster to remain active at all times). AWS Redshift also requires data be loaded into it, which sounds like it might be a very slow process, delaying our access to analytics
AWS Glue sounds like it may be able to take the raw messages as they arrive in S3 and convert them to Parquet files for more rapid querying via Athena
We could run a job to regularly batch messages to reduce the total number that must be processed. While a stream is live we'll receive one message every 10 seconds, but we really only care about the totals for a given viewer. This means that when a 2-hour stream concludes we can combine the 720 messages we've received from that player into a single "summary" message about the viewer's experience during the whole stream. This would massively reduce the amount of data we need to process, but exactly how and when to trigger this process isn't clear to me
The Ideal Architecture
This is a Big Data problem. The generic solution to Big Data problems is "don't take your data to your query, take your query to your data". If these messages were spread across 100 small storage nodes then each node could filter, sum, and count the subset of data they hold and pass these aggregates back to a central node which sums the sums and sums the counts. If each node is only operating on 1/100th of the data set then this kind of processing could theoretically be incredibly fast.
My Confusion
While I have a theoretical understanding of the "ideal" architecture, it's not clear to me if AWS works this way or how to construct a system that will function well like this.
S3 is a black box. It's not clear if Athena queries are run on individual nodes and aggregates are further reduced elsewhere, or if there's a system reading all of the data and aggregating it in a central location
Redshift requires the data by copied into a Redshift database. This doesn't sound fast, nor distributed
It's unclear to me how EMR works or if it will suit my purpose. Still researching
AWS Glue seems like it may need to be triggered by some event?
Parquet files seems to be like CSVs, where multiple records reside in a single file. Meanwhile I'm dumping one record per file. But perhaps there's a way to fix that? e.g. batching files every minute or every 5 minutes?
RDS or a similar service might be really good for this (indexing and whatnot) but would require a guaranteed schema (or necessitate migrating if our message schema changed) which is a concern. Migrating terabytes of data if we change our message schema sounds out of the question
Finally, along with wanting to get analytics results in as "real time" as possible (ideally we want to know within 1 minute when someone joins or leaves a stream), we want the dashboards to load quickly. Waiting 30 seconds to see the count of live viewers is horrendous. Dashboards should load in 2 seconds or less (ideally)
The plan is to use QuickSight to create dashboards (our old system had a hack-y Django app that read from our DynamoDB aggregates table, but I'd like to avoid creating more code for people to maintain)
I expect you are going to get a lot of different answers and opinions from the broad set of experts you have pinged with this. There is likely no single best answer to this as there are a lot of variables. Let me give you my best advice based on my experience in the field.
Kinesis to S3 is a good start and not moving data more than needed is the right philosophy.
You didn't mention Kinesis Data Analytics and this could be a solution for SOME of your needs. It is best for questions about what is happening in the data feed right now. The longer timeframe questions are better suited for the tools you mention. If you aren't too interested in what is happening in the past 10 minutes (or so) it could be good to omit.
S3 organization will be key to performing any analytic directly on the data there. You mention parquet formatting which is good but partitioning is far more powerful. Organizing the S3 data into "days" or "hours" of data and setting up the partitioning based on this can greatly speed up any query that is limited in the amount of time that is needed (don't read what you don't need).
Important safety note on S3 - S3 is an object store and as such there is overhead for each object you reference. Having many small objects (10,000+) treated as a single set of data is going to be slow no matter what solution you go with. You need to fix this before you go forward with any solution. You see it takes upwards of .5 sec to look up an object in S3 but if the file is small the transfer time is next to nothing. Now multiply .5 sec times all the objects you have and see how long it will take to read them. This is not a function of the downstream tool you choose but of the S3 organization you have. S3 objects as part of a Big Data solution should be at least 100M in size to not suffer greatly from the object lookup time. The choice of parquet or CSV files is mute without addressing object size and partitioning first.
Athena is good for occasional queries especially if the date ranges are limited. Is this the query pattern you expect? As you say "move the compute to the data" but if you use Athena to do large cross-sectional analytics where a large percentage of the data needs to be used, you are just moving the data to Athena every time you execute this query. Don't stop thinking about data movement at the point it is stored - think about the data movements to do the analytics also.
So a big question is how much data is needed and how often to support your analytics workloads and BI functions? This is the end result you are looking for. If a high percentage of the data is needed frequently then a warehouse solution like Redshift with the data loaded to disk is the right answer. The data load time to Redshift is quite fast as it parallel loads the data from S3 (you see S3 is a cluster and Redshift is a cluster and parallel loads can be done). If loading all your data into Redshift is what you need then the load time is not your main concern - the cost is. Big powerful tool with a price tag to match. The new RA3 instance type bends this curve down significantly for large data size clusters so could be a possibility.
Another tool you haven't mentioned is Redshift Spectrum. This brings several powerful technologies together that could be important to you. First is the power of Redshift with the ability to choose smaller cluster sizes that normally would be used for your data size. S3 filtering and aggregation technology allows Spectrum to perform actions on the data in S3 (yes initial compute actions of the query are performed inside of S3 potentially greatly reducing the data moved to Redshift). If your query patterns support this data reduction in S3 then the data movement will be small and the Redshift cluster can be small (cheap) too. This can be a powerful compromise point for IoT solutions like yours since complex data models and joining are not needed.
You bring up Glue and conversion to parquet. These can be good to do but as I mentioned before partitioning of the data in S3 is usually far more powerful. The value of parquet will increase as the width of your data increases. Parquet is a columnar format so it is advantaged if only a subset of "columns" are needed. The downside is the conversion time/cost and the loss of easy human readability (which can be huge during debug).
EMR is another choice you mention but I generally advise clients against going with EMR unless they need the flexibility it brings to the analytics and they have the skills to use it well. Without these EMR tends to be an unneeded costs sink.
If this is really going to be a Big Data solution then RDS (and Aurora) not good choices. They are designed for transactional workloads, not analytics. The data size and analytics will not fit well or be cost effective.
Another tool in the space is S3 Select. Not likely what you are looking for but something to remember exists and can be a tool in the toolbox.
Hybrid solutions are common in this space if there are variable needs based on some factor. A common one "is time of day" - no one is running extensive reports at 3am so the needed performance is much less. Another is user group - some groups need simple analytics while others need much more power. Another factor is timeliness of data - does everyone need "up to the second" information or is daily information sufficient? Trying to have one tool that does everything for everybody, all the time is often a path to an expensive, oversized solution.
Since Redshift Spectrum and Athena can point at the same S3 data (well organized since both will benefit) both tools can coexist on the same data. Also, Redshift is ideal for sifting through huge mounds of data, it is ideal for producing summary tables and then writing them (in partitioned parquet) to S3 for tools like Athena to use. All these cloud services can be run on schedules and this includes Redshift and EMR (Athena is query on demand) so they don't need to run all the time. Redshift with Spectrum can run a few hours a day to perform deep analytics and summarize data for writing to S3. Your data scientist can also use Redshift for their hardcore work while Athena supports dashboards using the daily summary data and Kinesis Data Analytics as source.
Lastly you bring up a 2 sec requirement for dashboards. This is definitely possible with Quicksight backed up by Redshift or Athena but won't be met for arbitrarily complex / data intensive queries. To meet this you will need the engine to have enough horsepower to produce the data in question. Redshift with local data storage is likely the fastest (Redshift Spectrum with some data pruning done in S3 wins in some cases) and Athena is the weakest / slowest. But the power doesn't matter if the work is small - see your query workload will be a huge deciding factor. The fastest will be to load the needed data into Quicksight storage (SPICE) but this is another localized / summarized version of the data so timeliness is again a factor (how often is this updated).
Based on designing similar systems and a bunch of guesses as to what you need I'd recommend that you:
Fix your object size (Kineses can be configured to do this)
Partition your data by day
Set up a small Redshift cluster (4 X dc2.large) and use Spectrum source address the data
Connect Quicksight to Redshift
Measure the performance (and cost) and compare to requirements (there will likely be gaps)
Adjust to solution (summary tables to S3, Athena, SPICE etc.) to meet goals
The alternative is to hire someone who has set up such systems before and have them review the requirements in detail and make a less "guess-based" recommendation.
I would look into Druid. Not an AWS offering, but easily runs on AWS, with good integration with S3 and Kinesis.
Capable of reading from Kinesis, at high speeds, and make the data available for querying right away. Can also flatten and transform the data as it reads it.
Capable of doing rollups/aggregation/compaction during ingestion (and further reduce data in an async manner). From what you wrote, it seems to me that it could easily reduce the number of rows in the DB by a very large factor.
Capable of fast queries, using standard SQL.
Smart partitioning of the data to scan only the relevant dates.
The down-side is that you will need to keep a cluster up and running for ingestion and for querying. It is pretty scalable, so you can start small.
On the up-side - you're not using 10 different technologies (Athena/Glue/EMR/etc.)
You might want to consider contacting Imply, which can ease the deployment.
A usual approach a lot of companies take is they do heavy weight lifting in athena or bigquery (or some other distributed sql environment) -> aggregate intermediate results into multiple indexed+partitioned postgres/mysql/redshift/clickhouse tables and then connect their APIs to read on those tables. Of course, this works fine except the fact that with an increased amount of intermediate-aggregated data, table indices grow and problems like cumulative sum or sorting become less and less efficient.
With your problem in hand, I think you can get a lot of help with AWS Lambda. AWS Lambda provides a very feasible serverless approach towards solving large granular data problems (if used correctly). For instance, assume that your pipelines partitions incoming stream by YYYYMMMDDHHMM and stores it into some S3 path which has a Lambda listening to it (as a trigger function) then your data ingest + aggregation becomes pretty much simultaneous processes. As soon as a minute is up, a new instance of the same Lambda function will be taking care of data landing into partition YYYYMMMDDHHMM+1. So, this way, you can run thousands of simultaneous processes with a good bunch of Lambda functions doing the same thing in parallel. Of course, this is a rough picture, but I think it can greatly help.
I am looking to get all of the Activity and Lead data in Marketo to be mirrored in an AWS S3 bucket so that I can build dashboards on it in Quicksight, so preferably I'd like to stream the data from Marketo into S3 in real-time, and then use Glue and Athena to connect the data to Quicksight. However, the only way to get large volumes of data out of Marketo appears to be their Bulk Extract tool (one for Leads, one for Activity data).
The problem is that these API interfaces make any attempt at near real-time streaming really clunky. Currently, I have Lambda functions being triggered every hour to pull the most recent hour of Lead/Activity data and saving it as a gzipped CSV in S3. But Marketo's Bulk Extract tool has a request queue and requests often take longer than 15 minutes to process (15 minutes being Lambda's max timeout length). So at least once a day my requests are getting dropped.
The solution seems to be to instead run this on an EC2 instance that can juggle multiple requests and patiently wait for Marketo's queue. But I'd rather not get into all the async and error-handling issues that that approach may entail if there is an easier way to accomplish this.
As an alternative solution, Amazon Appflow integrates with Marketo. But last I checked, it only works with Lead data, not Activity data. And there are restrictions on the filters you have to apply to the Lead data that make it clunky to work with anyway.
On Google I have found several companies that claim to offer seamless, reliable Marketo-to-S3 ETL, but I haven't yet researched their pricing or quality.
If anyone knows of a good approach to set up reliable and cost-efficient ETL between Marketo and S3 in a short period of time, I would very much appreciate it.
In a case like this, I would be tempted to recommend using an EC2 instance to run Singer with a Marketo input and CSV output, then set up something to move the CSV over to S3 as needed. That would be the absolute cheapest ETL solution, but this does suppose you have some comfort and familiarity with Python.
Also worth noting is that Stitch, Singers's paid product equivalent, supports native S3 export--you could always first test with a non-Marketo data source and see if that performs the way you'd like if you prefer money over time.
Having a very specific access pattern for my data, I wonder about the expected mapreduce performance of Cassandra. These are my requirements:
There will be 10 Million Documents (e.g. JSON, a couple of KB each)
in my database There will be occasional updates of the documents
Users want to create results from the whole dataset that require
processing of each document
Users will want to do this in a
semi-interactive fashion, trying out effects of changes they make to
the processing of each document. Waiting for the result a couple of
minutes is ok.
Users would like to be able to spend money (scaling up
or out) to increase interactive speed if there is a desire to
increase processing speed.
There will not be large user numbers,
processing needs to be done a couple of times per hour, maybe.
Durability is not a primary concern, as the data is replicated from a
source system anyway.
This sounds like a good Job for Cassandra and MapReduce but given that MapReduce is not intended to be used semi-interactively but rather as a background job, I wonder what performance possibilities I can expect using Cassandra.
My other options are plain MySQL with documents stored as CLOBS or partitioned Redis.
Can anyone provide clues on how to estimate the speed possibilities?
I'm searching for information on how ElasticSearch would scale with the amount of data in its indexes and am surprised how little I can find on that topic. Maybe some experience from the crowd here can help me.
We are currently using CloudSearch to index ≈ 7 million documents; in CloudSearch this results in 2 instances of type m2.xlarge. We are considering switching to ElasticSearch instead to reduce the cost. But all I find on the scaling of ElasticSearch is that it does scale well, can be distributed over several instances etc.
But what kind of machine (memory, disc) would I need for this kind of data?
How would that change if I increased the amount of data by the factor of 12 (≈ 80 million documents)?
As Javanna said, it depends. Mostly on: (1) rate of indexing; (2) size of documents; (3) rate and latency requirements for searches; and (4) type of searches.
Considering this, the best we can help is giving examples.
On our site (news monitoring) we:
Index more than 100 docs per minute. We have, currently, near 50 million documents. I've also heard of ES indexes with hundreds of millions of documents.
Documents are news articles with some metadata, not short but not that large.
Our search latency varies between ~50ms (for normal and rare terms) up to 800ms for common terms (stopwords, we index them). This variation is largely due to our custom scoring (thanks to Lucene/ES support for customizing it) and to the fact the dataset (inverted lists) do not fit entirely in memory (OS cache). So when it hits a cached inverted list, it's faster.
We do OR queries with a lot of terms which are one of the hardest. Also we do faceting on two single-valued fields. And have some experiments with date facet (to show rate of publication through time).
We do all this with 4 EC2's m1.large instances. And now we're planning moving to ES, just released, 0.9 version to get all the goodies and performance improvements of Lucene 4.0.
Now leaving examples aside. ElasticSearch is pretty scalable. It is very simple to create an index with N shards and M replicas, and then create X machines with ES. It will distribute all shards and replicas accordingly. You can change the number of replicas anytime you want (for each index).
One downside is that you can't change the number of shards after the index creation. But you can still "overshard" it beforehand to leave room for scaling when needed. Or create a new index with the right number of shards and reindex everything (we do this).
Finally, ElasticSearch (and also Solr) uses, under the hood, the Lucene Search library, which is very mature and well known library.
I've actually recently switched from using CloudSearch to a hosted ElasticSearch service at the company I work for. Our specific application has a little over 100 million documents and is growing daily. So far, our experience with ElasticSearch has been absolutely wonderful. Search performance averages at ~250ms, even with all the sorting, filtering, and faceting. Indexing documents is also relatively fast, despite the several MB load we pass through HTTP with the bulk API every couple of hours. Refresh rates seem to be near instant, as well.
For our ~100M doc / 12GB index, we used 4 shards / 2 replicas (will bump to 3 replicas if performance degrades) spread across 4 nodes. Prior to setting up the index, our team spent a couple of days researching ElasticSearch cluster deployment/maintenance, and opted to use http://qbox.io to save money and time. We were paralyzingly afraid of performance and scale issues choosing to host our index on a dedicated cluster like Qbox, but so far the experience has been seriously fantastic.
Since our index lives on a dedicated cluster, we don't have access to nuts-and-bolts node-level configuration settings, so my technical expertise with ES deployment is still pretty limited. That being said, I can't be sure of exactly what performance tweeks are needed for the performance we've experienced on our index. However, I do know Qbox's cluster uses SSD... so that could definitely have a significant impact.
Point in case, ElasticSearch has scaled seamlessly. I highly, highly recommend the switch (even if it's just to save $$, CloudSearch is crazy expensive). Hope this information helps!
CloudSearch recently dropped prices and may now be a cheaper alternative than maintaining your own Search infrastrcuture on EC2 - http://aws.amazon.com/blogs/aws/cloudsearch-price-reduction-plus-features/