I have a single local directory on the order of 1 terabyte. It is made up of millions of very small text documents. If I were to iterate through each file sequentially for my ETL, it would take days. What would be the fastest way for me to perform ETL on this data, ultimately loading it onto a distributed store like hdfs or a redis cluster?
Generically: try to use several/many parallel asynchronous streams, one per file. How many will depend on several factors (number of destination endpoints, disk IO for traversing/reading data, network buffers, errors and latency...)
Related
When saving to a delta table we avoid 'df.coalesce(1)' but when saving to csv or parquet we(my team) add 'df.coalesce(1)'. Is it a common practise? Why? Is it mandatory?
In most cases when I have seen df.coalesce(1) it was done to generate only one file, for example, import CSV file into Excel, or for Parquet file into the Pandas-based program. But if you're doing .coalesce(1), then the write happens via single task, and it's becoming the performance bottleneck because you need to get data from other executors, and write it.
If you're consuming data from Spark or other distributed system, having multiple files will be beneficial for performance because you can write & read them in parallel. By default, Spark writes N files into the directory where N is the number of partitions. As #pltc noticed, this may generate the big number of files that's often not desirable because you'll get performance overhead from accessing them. So we need to have a balance between the number of files and their size - for Parquet and Delta (that is based on Parquet), having the bigger files bring several performance advantages - you read less files, you can get better compression for data inside the file, etc.
For Delta specifically, having .coalesce(1) having the same problem as for other file formats - you're writing via one task. Relying on default Spark behaviour and writing multiple files is beneficial from performance point of view - each node is writing its data in parallel, but you can get too many small files (so you may use .coalesce(N) to write bigger files). For Databricks Delta, as it was correctly pointed by #Kafels, there are some optimizations that will allow to remove that .coalesce(N) and do automatic tuning achieve the best throughput (so called "Optimized Writes"), and create bigger files ("Auto compaction") - but they should be used carefully.
Overall, the topic of optimal file size for Delta is an interesting topic - if you have big files (1Gb is used by default by OPTIMIZE command), you can get better read throughput, but if you're rewriting them with MERGE/UPDATE/DELETE, then big files are bad from performance standpoint, and it's better to have smaller (16-64-128Mb) files, so you can rewrite less data.
TL;DR: it's not mandatory, it depends on the size of your dataframe.
Long answer:
If your dataframe is 10Mb, and you have 1000 partitions for example, each file would be about 10Kb. And having so many small files would reduce Spark performance dramatically, not to mention when you have too many files, you'll eventually reach OS limitation of the number of files. Any how, when your dataset is small enough, you should merge them into a couple of files by coalesce.
However, if your dataframe is 100G, technically you still can use coalesce(1) and save to a single file, but later on you will have to deal with less parallelism when reading from it.
I've been reading few questions regarding this topic and also several forums, and in all of them they seem to be mentioning that each of resulting .parquet files coming out from Spark should be either 64MB or 1GB size, but still can't make my mind around which case scenarios belong to each of those file sizes and the reasons behind apart from HDFS splitting them in 64MB blocks.
My current testing scenario is the following.
dataset
.coalesce(n) # being 'n' 4 or 48 - reasons explained below.
.write
.mode(SaveMode.Append)
.partitionBy(CONSTANTS)
.option("basepath", outputPath)
.parquet(outputPath)
I'm currently handling a total of 2.5GB to 3GB of daily data, that will be split and saved into daily buckets per year. The reasons behind 'n' being 4 or 48 is just for testing purposes, as I know the size of my testing set in advance, I try to get a number as close to 64MB or 1GB as I can. I haven't implemented code to buffer the needed data until I get the exact size I need prior saving.
So my question here is...
Should I take the size that much into account if I'm not planning to use HDFS and merely store and retrieve data from S3?
And also, which should be the optimal size for daily datasets of around 10GB maximum if I'm planning to use HDFS to store my resulting .parquet files?
Any other optimization tip would be really appreciated!
You can control the split size of parquet files, provided you save them with a splittable compression like snappy. For the s3a connector, just set fs.s3a.block.size to a different number of bytes.
Smaller split size
More workers can work on a file simultaneously. Speedup if you have idle workers.
More startup overhead scheduling work, starting processing, committing tasks
Creates more files from the output, unless you repartition.
Small files vs large files
Small files:
you get that small split whether or not you want it.
even if you use unsplittable compression.
takes longer to list files. Listing directory trees on s3 is very slow
impossible to ask for larger block sizes than the file length
easier to save if your s3 client doesn't do incremental writes in blocks. (Hadoop 2.8+ does if you set spark.hadoop.fs.s3a.fast.upload true.
Personally, and this is opinion, and some benchmark driven -but not with your queries
Writing
save to larger files.
with snappy.
shallower+wider directory trees over deep and narrow
Reading
play with different block sizes; treat 32-64 MB as a minimum
Hadoop 3.1, use the zero-rename committers. Otherwise, switch to v2
if your FS connector supports this make sure random IO is turned on (hadoop-2.8 + spark.hadoop.fs.s3a.experimental.fadvise random
save to larger files via .repartion().
Keep an eye on how much data you are collecting, as it is very easy to run up large bills from storing lots of old data.
see also Improving Spark Performance with S3/ADLS/WASB
I have a NodeJS application that needs to stream data from an RDS Postgres, perform some relatively expensive CPU operations on the data, and insert it into another database. The CPU intensive portion I've offloaded into an AWS Lambda, such that the Node application will get a batch of rows and immediately pass them to the Lambda for processing. The bottleneck appears to be the speed in which the data can be received from Postgres.
In order to utilize multiple connections to the DB, I have an algorithm which is effectively leapfrogging on sorted IDs, so that many concurrent connections can be maintained. Ex: 1 connection fetches ids 1-100, second one fetches ids 101-200, etc, and then when the first returns maybe it fetches ids 1001-1100. Is this relatively standard practice? Is there a faster method for pulling the data out for processing?
So long as I am below the database's max_connections, would it be arguably beneficial to add more, possibly as additional concurrent applications streaming data out of it? Both the application and the RDS are currently in the VPC, and the CPU utilization on the RDS gets to about 30%, with memory at 60%.
It would likely be MUCH faster to dump your Postgres database into a CSV file or export it directly to flat files, dump the flat files into S3 after splitting them up, then have workers process each batch of files on their own.
Streaming data out of Postgres (particularly if you're doing it for millions of items) will take a LOT of IO and a very long time.
I am writing software in C, on Linux running on AWS, that has to handle 240 terabytes of data, in 72 million files.
The data will be spread across 24 or more nodes, so there will only be 10 terabytes on each node, and 3 million files per node.
Because I have to append data to each of these three million files every 60 seconds, the easiest and fastest thing to do would to be able to keep each of these files open at one time.
I can't store the data in a database, because the performance in reading/writing the data will be too slow. I need to be able to read the data back very quickly.
My questions:
1) is it even possible to keep open 3 million files
2) if it is possible, how much memory would it consume
3) if it is possible, would performance be terrible
4) if it is not possible, I will need to combine all of the individual files into a couple of dozen large files. Is there a maximum file size in Linux?
5) if it is not possible, what technique should I use to append data every 60 seconds, and keep track of it?
The following is a very coarse description of an architecture that can work for your problem, assuming that the maximum number of file descriptors is irrelevant when you have enough instances.
First, take a look at this:
https://aws.amazon.com/blogs/aws/amazon-elastic-file-system-shared-file-storage-for-amazon-ec2/
https://aws.amazon.com/efs/
EFS provides a shared storage that you can mount as a filesystem.
You can store ALL your files in a single storage unit of EFS. Then, you will need a set of N worker-machines running at full capacity of filehandlers. You can then use a Redis queue to distribute the updates. Each worker has to dequeue a set of updates from Redis, and then will open necessary files and perform the updates.
Again: the maximum number of open filehandlers will not be a problem, because if you hit a maximum, you only need to increase the number of worker machines until you achieve the performance you need.
This is scalable, though I'm not sure if this is the cheapest way to solve your problem.
Given:
One can access an HSQLDB database concurrently using connections pooled with the help of the apache commons dbcp package.
I store files in a cached table in an embedded hsqldb database.
It is known that files on a conventional hard drive (as opposed to a solid state) should not be accessed from multiple threads, because we are likely to get performance degradation rather than boost. This is because of the time it takes to move the mechanical reading head back and forth between the files with each thread context switch.
Question:
Does this rule hold to files managed in an HSQLDB database? The file sizes may range from several KB to several MB.
HSQLDB accesses two files for data storage during operations. One file for all CACHED table data, and another file for all the lobs. It manages access to these files internally.
With multiple threads, there is a possibility of reduced access speed in the following circumstances.
Simultaneous read and write access to large tables.
Simultaneous read and write access to lobs larger than 500KB.