英文:
Fastest way to find files modified in last 'x' minutes
问题
I have a requirement to find files modified in the last 10 minutes in a directory. The directory keeps getting updated and it will have around 50k-60k files every time. I'm using the below code to get the files:
import java.io.File
import java.time.Instant
val dir = new File("/path/to/dir")
val files = dir.listFiles.toList.filter(f => f.getName.matches("some filter"))
files.filter(f => f.isFile && f.exists &&
Instant.ofEpochMilli(f.lastModified).plus(10, MINUTES).isAfter(Instant.now))
.toList.sortBy(_.lastModified)
This takes around 20-30 minutes to run. But I want to get the results in less than 10 minutes.
I even tried running this in our hadoop cluster using spark. This is the spark code:
val sparkConfig = new SparkConf()
.setAppName("findRecentFiles")
.set("spark.serializer", "org.apache.spark.serializer.KryoSerializer")
.set("spark.shuffle.compress", "true")
.set("spark.rdd.compress", "true")
val sc = new SparkContext(sparkConfig)
val rdd = sc.parallelize(files)
rdd.filter(f => f.isFile && f.exists &&
Instant.ofEpochMilli(f.lastModified).plus(10, MINUTES).isAfter(Instant.now))
.collect.toList.sortBy(_.lastModified)
Still, it takes the same time. And one thing I noticed is that filtering based on the file name is fast. But adding the lastModified filter makes it slow. Is there any better way so that I can get the results faster?
UPDATE
I updated the spark configs and now I'm able to get the results in less than 10 minutes. Earlier, I was running the jar like this:
spark-submit myJar.jar
I changed it to this:
spark-submit --deploy-mode client --queue SomeNonDefaultQueue --executor-memory 16g --num-executors 10 --executor-cores 1 --master yarn myJar.jar
Also removed set("spark.rdd.compress", "true")
from code as it increases the CPU time, as explained here - https://spark.apache.org/docs/2.3.0/configuration.html#compression-and-serialization
英文:
I have a requirement to find files modified in last 10 minutes in a directory. The directory keeps getting updated and it will have around 50k-60k files every time. I'm using the below code to get the files:
import java.io.File
import java.time.Instant
val dir = new File("/path/to/dir")
val files = dir.listFiles.toList.filter(f => f.getName.matches("some filter"))
files.filter(f => f.isFile && f.exists &&
Instant.ofEpochMilli(f.lastModified).plus(10, MINUTES).isAfter(Instant.now))
.toList.sortBy(_.lastModified)
This takes around 20-30 minutes to run. But I want to get the results in less than 10 minutes.
I even tried running this in our hadoop cluster using spark. This is the spark code:
val sparkConfig = new SparkConf()
.setAppName("findRecentFiles")
.set("spark.serializer", "org.apache.spark.serializer.KryoSerializer")
.set("spark.shuffle.compress", "true")
.set("spark.rdd.compress", "true")
val sc = new SparkContext(sparkConfig)
val rdd = sc.parallelize(files)
rdd.filter(f => f.isFile && f.exists &&
Instant.ofEpochMilli(f.lastModified).plus(10, MINUTES).isAfter(Instant.now))
.collect.toList.sortBy(_.lastModified)
Still it takes the same time. And one thing I noticed is that filtering based on the file name is fast. But adding the lastModified filter makes it slow. Is there any better way so that I can get the results faster?
UPDATE
I updated the spark configs and now I'm able to get the results in less than 10 minutes. Earlier, I was running the jar like this:
spark-submit myJar.jar
I changed it to this:
spark-submit --deploy-mode client --queue SomeNonDefaultQueue --executor-memory 16g --num-executors 10 --executor-cores 1 --master yarn myJar.jar
Also removed set("spark.rdd.compress", "true")
from code as it increases the CPU time, as explained here - https://spark.apache.org/docs/2.3.0/configuration.html#compression-and-serialization
答案1
得分: 1
问题出在stat()
检查获取最后修改时间的操作是在线性搜索目录以查找名称之后执行的。如果您可以更改目录格式,请添加子目录(由文件名计算)并尝试将每个子目录中的条目数分组到约1000个。
否则,创建一个名称:lastModified的映射并使用WatchService
在触发事件时更新映射。
英文:
The problem is the stat()
check to get last modified comes after a linear search through the directory to look-up the name. If you can change the directory format, add subdirectories (calculated by file name) and try to group the number of entries in each subdirectory to ~1000.
Otherwise, create a map of name:lastModified and use the WatchService
to update the map whenever an event is fired.
通过集体智慧和协作来改善编程学习和解决问题的方式。致力于成为全球开发者共同参与的知识库,让每个人都能够通过互相帮助和分享经验来进步。
评论