I am wondering if someone can explain how the distributed cache works in Hadoop. I am running a job many times, and after each run I notice that the local distributed cache folder on each node is growing in size.
Is there a way for multiple jobs to re-use the same file in the distributed cache? Or is the distributed cache only valid for the lifetime of any individual job?
The reason I am confused is that the Hadoop documentation mentions that "DistributedCache tracks modification timestamps of the cache files", so this leads me to believe that if the time stamp hasn't changed, then it should not need to re-cache or re-copy the files to the nodes.
I am adding files successfully to the distributed cache using:
DistributedCache.addFileToClassPath(hdfsPath, conf);
DistributedCache uses reference counting to manage the caches. org.apache.hadoop.filecache.TrackerDistributedCacheManager.CleanupThread is in charge of cleaning up the CacheDirs whose reference count is 0. It will check every minute (default period is 1 minute, you can set it by "mapreduce.tasktracker.distributedcache.checkperiod").
When a Job finishes or fails, JobTracker will send a org.apache.hadoop.mapred.KillJobAction to the TaskTrackers. Then if a TaskTracker receives a KillJobAction, it puts the action to tasksToCleanup. In the TaskTracker, there is a background Thread called taskCleanupThread which takes the action from tasksToCleanup and do the cleanup work. For a KillJobAction, it will invoke purgeJob to clean up the Job. In this method, it will decrease the reference count used by this Job (rjob.distCacheMgr.release();).
The above analysis bases on hadoop-core-2.0.0-mr1-cdh4.2.1-sources.jar. I also checked the hadoop-core-0.20.2-cdh3u1-sources.jar and found there was a litte difference between this two versions. For example, there was not a org.apache.hadoop.filecache.TrackerDistributedCacheManager.CleanupThread in 0.20.2-cdh3u1. When initializing a Job, TrackerDistributedCacheManager will check if there is enough space to put the new caches files for this Job. If not, it will delete the caches which have 0 reference count.
If you are using cdh4.2.1, you can increase "mapreduce.tasktracker.distributedcache.checkperiod" to let the clean up work delay. Then the probability that multiple Jobs use the same distributed cache is increased.
If you are using cdh3u1, you can increase the limitation of the cache size("local.cache.size", default is 10G) and the max directories for caches("mapreduce.tasktracker.cache.local.numberdirectories", default is 10000). This can be also applied to cdh4.2.1.
If you look closely at what this book says, is that there is a limit of what can be stored in Distributed Cache. By default it's 10GB (configurable). There can be multiple different jobs running in the cluster concurrently. Furthermore, Hadoop kind of guarantees the files stay available in the cache for a single job as it is maintained by reference count done by the tasktracker for different tasks accessing the files in cache. In your case, for subsequent Jobs, the files may not be there as they are already marked for deletion.
Please correct me if you disagree anywhere. I'll be glad to discuss this further.
According to this: http://www.datasalt.com/2011/05/handling-dependencies-and-configuration-in-java-hadoop-projects-efficiently/
You should be able to do this via DistributedCache API instead of "-libjars"
Related
I'm running a hive query against a hadoop cluster of 3 nodes. And I am getting an error which says "Too many fetch failures". My hive query is:
insert overwrite table tablename1 partition(namep)
select id,name,substring(name,5,2) as namep from tablename2;
that's the query im trying to run. All i want to do is transfer data from tablename2 to tablename1. Any help is appreciated.
This can be caused by various hadoop configuration issues. Here a couple to look for in particular:
DNS issue : examine your /etc/hosts
Not enough http threads on the mapper side for the reducer
Some suggested fixes (from Cloudera troubleshooting)
set mapred.reduce.slowstart.completed.maps = 0.80
tasktracker.http.threads = 80
mapred.reduce.parallel.copies = sqrt (node count) but in any case >= 10
Here is link to troubleshooting for more details
http://www.slideshare.net/cloudera/hadoop-troubleshooting-101-kate-ting-cloudera
Update for 2020 Things have changed a lot and AWS mostly rules the roost. Here is some troubleshooting for it
https://docs.aws.amazon.com/emr/latest/ManagementGuide/emr-troubleshoot-error-resource-1.html
Too many fetch-failures
PDF
Kindle
The presence of "Too many fetch-failures" or "Error reading task output" error messages in step or task attempt logs indicates the running task is dependent on the output of another task. This often occurs when a reduce task is queued to execute and requires the output of one or more map tasks and the output is not yet available.
There are several reasons the output may not be available:
The prerequisite task is still processing. This is often a map task.
The data may be unavailable due to poor network connectivity if the data is located on a different instance.
If HDFS is used to retrieve the output, there may be an issue with HDFS.
The most common cause of this error is that the previous task is still processing. This is especially likely if the errors are occurring when the reduce tasks are first trying to run. You can check whether this is the case by reviewing the syslog log for the cluster step that is returning the error. If the syslog shows both map and reduce tasks making progress, this indicates that the reduce phase has started while there are map tasks that have not yet completed.
One thing to look for in the logs is a map progress percentage that goes to 100% and then drops back to a lower value. When the map percentage is at 100%, this does not mean that all map tasks are completed. It simply means that Hadoop is executing all the map tasks. If this value drops back below 100%, it means that a map task has failed and, depending on the configuration, Hadoop may try to reschedule the task. If the map percentage stays at 100% in the logs, look at the CloudWatch metrics, specifically RunningMapTasks, to check whether the map task is still processing. You can also find this information using the Hadoop web interface on the master node.
If you are seeing this issue, there are several things you can try:
Instruct the reduce phase to wait longer before starting. You can do this by altering the Hadoop configuration setting mapred.reduce.slowstart.completed.maps to a longer time. For more information, see Create Bootstrap Actions to Install Additional Software.
Match the reducer count to the total reducer capability of the cluster. You do this by adjusting the Hadoop configuration setting mapred.reduce.tasks for the job.
Use a combiner class code to minimize the amount of outputs that need to be fetched.
Check that there are no issues with the Amazon EC2 service that are affecting the network performance of the cluster. You can do this using the Service Health Dashboard.
Review the CPU and memory resources of the instances in your cluster to make sure that your data processing is not overwhelming the resources of your nodes. For more information, see Configure Cluster Hardware and Networking.
Check the version of the Amazon Machine Image (AMI) used in your Amazon EMR cluster. If the version is 2.3.0 through 2.4.4 inclusive, update to a later version. AMI versions in the specified range use a version of Jetty that may fail to deliver output from the map phase. The fetch error occurs when the reducers cannot obtain output from the map phase.
Jetty is an open-source HTTP server that is used for machine to machine communications within a Hadoop cluster
I am trying to find out how many MASTER, CORE, TASK instances are optimal to my jobs. I couldn't find any tutorial that explains how do I figure it out.
How do I know if I need more than 1 core instance? What are the "symptoms" I would see in EMR's console in the metrics that would hint I need more than one core? So far when I tried the same job with 1*core+7*task instances it ran pretty much like on 8*core, but it doesn't make much sense to me. Or is it possible that my job is so much CPU bound that the IO is such minor? (I have a map-only job that parses apache log files into csv file)
Is there such a thing to have more than 1 master instance? If yes, when is it needed? I wonder, because my master node pretty much is just waiting for the other nodes to do the job (0%CPU) for 95% of the time.
Can the master and the core node be identical? I can have a master only cluster, when the 1 and only node does everything. It looks like it would be logical to be able to have a cluster with 1 node that is the master and the core , and the rest are task nodes, but it seems to be impossible to set it up that way with EMR. Why is that?
The master instance acts as a manager and coordinates everything that goes in the whole cluster. As such, it has to exist in every job flow you run but just one instance is all you need. Unless you are deploying a single-node cluster (in which case the master instance is the only node running), it does not do any heavy lifting as far as actual MapReducing is concerned, so the instance does not have to be a powerful machine.
The number of core instances that you need really depends on the job and how fast you want to process it, so there is no single correct answer. A good thing is that you can resize the core/task instance group, so if you think your job is running slow, then you can add more instances to a running process.
One important difference between core and task instance groups is that the core instances store actual data on HDFS whereas task instances do not. In turn, you can only increase the core instance group (because removing running instances would lose the data on those instances). On the other hand, you can both increase and decrease the task instance group by adding or removing task instances.
So these two types of instances can be used to adjust the processing power of your job. Typically, you use ondemand instances for core instances because they must be running all the time and cannot be lost, and you use spot instances for task instances because losing task instances do not kill the entire job (e.g., the tasks not finished by task instances will be rerun on core instances). This is one way to run a large cluster cost-effectively by using spot instances.
The general description of each instance type is available here:
http://docs.aws.amazon.com/ElasticMapReduce/latest/DeveloperGuide/InstanceGroups.html
Also, this video may be useful for using EMR effectively:
https://www.youtube.com/watch?v=a5D_bs7E3uc
We have a system that receives archives on a specified directory and on a regular basis it launches a mapreduce job that opens the archives and processes the files within them. To avoid re-processing the same archives the next time, we're hooked into the close() method on our RecordReader to have it deleted after the last entry is read.
The problem with this approach (we think) is that if a particular mapping fails, the next mapper that makes another attempt at it finds that the original file has been deleted by the record reader from the first one and it bombs out. We think the way to go is to hold off until all the mapping and reducing is complete and then delete the input archives.
Is this the best way to do this?
If so, how can we obtain a listing of all the input files found by the system from the main program? (we can't just scrub the whole input dir, new files may be present)
i.e.:
. . .
job.waitForCompletion(true);
(we're done, delete input files, how?)
return 0;
}
Couple comments.
I think this design is heartache-prone. What happens when you discover that someone deployed a messed up algorithm to your MR cluster and you have to backfill a month's worth of archives? They're gone now. What happens when processing takes longer than expected and a new job needs to start before the old one is completely done? Too many files are present and some get reprocessed. What about when the job starts while an archive is still in flight? Etc.
One way out of this trap is to have the archives go to a rotating location based on time, and either purge the records yourself or (in the case of something like S3) establish a retention policy that allows a certain window for operations. Also whatever the back end map reduce processing is doing could be idempotent: processing the same record twice should not be any different than processing it once. Something tells me that if you're reducing your dataset, that property will be difficult to guarantee.
At the very least you could rename the files you processed instead of deleting them right away and use a glob expression to define your input that does not include the renamed files. There are still race conditions as I mentioned above.
You could use a queue such as Amazon SQS to record the delivery of an archive, and your InputFormat could pull these entries rather than listing the archive folder when determining the input splits. But reprocessing or backfilling becomes problematic without additional infrastructure.
All that being said, the list of splits is generated by the InputFormat. Write a decorator around that and you can stash the split list wherever you want for use by the master after the job is done.
The simplest way would probably be do a multiple input job, read the directory for the files before you run the job and pass those instead of a directory to the job (then delete the files in the list after the job is done).
Based on the situation you are explaining I can suggest the following solution:-
1.The process of data monitoring I.e monitoring the directory into which the archives are landing should be done by a separate process. That separate process can use some metadata table like in mysql to put status entries based on monitoring the directories. The metadata entries can also check for duplicacy.
2. Now based on the metadata entry a separate process can handle the map reduce job triggering part. Some status could be checked in metadata for triggering the jobs.
I think you should use Apache Oozie to manage your workflow. From Oozie's website (bolding is mine):
Oozie is a workflow scheduler system to manage Apache Hadoop jobs.
...
Oozie Coordinator jobs are recurrent Oozie Workflow jobs triggered by time (frequency) and data availabilty.
When do the outputs for a mapper task get deleted from the local filesystem? Do they persist until the entire job completes or do they get deleted at an earlier time than that?
In addition to the map and reduce tasks, two further tasks are created: a job setup task
and a job cleanup task. These are run by tasktrackers and are used to run code to setup
the job before any map tasks run, and to cleanup after all the reduce tasks are complete.
The OutputCommitter that is configured for the job determines the code to be run, and
by default this is a FileOutputCommitter. For the job setup task it will create the final
output directory for the job and the temporary working space for the task output, and
for the job cleanup task it will delete the temporary working space for the task output.
Have a look at OutputCommitter.
If your hadoop.tmp.dir is set to a default setting (say, /tmp/), it will most likely be subject to tmpwatch and any default settings in your OS. I would suggest poking around in /etc/cron.d/, /etc/cron.daily, etc/cron.weekly/, etc., to see exactly what your OS default is like.
One thing to keep in mind about tmpwatch is that, by default, it will key on access time, not modification time (i.e., files that have not been 'touched' since X will be considered 'stale' and subject to removal). However, it's a common practice with Hadoop to mount filesystems with the noatime and nodiratime flags, meaning that access times will not get updated and thus skewing your tmpwatch behaviors.
Otherwise, Hadoop will purge task attempt logs older than 24 hours (after task completion), by default. While a few years old, this writeup has some great info on the default behaviors. Take a look in particular at the sections that refer to mapreduce.job.userlog.retain.hours.
EDIT: responding to OP's comment, which clears up my misunderstanding of the question:
As far as the intermediate output of map tasks which is spilled to disk, used by any combiners, and copied to any reducers, the Hadoop Definitive Guide has this to say:
Tasktrackers do not delete map outputs from disk as soon as the first
reducer has retrieved them, as the reducer may fail. Instead, they
wait until they are told to delete them by the jobtracker, which is
after the job has completed.
Source
I've also +1'd #mgs answer below, as they have linked the source code that controls this and described the Job cleanup task.
So, yes, the map output data is deleted immediately after the job completes, successfully or not, and no sooner.
"Tasktrackers do not delete map outputs from disk as soon as the first reducer has retrieved them, as the reducer may fail. Instead, they wait until they are told to delete them by the jobtracker, which is after the job has completed"
Hadoop: The Definitive Guide ( Section 6.4)
Does it have any measurable effect on resources whether I submit a bunch of hadoop jobs from different client servers or all from the same one? I would think not since all the work is done in the cluster. Is this correct?
The only thing which is resource intensive on the client submitting to the Hadoop cluster is the calculation of the input splits. When the input data is huge or when too many jobs are submitted from the same client then because of the input split calculations, the job submission might become a bit slow.
I am not able to recall the Hadoop release or the parameter, but a configurable parameter was included to move the calculation of the input splits from the client submitting a job to the Hadoop cluster.
It really shouldn't matter where you submit your jobs from. The client itself doesn't do much, it uses RPC protocol to contact the services, and then just sits idle until the job is finished.
Also, the most important is what kind of scheduler you use to allocate resource, which is probably going to make the most significant difference and decide which resources to allocate to which job. More on job scheduling here.
I don't think you can move the input split calculation into Job Tracker in 'Classic' version. In YARN, you can move it using
"yarn.app.mapreduce.am.compute-splits-in-cluster"
I am guessing, Hadoop people didn't want to overload Job tracker with input split creation. Similar to the design decision of not assigning too much work for Namenode in HDFS.
In YARN, every job gets its own Application Master, so no worries about overloading a SPOF/bottleneck master like job tracker.
In reference to the original question, the client job would have to reach out to the namenode to get the block locations (I have see parts of code on block storage class calling data node for some meta data...not sure whether these happen during input split creation or in task tracker node) . This can become an issue if you are handling a lot of jobs on the same client node.
If you are using YARN, there would be a slight performance increase if all these communications happen inside the cluster.
Need to check how Oozie handles this issue.
Hopefully, this helps!
Arun