We are considering using airflow to replace our currently custom rq based workflow but I am unsure of the best way to design it. Or if it even makes sense to use airflow.
The use case is:
We get a data upload from a user.
Given the data types received we optionally run zero or more jobs
Each job runs if a certain combination of datatypes was received. It runs for that user for a time frame determined from the data received
The job reads data from a database and writes results to a database.
Further jobs are potentially triggered as a result of these jobs.
e.g.
After a data upload we put an item on a queue:
upload:
user: 'a'
data:
- type: datatype1
start: 1
end: 3
- type: datatype2
start: 2
end: 3
And this would trigger:
job1, user 'a', start: 1, end: 3
job2, user 'a', start: 2, end: 3
and then maybe job1 would have some clean up job that runs after it.
(Also it would be good if it was possible to be able to restrict jobs to only run if there are not other jobs running for the same user.)
Approaches I have considered:
1.
Trigger a DAG when data upload arrives on message queue.
Then this DAG determines which dependent jobs to to run and passes as arguments (or xcom) the user and the time range.
2.
Trigger a DAG when data upload arrives on message queue.
Then this DAG dynamically creates DAGS for the jobs based on datatypes and templates in the user and timeframe.
So you get dynamic DAGs for each user, job, timerange combo.
I'm not even sure how to trigger DAGs from a message queue... And finding it hard to find examples similar to this use case. Maybe that is because Airflow is not suited?
Any help/thoughts/advice would be greatly appreciated.
Thanks.
Airflow is built around time based schedules. It is not built to trigger runs based on the landing of data. There are other systems designed to do this instead. I heard something like pachyderm.io or maybe dvs.org. Even repurposing a CI tool or customizing a Jenkins setup could trigger based on file change events or a message queue.
However you can try to work it with Airflow by having an external queue listener use rest API calls to Airflow to trigger DAGs. EG if the queue is an AWS SNS queue you could have an AWS Lambda listener in simple Python do this.
I would recommend one DAG per job type (or is it user, whichever is less) which the trigger logic determines is correct based on the queue. If there's common clean up tasks and the like, the DAG might use a TriggerDagRunOperator to start those, or you might just have a common library of those clean up tasks that each DAG includes. I think the latter is cleaner.
DAGs can have their tasks limited to certain pools. You could make a pool per user so as to limit the runs of jobs per user. Alternatively if you have a DAG per user, you could set your max concurrent DAG runs for that DAG to something reasonable.
Related
I have a Azure Data Factory Pipeline. My trigger has been set for every each 5 minutes.
Sometimes my Pipeline takes more than 5 mins to finished its jobs. In this case, Trigger runs again and creates another instance of my Pipeline and two instances of the same pipeline make problem in my ETL.
How can I be sure than just one instance of my pipeline runs at time?
As you can see there are several instances running of my pipelines
Few options I could think of:
OPT 1
Specify 5 min timeout on your pipeline activities:
https://learn.microsoft.com/en-us/azure/data-factory/concepts-pipelines-activities
https://learn.microsoft.com/en-us/azure/data-factory/concepts-pipelines-activities#activity-policy
OPT 2
1) Create a 1 row 1 column sql RunStatus table: 1 will be our "completed", 0 - "running" status
2) At the end of your pipeline add a stored procedure activity that would set the bit to 1.
3) At the start of your pipeline add a lookup activity to read that bit.
4) The output of this lookup will then be used in if condition activity:
if 1 - start the pipeline's job, but before that add another stored procedure activity to set our status bit to 0.
if 0 - depending on the details of your project: do nothing, add a wait activity, send an email, etc.
To make a full use of this option, you can turn the table into a log, where the new line with start and end time will be added after each successful run (before initiating a new run, you can check if the previous run had the end time). Having this log might help you gather data on how much does it take to run your pipeline and perhaps either add more resources or increase the interval between the runs.
OPT 3
Monitor the pipeline run with SDKs (have not tried that, so this is just to possibly direct you):
https://learn.microsoft.com/en-us/azure/data-factory/monitor-programmatically
Hopefully you can use at least one of them
It sounds like you're trying to run a process more or less constantly, which is a good fit for tumbling window triggers. You can create a dependency such that the trigger is dependent on itself - so it won't run until the previous run has completed.
Start by creating a trigger that runs a pipeline on a tumbling window, then create a tumbling window trigger dependency. The section at the bottom of that article discusses "tumbling window self-dependency properties", which shows you what the code should look like once you've successfully set this up.
Try changing the concurrency of the pipeline to 1.
Link: https://www.datastackpros.com/2020/05/prevent-azure-data-factory-from-running.html
My first thought is that the recurrence is too frequent under these circumstances. If the graph you shared is all for the same pipeline, then most of them take close to 5 minutes, but you have some that take 30, 40, even 60 minutes. Situations like this are when a simple recurrence trigger probably isn't sufficient. What is supposed to happen while the 60 minute one is running? There will be 10-12 runs that wouldn't start: so they still need to run or can they be ignored?
To make sure all the pipelines run, and manage concurrency, you're going to need to build a queue manager of some kind. ADF cannot handle this itself, so I have built such a system internally and rely on it extensively. I use a combination of Logic Apps, Stored Procedures (Azure SQL), and Azure Functions to queue, execute, and monitor pipeline executions. Here is a high level break down of what you probably need:
Logic App 1: runs every 5 minutes and queues an ADF job in the SQL database.
Logic App 2: runs every 2-3 minutes and checks the queue to see if a) there is not a job currently running (status = 'InProgress') and 2) there is a job in the queue waiting to run (I do this with a Stored Procedure). IF this state is met: execute the next ADF and update its status to 'InProgress'.
I use an Azure Function to submit jobs instead of the built in Logic App activity because I have better control over variable parameters. Also, they can return the newly created ADF RunId, which I rely in #3.
Logic App 3: runs every minute and updates the status of any 'InProgress' jobs.
I use an Azure Function to check the status of the ADF pipeline based on RunId.
I currently have a setup that creates a job and then collect some metrics about the tasks in the job. I want to do something similar, but by setting a job schedule instead. In particular, I want to set a job schedule that wakes up at a recurrence interval that I specify, and run the same code that I was running when creating a job. What's the best way to go about doing that?
It seems that there is a CloudJobSchedule that I could use to set up my job schedule, but this only lets me create say a job manager task, and specify few properties. How can I run external code on the jobs created by the Job schedule?
It could also help to clarify how the CloudJobSchedule works. Specifically, after I commit my job schedule, what would happen programmatically. Does the code just move sequentially and run the rest of the code. In this case, does it make sense to get a reference to the last job created by the job schedule and run code on the job returned?
You'll want to create a CloudJobSchedule. You can specify the recurrence in the Schedule.
If you only need to run a single task per recurrence, your job manager task can simply be the task you need to run. If you need to run multiple tasks per job recurrence, your job manager needs to have logic to submit tasks to Batch and monitor for completion (if necessary).
When you submit a job schedule to Batch, your client side code will continue running. The behavior is no different than if you were submitting a regular job. You can retrieve the last job run via JobScheduleExecutionInformation and the RecentJob property.
I already saw this question How to implement custom job listener/tracker in Spark? and checked the source code to find out how to get the number of stages per job but is there any way to track programatically the % of jobs that got completed in a Spark app?
I can probably get the number of finished jobs with the listeners but I'm missing the total number of jobs that will be run.
I want to track progress of the whole app and it creates quite a few jobs but I can't find to find it anywhere.
#Edit: I know there's a REST endpoint for getting all the jobs in an app but:
I would prefer not to use REST but to get it in the app itself (spark running on AWS EMR/Yarn - getting the address probably is doable but I'd prefer to not do it)
that REST endpoint seems to be returning only jobs that are running/finished/failed so not total number of jobs.
After going through the source code a bit I guess there's no way to see upfront how many jobs will there be since I couldn't find any place where Spark would be doing such analysis upfront (as jobs are submitted in each action independently Spark doesn't have a big picture of all the jobs from the start).
This kind of makes sense because of how Spark divides work into:
jobs - which are started whenever the code which is run on the driver node encounters an action (i.e. collect(), take() etc.) and are supposed to compute a value and return it to the driver
stages - which are composed of sequences of tasks between which no data shuffling is required
tasks - computations of the same type which can run in parallel on worker nodes
So we do need to know stages and tasks upfront for a single job to create the DAG but we don't necessarily need to create a DAG of jobs, we can just create them "as we go".
I am sorry about being a little general here, but I am a little confused about how job scheduling works internally in spark. From the documentation here I get that it is some sort of implementation of Hadoop Fair Scheduler.
I am unable to come around to understand that who exactly are users here (are the linux users, hadoop users, spark clients?). I am also unable to understand how are the pools defined here. For example, In my hadoop cluster I have given resource allocation to two different pools (lets call them team 1 and team 2). But in spark cluster, wont different pools and the users in them instantiate their own spark context? Which again brings me to question that what parameters do I pass when I am setting property to spark.scheduler.pool.
I have a basic understanding of how driver instantiates a spark context and then splits them into task and jobs. May be I am missing the point completely here but I would really like to understand how Spark's internal scheduler works in context of actions, tasks and job
I find official documentation quite thorough and covering all your questions. However, one might find it hard to digest from the first time.
Let us put some definitions and rough analogues before we delve into details. application is what creates SparkContext sc and may be referred to as something you deploy with spark-submit. job is an action in spark definition of transformation and action meaning anything like count, collect etc.
There are two main and in some sense separate topics: Scheduling Across applications and Scheduling Within application. The former relates more to Resource Managers including Spark Standalone FIFO only mode and also concept of static and dynamic allocation.
The later, Scheduling Within Spark application is the matter of your question, as I understood from your comment. Let me try to describe what happens there at some level of abstraction.
Suppose, you submitted your application and you have two jobs
sc.textFile("..").count() //job1
sc.textFile("..").collect() //job2
If this code happens to be executed in the same thread there is no much interesting happening here, job2 and all its tasks get resources only after job1 is done.
Now say you have the following
thread1 { job1 }
thread2 { job2 }
This is getting interesting. By default, within your application scheduler will use FIFO to allocate resources to all the tasks of whichever job happens to appear to scheduler as first. Tasks for the other job will get resources only when there are spare cores and no more pending tasks from more "prioritized" first job.
Now suppose you set spark.scheduler.mode=FAIR for your application. From now on each job has a notion of pool it belongs to. If you do nothing then for every job pool label is "default". To set the label for your job you can do the following
sc.setLocalProperty("spark.scheduler.pool", "pool1").textFile("").count() // job1
sc.setLocalProperty("spark.scheduler.pool", "pool2").textFile("").collect() // job2
One important note here is that setLocalProperty is effective per thread and also all spawned threads. What it means for us? Well if you are within the same thread it means nothing as jobs are executed one after another.
However, once you have the following
thread1 { job1 } // pool1
thread2 { job2 } // pool2
job1 and job2 become unrelated in the sense of resource allocation. In general, properly configuring each pool in fairscheduler file with minShare > 0 you can be sure that jobs from different pools will have resources to proceed.
However, you can go even further. By default, within each pool jobs are queued up in a FIFO manner and this situation is basically the same as in the scenario when we have had FIFO mode and jobs from different threads. To change that you you need to change the pool in the xml file to have <schedulingMode>FAIR</schedulingMode>.
Given all that, if you just set spark.scheduler.mode=FAIR and let all the jobs fall into the same "default" pool, this is roughly the same as if you would use default spark.scheduler.mode=FIFO and have your jobs be launched in different threads. If you still just want single "default" fair pool just change config for "default" pool in xml file to reflect that.
To leverage the mechanism of pools you need to define the concept of user which is the same as setting "spark.scheduler.pool" from a proper thread to a proper value. For example, if your application listens to JMS, then a message processor may set the pool label for each message processing job depending on its content.
Eventually, not sure if the number of words is less than in the official doc, but hopefully it helps is some way :)
By default spark works with FIFO scheduler where jobs are executed in FIFO manner.
But if you have your cluster on YARN, YARN has pluggable scheduler, it means in YARN you can scheduler of your choice. If you are using YARN distributed by CDH you will have FAIR scheduler by deafult but you can also go for Capacity scheduler.
If you are using YARN distributed by HDP you will have CAPACITY scheduler by default and you can move to FAIR if you need that.
How Scheduler works with spark?
I'm assuming that you have your spark cluster on YARN.
When you submit a job in spark, it first hits your resource manager. Now your resource manager is responsible for all the scheduling and allocating resources. So its basically same as that of submitting a job in Hadoop.
How scheduler works?
Fair scheduling is a method of assigning resources to jobs such that all jobs get, on average, an equal share of resources over time. When there is a single job running, that job uses the entire cluster. When other jobs are submitted, tasks slots that free up are assigned to the new jobs, so that each job gets roughly the same amount of CPU time(using preemption killing all over used tasks). Unlike the default Hadoop scheduler(FIFO), which forms a queue of jobs, this lets short jobs finish in reasonable time while not starving long jobs. It is also a reasonable way to share a cluster between a number of users. Finally, fair sharing can also work with job priorities - the priorities are used as weights to determine the fraction of total compute time that each job should get.
The CapacityScheduler is designed to allow sharing a large cluster while giving each organization a minimum capacity guarantee. The central idea is that the available resources in the Hadoop Map-Reduce cluster are partitioned among multiple organizations who collectively fund the cluster based on computing needs. There is an added benefit that an organization can access any excess capacity no being used by others. This provides elasticity for the organizations in a cost-effective manner.
Spark internally uses FIFO/FCFS job scheduler. But, when you talk about the tasks, it works in a Round Robin fashion. It will be clear if we concentrate on the below example:
Suppose, the first job in Spark's own queue doesn't require all the resources of the cluster to be utilized; so, immediately second job in the queue will also start getting executed. Now, both jobs are running simultaneously. Each job has few tasks to be executed in order to execute the whole job. Assume, the first job assigns 10 tasks and the second one assigns 8. Then, those 18 tasks will share the CPU cycles of the whole cluster in a preemptive manner. If you want to further drill down, lets start with executors.
There will be few executors in the cluster. Assume the number is 6. So, in an ideal condition, each executor will be assigned 3 tasks and those 3 tasks will get same CPU time of the executors(separate JVM).
This is how spark internally schedules the tasks.
I am trying to use beanstalk for queuing a large number of periodic
tasks (for example, tasks need processed every N minutes), for each
task, if the last queued job is not completed (not reserved, i mean)
when current job to be added, the last queued job should be replaced
with current job, in other words, only the latest queued job of a task
should be processed.
how can i achieve that using beanstalk?
Ideas i have got right now is:
for each task, use memcached store its latest timestamps (set this
when add jobs to queue),
every time the worker reserved a job successfully, it first checks
timestamps for this task in memcached,
if timestamps of the job is same as timestamps in memcached, then
process this job,
otherwise skip this job, and delete it from the queue.
So is there better ways to do such work? please give your suggestions,
thanks.
I found a memcache/beanstalk combination also the best solution for an implementation where I didnt want a newer but identical job entering a queue.
Until 'named jobs' are done and the software released, that may be one of the better solutions.