My use case is as follows:
I have a python script which:
1. reads a file from S3
2. processes the file and outputs a new file
3. saves the output file to S3 (or maybe a database)
The python script has some dependencies which are managed via virtualenv.
What is the recommended/easiest way of running these scripts in parallel on AWS?
I see the following options:
AWS Batch: Looks really complicated - I have to build my own Docker container, set up 3 different users, it's not easy to debug.
AWS Lambda: A bit easier to set up, but I still have to wrap my script up into a Lambda function. Debugging doesn't seem too straightforward
Slurm on manually spun up EC2 instances - From a user perspective, this is ideal - all I would have to do is just create a jobs.sbatch file which loads the virtualenv and runs the script. The main drawback is that I have to install and configure slurm.
What is the recommended way of handing this workflow?
Lambda will be suitable for you because you won't have to look at scaling nor get into setting up all the things. About the debugging, you can easily do it using sls wsgi serve
I think you can use a publish/subscribe mechanism by using an SQS queue containing the object key to work on. Then you can have a group of EC2 instances or ECS each subscribing the queue and performing the single operation. With the queue you ensure each process work on a single instance of the problem. I think it is possible to create an auto scaling group in ECS and you probably can change the number of machines to tune the performance/cost.
Related
I have scheduled 2 cronjobs for my application.
My Application server is in an autoscaling group and I kept a minimum of 2 instances because of High availability. Everything working is fine but cron job is running multiple times because of 2 instances in autoscaling.
I could not limit the instance size to 1 because already my application in the production environment I prefer to have HA.
How should I have to limit execute cron job on a single instance? or should i have to use other services like AWS Lamda or AWS ELasticBeanstalk
Firstly you should consider whether running the crons on these instances is suitable. If you're trying to keep this highly available and it is directly interacted via customers what will the impact of the crons performance be?
Perhaps consider using a separate autoscaling group or instance with a total of 1 instances to run these crons? You could launch the instance or update the autoscaling group just before the cron needs to run and then automate the shutdown after it has completed.
Otherwise you would need to consider using a locking mechanism for your script. By using this your script write a lock to confirm that it is in process, at the beginning of the script run it would check whether there was any script lock in progress. To further prevent the chance of a collision between multiple servers consider adding jitter (random seconds of sleep) to the start of your script.
Suitable technologies for writing a lock are below:
DynamoDB using strongly consistent reads.
EFS for a Linux application, or FSX for a Windows application.
S3 using strong consistency.
Solutions suggested by Chris Williams sound reasonable if using lambda function is not an option.
One way to simulate cron job is by using CloudWatch Events (now known as EventBridge) in conjunction with AWS Lambda.
First you need to write a Lambda function with the code that needs to be executed on a schedule. Lambda supports cron expressions.
You can then use Schedule Expressions with EventBridge/CloudWatch Event in the same way as a cron tab and mention the Lambda function as target.
you can enable termination protection on of the instance. Attach necessary role & permission for system manager. once the instance is available under managed instance under system manager you can create a schedule event in cloudwatch to run ssm documents. if you are running a bash script convert that to ssm document and set this doc as targate. or you can use shellscript document for running commands
Currently have some task-based automation for ECS that run on a scheduled basis, however sometimes there is a need to run only run task or re-run tasks for only a certain kinds of tasks (for example sql tasks or datadog tasks).
I know this can be done via console, but it's inefficient. Was thinking of a bash script that calls to start a task from a CLI. Currently I know I can do this with the AWS CLI using '--task-definition', but it's not much better. I don't usually write scripts, so I'm basically here to help with brainstorming. I'm wondering if there is a way to make an API call to start tasks. Would I need to type in the ARN every time? Can I just list the tasks on the AWS CLI and have the exported to the script? Would network-config need to be hard-coded?
Thanks!
The AWS API calls to start a task are:
StartTask:
Starts a new task from the specified task definition on the specified container instance or instances.
RunTask:
Starts a new task using the specified task definition. You can allow Amazon ECS to place tasks for you, or you can customize how Amazon ECS places tasks using placement constraints and placement strategies.
Since this is AWS API calls, there are equivalent calls in CLI and SDK.
I have few python scripts which need to be executed in sequence on AWS Cloud so what are the best and simplest options? These script files are proof of concept so little bit dirty also but need to run overnight. Most of the script finishes within 10 mins but couple of them can take up to 1 hour running on a single core.
We do not have any servers like Jenkins, airflow etc...we are planning to use existing aws services.
Please let me know, Thanks.
1) EC2 Instance (Manually controlled)
Upload your scripts to an S3 bucket Use default VPC
launch EC2 Instance
Use SSM Remote session to log in
Run AWS CLI (AWS S3 Sync to download from S3)
Run them Manually
stop instance when done.
To be clean, make a SH file (or master .py file) to do the work. If you want it to stop charging you money afterwards, add command to stop instance when complete.
Least amount of work
2) If you want to run scripts daily
- Script out the work above (include modifying the Autoscale group at end to go to one box)
- Create an EC2 Auto Scale Group and launch it on a CRON job schedule.
It will start up, do the work, and then shut down and stop charging you.
3) Lambda
Pretty much like option 2, but AWS will do most of the work for you.
Either put all your scripts into one lambda..or put each script into its own lambda and have a master that does sync invoke of each script in the order you want.
You have a cloudwatch alarm trigger daily and does the work
I would say that if you are in POC mode, option 1 is best decision. It is likely closest to what you already do where you are currently executing. This is what #jarmod recommended already.
You didn't mention anything about which AWS resources your python scripts need to access or at least the purpose of the scripts, so it is difficult to provide a solution.
However a good option is to use AWS Batch.
Currently working on an environment requirement where we are to push the same file out to multiple EC2 instances running Windows on a scheduled interval. As it stands now, I see a few options and have tried each:
Windows Task Manager: run a basic task on a set schedule invoking the S3 Sync CLI tool
Cons I can see here include: setting up the task on each EC2 instance (there are many).
Lambda: scheduled lambda job that utilizes SSM to run commands on each server in a resource group
Cons: introducing another layer required to execute this task.
Run Command: using an AWS-RunRemoteScript document, run the script (stored in S3) bucket on target instances.
Cons: I'm not positive you can automate these commands on a schedule without adding another layer.
What is the most scalable path forward? Thanks in advance for your help.
Using the Run Command feature of AWS Systems Manager together with either the Maintenance Window feature of AWS Systems Manager or using CloudWatch Events to schedule the execution of Run Command should be useful here.
If you also tag instances appropriately, you can use the tag targeting feature of Run Command to ensure that all instances run the command (including new instances launched in the future as long as they are tagged).
/Mats
I have an EC2 instance that is running a few processes. I also have a Lambda script that is triggered through various means. I would like this Lambda script to talk to my EC2 instance and get a list of running processes from it (Essentially run ps aux on the EC2 box, and read the output).
Now this is easy enough with just one instance and its instance-id. Just SSH in, run the command, get the output, and be on my way. However, I would like to scale this to multiple EC2 instances, for which only the instance-id is known and SSH keys may not be given.
Is such a configuration possible with Lambda and Boto (or other libraries)? Or do I just have to run a microserver on each of my instances that will reply with the given information (something I'm really trying to avoid)
You can do this easily with AWS Systems Manager - Run Command
AWS Systems Manager provides you safe, secure remote management of your instances at scale without logging into your servers, replacing the need for bastion hosts, SSH, or remote PowerShell.
Specifically:
Use the send-command API from Lambda function to get list of all processes on a group of instances. You can do this by providing a list of instances or even a tag query
You can also use CloudWatch Events to trigger a Run Command directly
I don't think there is something available out of the box for this scenario.
Instead of querying, try an alternate approach. Install an agent on all ec2 instances, which reports the required information to a central service or probably a DynamoDB table, with HashKey as InstanceId.
You may want to bake this script as a cron job, (executed probably hourly?) in the AMI itself.
With this implementation, you reduce the complexity of managing and running a separate web service on each EC2 instance.
Query the DynamoDB table on demand. There will be a lag, as data may not be real time, but you can always reduce the CRON interval per your needs.
Like Yeshodhan mentioned, There is no direct approach for this.
However, There is one more approach.
1) Save your private key file to an s3 bucket, Create a lambda function and use python fabric module to login to the remote machines from lambda function and execute commands.
The above-mentioned approach is possible but I highly recommend launching a separate machine and use a configuration management system (Preferably ansible) and get the results from remote machines.