Even after going through the AWS documentation and various blogs, I still don't understand how AWS lambda would behave when it is configured with VPC.
When AWS lambda configured with VPC, does that mean all instances of lambda would get the IP address from the specified subnet of that VPC?
How the ENI plays the role in AWS Lambda-VPC configuration?
The formula for ENI capacity from AWS doc -
Projected peak concurrent executions * (Memory in GB / 3GB)
Does it means AWS lambda’s running instance used to have 3 GB memory? And when that exceeds another ENI would get attached?
Most the AWS Lambda-VPC configuration related architecture diagrams shows Lambda inside VPC. Does that means Lambda would run inside VPC?
Here, I’m sure I’m missing a few pieces of information. Any pointers would be helpful.
When you configure a Lambda function to run in the VPC it uses an ENI that is created with and IP address in one of the subnets you select. Based on the formula of expected ENIs needed it seems that ENIs can be shared between lambdas.
There are only two reasons that I know of for running your lambda in a VPC.
It needs to access resources inside your VPC that do not have a public endpoint, e.g. Redis/Memcached caching clusters (Elasticache) or an RDS/Redshift cluster that doesn't have a public ip (good idea to not have public ip's on databases). When you lambda runs inside the VPC it uses a private ip and can connect to the private resources in your VPC
If you need to have your lambda's have a consistent IP address (perhaps a service that only allows whitelisting of IPs for authentication). This is achieved by using a NAT gateway.
Lambda functions cannot received inbound connections in any case.
Disadvantages of putting your lambda in a VPC are
Slower cold start times since a ENI might need to be provisioned.
You need a NAT gateway (or VPC endpoint) to access external resources
Needing to manage concurrency and available ip addresses more closely.
Related
I have a single AWS lambda function that connects to a single AWS RDS Postgres db and simply returns a json list of all records in the db.
If I don't assign a VPC to the lambda function, it is able to access the AWS RDS db. However, if I assign a VPC to the lambda function it can no longer access the db.
The VPC is the same for both the lambda function and the RDS db. I've also opened all traffic on port 0.0.0.0/0 for inbound and outbound connections temporarily to find the issue, but I am still unable to connect.
I believe it might be a role permission related to VPC for the lambda function, but I've already assigned the policy AmazonVPCFullAccess to the lambda role.
The fact that the lambda can access the DB when not in a VPC is a bit troubling in the sense that the DB is then probably public.
A common mistake that often happens is that lambda is deployed to a public subnet. Lambda's only get assigned private IP addresses in a VPC. When deployed to a public subnets, it's only route to the internet is the internet gateway. That doesn't really work well if the lambda itself has a private ip address (the internet couldn't route traffic back to you :P).
One part of the solution is to make sure your lambda is deployed to a private subnet instead with a route to a NAT gateway if it needs access to public resources.
However, the better part of the solution is actually put the database in the private subnet WITHOUT a public IP adresss.
Because I've seen many mistakes with this with my customers, and because it can't be stressed enough: I'd strongly suggest you follow a three-tier networking model with your VPC's. This basically means:
Don't use the default VPC. Create your own.
Create 9 subnets:
3 public
3 private. Put your private lambda's here.
3 isolated. Put your database here.
There are lot's of articles / templates available that do this for you. A quick google search gives me
https://github.com/aws-samples/vpc-multi-tier
https://www.wellarchitectedlabs.com/reliability/100_labs/100_deploy_cloudformation/1_deploy_vpc/
I am running AWS Lambda functions in a VPC.
And during the course of the project I have hit problems because:
no access to my database - had to solve this somehow
no access to AWS SES - had to find workaround
no access to AWS SQS -removed all queuing functionality from Lambda functions
no access to external Internet - still don't know how to implement ReCapthca
without Internet access
no access to AWS Cognito - cannot get
information about logged in users
I COULD implement a NAT gateway in the VPC but what is the point of serverless if I have to run a NAT server instance? That's not serverless.
So finally AWS has worn me down and I have decided to give up on running my AWS Lambda functions in a VPC - without endpoints for Internet proxying and the various AWS services its just too hard.
SO my question is - what is the downside/disadvantage of running my AWS Lambda functions with no VPC?
If you need access to resources within a VPC, then run your AWS Lambda function within a VPC. If you do not require this access, then do not run it within a VPC.
If you require Internet access, then you should connect your Lambda functions to a Private Subnet and use a NAT Gateway, which is a fully-managed NAT so you can remain serverless. It will solve the problems you listed.
AWS has provided a reference document for Lambda deployments: Serverless Application Lens, AWS Well-Architected Framework. In it they provide the following decision tree:
The only major downside noted is that a Lambda outside of a VPC cannot directly access private resources within a VPC.
One reason to create a Lambda in a VPC would be that you have a specific IP or IP range for it. This could be the case if a system just accepts calls from a specific IP which would need to be whitlistet for it.
Fix IP for Lambda function is discussed here: Is there a way to assign a Static IP to a AWS Lambda without VPC?
Downside of not having Lambda in VPC: Not having specific IP / IP-range for your Lambda function.
In the end I stayed with the VPC but I added an EC2 instance into the VPC and ran TinyProxy on it. I then configured my AWS Lambda functions with the environment variable:
HTTPS_PROXY https://ip-10-0-1-53.eu-west-1.compute.internal:8888
boto3 picked up the environment variable and sent all requests to the proxy. This seems to work fine without the complexity of a NAT gateway.
I have a Lambda function that needs to read messages from an SQS queue using it's URL. Then it needs to insert that data to Cassandra running on a server inside a VPC.
I am able to access the Cassandra server from my Lambda function, using it's private IP and configuring the security groups correctly.
However, I am not able to read messages from the SQS Queue. When I change the configuration of Lambda function to No VPC, then I am able to read the messages from the SQS Queue. However, with VPC settings, it just times out.
How can I overcome this ? I have checked the security group of my Lambda function has full outbound access to all IP addresses.
At the end of 2018, AWS announced support for SQS endpoints which provide
connectivity to Amazon SQS without requiring an internet gateway, network address translation (NAT) instance, or VPN connection.
There is a tutorial for Sending a Message to an Amazon SQS Queue from Amazon Virtual Private Cloud
See also the SQS VPC Endpoints Documentation for more information.
Its important to note that if you want to access SQS within the Lambda VPC there are a couple other things you need to do:
Make sure to specify the SQS region in your code. For example, I had to set my endpoint_url to "https://sqs.us-west-2.amazonaws.com"
Make sure that you have attached a "wide open" security group to the SQS VPC Interface, otherwise SQS will not work.
Make sure that your subnets in your Lambda VPC match what you have set up for your SQS VPC Interface.
Some services (e.g. S3) are offering VPC endpoints to solve this particular problem but SQS is not one of them. I think the only real solution to this problem is to run a NAT inside your VPC so the network traffic from the Lambda function can be routed to the outside world.
I ran into the same kind of problem when I was running lambda function with access to elasticache on the VPC. While the function was configured to run in the VPC, I wasnt able to talk to any other service (specifically codedeploy for me).
As #garnaat pointed out NAT seems to be the only way to go about solving this problem for services without VPC endpoints.
And like you pointed out, I also ran into the same trouble where I could'nt SSH into the machine(s) once I replaced the entry with the IGW in the route table. Seems like detaching the IGW starves the VPC of either the incoming traffic (mostly) or the outgoing traffic from or to the internet respectively. So here's what I did and it worked for me:
Create a new Subnet within the VPC
Now, when lambda runs, make sure lambda operates from this subnet.
You can do this by using aws-cli like so:
aws lambda update-function-configuration --function-name your-function-name --vpc-config SubnetIds="subnet-id-of-created-subnet",SecurityGroupIds="sg-1","sg-2"
Make sure you add all the security groups whose inbound and outbound traffic rules apply for your lambda function.
Next, go to Route Tables in the VPC console and create a new route table.
Here is where you add the NAT gateway to the target.
finally go to the Subnet Associations tab in the new route table and add the newly created subnet there.
Thats all this should get it working . Mind you, please treat this as only a workaround. I haven't done much digging and I have a very limited idea on how things get resolved internally while doing this. This might not be an ideal solution.
The ideal solution seems to be to design the VPC before hand. Use subnets to isolate resources/instances that need internet access and that dont(private and public subnets) and place appropriate gateways where needed.( so that you may not have to create a seperate subnet for this purpose later). Thanks
I was unable to get either of the other two answers to this question to work. Perhaps this is due to one or more mistakes on my part. Regardless, I did find a workaround that I wanted to share, in case I'm not alone with this problem.
Solution: I created two Lambda functions. The first Lambda function runs inside my VPC and performs the desired work (in mandeep_m91's case, that's a data insertion to Cassandra; in my case is was accessing an RDS instance). The second Lambda function lives outside the VPC, so I could hook it up to the SQS queue. I then had the second Lambda function call the first, using the information found this this StackOverflow Q&A answer. Note, the linked question has both node.js and Python examples in the answers.
This will effectively double the cost of making a function call, since each call results in two function executions. However, for my situation, the volume is so low it won't make a real difference.
To clarify a point above about a "wide open" security group, the group set on the endpoint needs to allow inbound access to SQS from your lambda function.
I created a security group for my endpoint that only opened 443 to my lambda's security group.
I am running into a security related issue with AWS lambda and not sure what is the right way to resolve this.
Consider an EC2 instance A accessing the database on another EC2 instance B. If I want to restrict the accessibility of the DB on instance B to instance A only, I would modify the security group and add a custom TCP rule to allow access to only the public IP of instance A. So, this way, AWS will take care of everything and the DB server will not be accessible from any other IP address.
Now let us replace instance A by a lambda function. Since it is no longer an instance, there is no definite IP address. So, how do I restrict access to only the lambda function and block any other traffic ?
Have the Lambda job determine its IP, and dynamically update the instance B security group, then reset the security group when done.
Until there is support for Lambda running within a VPC this is the only option. Support for that has been announced for later this year. The following quote is from the referenced link above.
Many AWS customers host microservices within a Amazon Virtual Private
Cloud and would like to be able to access them from their Lambda
functions. Perhaps they run a MongoDB cluster with lookup data, or
want to use Amazon ElastiCache as a stateful store for Lambda
functions, but don’t want to expose these resources to the Internet.
You will soon be able to access resources of this type by setting up
one or more security groups within the target VPC, configure them to
accept inbound traffic from Lambda, and attach them to the target VPC
subnets. Then you will need to specify the VPC, the subnets, and the
security groups when your create your Lambda function (you can also
add them to an existing function). You’ll also need to give your
function permission (via its IAM role) to access a couple of EC2
functions related to Elastic Networking.
This feature will be available later this year. I’ll have more info
(and a walk-through) when we launch it.
I believe the below link will explain lambda permission model for you.
http://docs.aws.amazon.com/lambda/latest/dg/intro-permission-model.html
I have for a long time a VPC (with 1 subnet) on Amazon Web Services (AWS) with several instances each having an Elastic IP address.
For new needs, I have defined a second VPC (with 1 subnet also) on my same account: for some reasons, I can't associate EIP (which is allocated with no problem) to instances launched in VPC #2: the interactive wizard of the console only presents me the instances of the first VPC.
Is it a known limitation or am I doing something wrong?
Two questions:
How many EIP's do you have on your account?
Is the 2nd VPC using a NAT instance to access the Internet?
EIP addresses should only be used on instances in subnets configured to route their traffic directly to the Internet Gateway. EIPs cannot be used on instances in subnets configured to use a NAT instance to access the Internet. (aws.amazon.com)