问题
I have a AWS Lambda function which:
- checks a Redis Elasticache instance,
- if the item is not found in the cache, goes to Google Places API service.
The Redis instance is in a private subnet; so, to fetch it, I added the VPC and the subnet in which the instance resides. I also specified the security group which allows all the outbound traffic. The Network ACL is the default one which is supposed to all the inbound and the outbound traffic.
When adding VPC to Lambda function like that via the console, it prompts:
When you enable VPC, your Lambda function will lose default internet access. If you require external internet access for your function, ensure that your security group allows outbound connections and that your VPC has a NAT gateway.
So, in the Route Table of the private subnet, I added a NAT gateway too. However, at the point where the Google Places API service call is made from the Lambda function it is always doomed to result in timeout.
In short, I doubt that the NAT gateway properly allows internet access of the Lambda function. How can I check what goes wrong with it?
Do NAT Gateways log the calls or the call attempts being tried through it somehow in CloudWatch etc.?
回答1:
The following steps are required
- An IAM role with full VPC permission assigned to your lambda function.
- VPC with public and private subnet
- while creating a NAT Gateway a)the subnet has to be public subnet b)Elastic IP creat a new one or allocate one
- Create the route table and add another route with target as our NAT gateway we created above. And your lambda should be happy now
回答2:
The problem for my case turned out to the fact that, I had created the NAT Gateway in the private subnet.
Make sure you place the NAT Gateways in the public subnet.
By the way, there are metrics but no direct logging records available in CloudWatch for NAT Gateway.
回答3:
I want to elaborate on the answer from @vahdet. I was losing my mind trying to reconcile how the NAT Gateway was supposed to be in the public and private subnets simultaneously. It seemed like the official AWS documentation here was wrong, but of course it's not. There is a very subtle detail that myself and others have missed.
The NAT Gateway has to be "hot-wired" across two different subnets simultaneously in order to bridge private addresses to a public one that is internet facing.
First, I tried to put the NAT Gateway in the same route table as the IGW, but of course that doesn't work because you can't have two identical routes (0.0.0.0/0) with different targets.
The guide was saying to put the NAT Gateway in the route table for the Private Network, which I did, but that didn't seem to work.
The critical detail I was missing was that the NAT Gateway has to be created in a public subnet. The documentation actually says this, but it seems confusing because we are later told to put the route for NAT Gateway in the private table.
Both things are true. Create the NAT Gateway in the public subnet and then only add a route table entry in the private route table.
The documentation tells you to create the following network resources in the VPC:
- two new subnets
- two new route tables
- one new NAT Gateway
I already had a route table and some subnets, so I tried to only add one new subnet and one new route table and this is where I got into trouble. It really was better to create two of each as documented.
Here's what it the subnets look like for me:
subnet-public 10.8.9.0/24 us-east-1a
subnet-private 10.8.8.0/24 us-east-1a
Then create the NAT Gateway in subnet-public. It will be pending for a couple of minutes, which is important, because it must go to available status before it can be referenced in a route table entry.
Here are the route tables:
route-table-public
10.8.0.0/16 local
0.0.0.0/0 igw-xyz
subnet-association: subnet-public
route-table-private
10.8.0.0/16 local
0.0.0.0/0 nat-abc
subnet-association: subnet-private
Do you see what happened there? It's really subtle. The NAT Gateway is cross-wired. It "lives" in the public subnet it was created in, but all traffic in the private subnet gets routed to it.
If you create the NAT Gateway in the private subnet like I did at first, then the NAT Gateway is just as isolated as everything else in the private subnet, and has no way to route traffic out to the internet. It must be created in the public subnet to have internet access, because it must have an IP address inside the public subnet. My NAT Gateway got an internal IP of 10.8.9.127 and an external IP in the 54.X.X.X range.
By making the NAT Gateway the 0.0.0.0/0 route in the private routing table, we are telling all non-10.8.0.0/16 traffic to go straight to the NAT Gateway, even though it isn't actually inside the private subnet.
The VPC itself knows how to bridge traffic across its own subnets, and is able to send the 10.8.8.X traffic to the NAT Gateway's 10.8.9.X IP. It then acts as a bridge, and translates all of that traffic across it's internal IP to its external IP. Because it is in a public subnet that is in a route table with an internet gateway, the external IP has a clear path to the internet.
Now my private VPC lambda in subnet-private has verified internet access through the NAT Gateway.
来源：https://stackoverflow.com/questions/52155921/aws-lambda-nat-gateway-internet-access-results-in-timeout