TL;DR: AWS now has a managed connection pooling service for MySQL on RDS only (public preview right now). The article talks about using it in conjunction with Lambda.
It took me over half the article before I could parse "Amazon RDS Proxy for AWS Lambda" as a coherent sentence rather than a series of semi-related words strung together.
You may already know all this, but I thought I’d add to what you said.
JDBC doesn’t provide connection pooling out of the box. There are connection pooling libraries that layer into JDBC. Most frameworks and application servers do bundle a connection pool out of the box.
Lambda is just a different paradigm than a single monolithic app with a connection pool. I assume that’s what you’re getting at. :)
There are other use cases though. Tools like pgBouncer are fairly useful in a large enterprise where a single database has multiple clients. They provide a single point to coordinate failovers and manage resource limits. I generally prefer a 1:1 relationship between an application and a database, but sometimes you have to play the cards you’re dealt.
I really hope it can replace PgBouncer. We've spent way too much time getting it running in ECS, mostly due to us needing a separate PgBouncer service for the primary and replica RDS instances. And then a third ECS service running HAProxy in TCP mode to load balance between the PgBouncer services, to provide a read-only endpoint that balances between the primary and replica. Add issues like: having to patch in HAProxy TCP keep-alive
parameter support (it only allows enabling/disabling it, leaving params up to the OS, which you can't control on ECS Fargate, _and_ the AWS NLB having a 350-second TCP flow idle timeout - good luck with long-running queries); AWS NLBs having horrible timing issues when handling target group changes... it was not a great experience.
We also set up pgbouncer, but after testing we realized that running it in a container added extra latency so went with ec2 instances behind ASG.
For load blanacing we used ELBv1, we also tried to use NLB, but 350 second timeout, and broken behavior when after timeout it just silently closes the connections was a no-no. ELBv1 doesn't silently close connections, and idle timeout can be set to 1h. After that we configured pgbouncer to close idle connections after 55 minutes. ELBv1 also doesn't have the timing issues with health checks.
IMO NLB is maybe performant, but that's the only good thing about it, in every other category it is crap and for this use case ELBv1 isn't a bottleneck.
If you will plan to revise your setup, be aware that pgbouncer is single threaded so throwing multicore machines at it won't do much good. If CPU ends up being a bottleneck it is better to use a large single core instance and then adding more instances. The only problem with that is that as you're adding more instances you're increasing number of connections to the database, but you could have additional process that could monitor ASG and set the number of instances and dynamically adjust number of server-side connections (doesn't require restarting pgbouncer).
BTW: Since version 1.12.0 you can also run multiple pgbouncer instances listening on the same port so that can help make a better use of extra cores if you chose multicore instances.
If I were to do it again, I'd consider either that or building it directly on EC2, preparing the AMI with Packer or something, and handling scaling with an autoscaling group.
I, too, am curious how it'll fare compared to DIY PgBouncer. Related, I am building something related that solves a superset of these classes of struggles. I'd love to pick your brain, if you're interested! Do you mind shooting me a message? (email in profile)
As an aside, why isn't pgbouncer basically built into Postgres, at least as an option? I'm sure there's a good reason, but as a lay-user, pgbouncer seems to just do connections better than Postgres. It's just a pain to both with it when cloud databases are so easy to launch in comparison.
Culture in Pg community once favored separate, single purpose solutions. This was the replication story for a long time. So here's hoping they'll build-in something to help with the high costs of individual connections.
Finally, this has been a long time coming. Connecting to existing RDS instances from lambda functions has been a pain for users looking to adopt serverless tech into their existing applications.
Does anyone know why RDS Postgres always seems to lag behind their MySQL offering? I can imagine Postgres is that much more difficult to develop a connection pooling service for, especially since PGBouncer and PGPool-II both already exist and are widely used.
We use Aurora Serverless and this is a sorely needed feature for us. We get large spikes of traffic that also cause large increases in connections to our DB, but Aurora Serverless cannot scale fast enough to cater for this. Unfortunately it doesn't seem we can enable this proxy for serverless clusters.
Aurora serverless already has a proxy by default - if it can’t scale fast enough it’s likely because your queries are too long and interleaved to find a logical switchover point.
This proxy is probably an offshoot of the serverless proxy.
It has a hard limit on max_connections based on the "Aurora Capacity Unit", e.g. for 16 ACU's max_connections it is 2,000. We see spikes of 10x to 100x concurrent users in a matter of minutes (think hug of death style). And although our queries run fast and we see limited increase of CPU on the cluster we almost instantly hit the connection limit.
Free til the end of 2019, 0.015 per vCPU (of the underlying RDS instance) per hour the proxy is enabled. So it is not scaled down to 0 when not in use. For comparison, a t3.micro (2 vCPU, 1gb ram) is 0.0104 per hour. If you're a big project it might be worth it, but just keep in mind rolling your own proxy or using the RDS serverless APIs are cheaper options.
I don't understand why this isn't launching with data api support? AWS specifically built data apis into the Aurora Serverless product for this exact usecase...
The same functionality could have been accomplished by just allowing us to specify the max number of concurrent instances of a function that can be active at a time. With proper request queuing the problem would be solved. As usual, AWS solves a problem they failed to design for by charging for an overly complex solution. Flame me if you want but there are K8s serverless frameworks that address this out of the box.
31 comments
[ 4.1 ms ] story [ 65.7 ms ] threadHowever I do wish they’d include how this will evolve alongside last years announced data API.
Would love to see this come out for Postgres and come to Google Cloud :)
JDBC doesn’t provide connection pooling out of the box. There are connection pooling libraries that layer into JDBC. Most frameworks and application servers do bundle a connection pool out of the box.
Lambda is just a different paradigm than a single monolithic app with a connection pool. I assume that’s what you’re getting at. :)
There are other use cases though. Tools like pgBouncer are fairly useful in a large enterprise where a single database has multiple clients. They provide a single point to coordinate failovers and manage resource limits. I generally prefer a 1:1 relationship between an application and a database, but sometimes you have to play the cards you’re dealt.
For load blanacing we used ELBv1, we also tried to use NLB, but 350 second timeout, and broken behavior when after timeout it just silently closes the connections was a no-no. ELBv1 doesn't silently close connections, and idle timeout can be set to 1h. After that we configured pgbouncer to close idle connections after 55 minutes. ELBv1 also doesn't have the timing issues with health checks.
IMO NLB is maybe performant, but that's the only good thing about it, in every other category it is crap and for this use case ELBv1 isn't a bottleneck.
If you will plan to revise your setup, be aware that pgbouncer is single threaded so throwing multicore machines at it won't do much good. If CPU ends up being a bottleneck it is better to use a large single core instance and then adding more instances. The only problem with that is that as you're adding more instances you're increasing number of connections to the database, but you could have additional process that could monitor ASG and set the number of instances and dynamically adjust number of server-side connections (doesn't require restarting pgbouncer).
BTW: Since version 1.12.0 you can also run multiple pgbouncer instances listening on the same port so that can help make a better use of extra cores if you chose multicore instances.
The result format is a bit clunky but there's all the column information you'd expect.
I found this golang database interface which ended up dropping in as expected with a few tweaks.
https://github.com/Clever/rds/tree/birthday
My bill for Aurora Serverless Postgres last month was less than a dollar.
This proxy is probably an offshoot of the serverless proxy.
https://www.jeremydaly.com/aurora-serverless-the-good-the-ba...
Free til the end of 2019, 0.015 per vCPU (of the underlying RDS instance) per hour the proxy is enabled. So it is not scaled down to 0 when not in use. For comparison, a t3.micro (2 vCPU, 1gb ram) is 0.0104 per hour. If you're a big project it might be worth it, but just keep in mind rolling your own proxy or using the RDS serverless APIs are cheaper options.
https://docs.aws.amazon.com/AmazonRDS/latest/AuroraUserGuide...
If I'm using Aurora Serverless right now and want to switch over to RDS Proxy I'm going to have to rewrite my calls.
This does nothing to solve the issue however. There are many use cases where you can't delay handling requests.