From what I can tell, the 'Metal' offering runs on nodes with directly attached NVMe rather than network-attached storage. That means there isn't a per-customer IOPS cap – they actually market it as 'unlimited I/O' because you hit CPU before saturating the disk. The new $50 M-class clusters are essentially smaller versions of those nodes with adjustable CPU and RAM in AWS and GCP .
RE: EC2 shapes, it's not a shared EBS volume but a dedicated instance with local storage. BUT you'll still want to monitor capacity since the storage doesn't autoscale.
ALSO this pricing makes high-throughput Postgres accessible for indie projects, which is pretty neat.
Just want to add that you don't necessarily need to invest in fancy disk-usage monitoring as we always display it in the app and we start emailing database owners at 60% full to make sure no one misses it.
Yes, you can certainly use up your CPU allocation on an M-10 database (at which point we offer online resizing as large as you want to go, all the way up to 192 CPUs and 1.5TiB RAM). Even still, I've been able to coax more than 10,000 IOPS from an M-10. (Actually, out of dozens of M-10s colocated on the same hardware all hammering away.)
You can get a lot more out of that CPU allocation with the fast I/O of a local NVMe drive than from the slow I/O of an EBS volume.
Doesnt "Metal" infer you get the whole box to yourself? Curious if my definitions are different to others here because I don't get what's "Metal" about sharing an instance with others.
You're still sharing nvme IO, cpu, memory bandwidth, etc. Not having a VM isn't really the point. (EDIT: and could have been done with non-metal aws instances with direct-attached nvme anyway)
Within PlanetScale's product lineup, Metal refers to the use of local NVMe drives. Nothing more. These extremely affordable sizes are indeed slices of larger boxen, though no resources are overcommitted.
It might be slightly off topic but I have a hard time understanding the layout of the website on mobile, it's not clear what is clickable and what's not.
Do such small caps on CPU/RAM mean that multiple customers are sharing the same server? Is there concern for noisy neighbors here, either IOPS or in case another customer's workload grows to take the full available storage on the NVMe? What kind of downtime would be needed to switch to a larger size?
We've engineered in protections from noisy neighbors in both CPU and I/O usage and we do not over-commit resources.
If your or another customer's workload grows and needs to size up we launch three whole new database servers of the appropriate size (whether that's more CPU+RAM, more storage, or both), restore the most recent backups there, catch up on replication, and then orchestrate changing the primary.
Downtime when you resize typically amounts to needing to reconnect i.e. it's negligible.
Why is Metal not offered for single instance deploys? Our app does not need this kind of uptime. We would be happy with a node going down once in a while (no data loss, of course) with a little bit of downtime to save 66% on the cost of running 2 additional nodes that will never see action.
It's a durability thing, we need to make sure writes are replicated off to at least one node. There might be avenues to get Metal down to single node in the future.
I definitely think there are use-cases out there which are fine with daily backups. Not every use-case requires high availability or high durability.
Even to take a case in point where durability is irrelevant - people building caches in Postgres (so as to only have one datastore / not need Redis as well). Not a big deal if the cache blows up - just force everyone to login again. Would love to see the vendor reduce complexity on their end and pass through the savings to the customer.
edit: per your other reply re. using replication to handle resizing, maybe being upfront with customers about additional latency / downtime being necessary with single-node discounts, then for resizing you could break connections, take a backup, then restore the backup on a resized node?
I haven't read HN for a while, this appears to just be an advertisement, did the rules change and advertisements for new products are promoted like product placement in movies?
It is a product / feature announcement. Much like blog post talking about their products or AWS announcing new features at their summit. Apple Announcing new MacBook Pro.
If anyone from Planetscale is reading this, please know I hate what you did to your website. I previously had it bookmarked as an example of excellent, usable website design. About a year ago it turned into a plaintext nightmare. The first time I saw the new design I genuinely thought that a CSS file had failed to load in my browser. It's awful.
*Edit:* It also fails to load other pages if you have JavaScript or XHR disabled.
Same. Love PlanetScale, love their previous website design. I struggle reading white text on black backgrounds, so I don't even try to read their product pages or blog posts since there's no light mode :( yes I know about reader mode
It feels it went from "professional Stripe level design that you admire and it inspires you" to just "hard to read black website", not sure what for.
There's definitely a light mode for planetscale.com (the docs, the blog, the changelog, and the UI). Should work on both desktop and mobile. Make sure your browser is requesting light mode. The browser doesn't always follow your OS-level preferences.
Was curious what it looks like now, and yea, not a fan of the fake hacker "we don't do CSS or styling". But then again maybe I was just used to their old design
I don't think either is a bad choice, but Aurora has some advantages if you're not a DB expert. Starting with Aurora Serverless:
- Aurora storage scales with your needs, meaning that you don't need to worry about running out of space as your data grows.
- Aurora will auto-scale CPU and memory based on the needs of your application, within the bounds you set. It does this without any downtime, or even dropping connections. You don't have to worry about choosing the right CPU and memory up-front, and for most applications you can simply adjust your limits as you go. This is great for applications that are growing over time, or for applications with daily or weekly cycles of usage.
The other Aurora option is Aurora DSQL. The advantages of picking DSQL are:
- A generous free tier to get you going with development.
- Scale-to-zero and scale-up, on storage, CPU, and memory. If you aren't sending any traffic to your database it costs you nothing (except storage), and you can scale up to millions of transactions per second with no changes.
- No infrastructure to configure or manage, no updates, no thinking about replicas, etc. You don't have to understand CPU or memory ratios, think about software versions, think about primaries and secondaries, or any of that stuff. High availability, scaling of reads and writes, patching, etc is all built-in.
Perhaps a naive question — but why would someone use a dedicated database provider and connect from another cloud provider's application service? ...as opposed to using the same provider's db + app service offering?
Wouldn't this introduce additional latency among other issues?
I had the same latency concerns when I heard about this PaaS DB trend, but you’ll note that this runs in the AWS (soon GCP) region of your choice, so if you’re hosted there, it should be about the same latency as using their managed DB service.
If you aren’t hosting the app in the same AWS/GCP region then I still have the same question.
> so if you’re hosted there, it should be about the same latency as using their managed DB service.
yes and no. In my AWS account I can explicitly pick an AZ (us-east-2a, us-east-2b or us-east-2c) but Availability Zones are not consistent between AWS accounts.
But that's exactly why they introduced the AZ IDs (use1-az1 as opposed to us-east-1a), so you can tell whether you're really in the same zone, regardless of the name you see in a particular account.
From the PlanetScale perspective keep in mind the ability to shard. What happens when the largest single node Aurora instance can no longer keep up with application/traffic demands?
I ask because we see it more often than not, and for that situation sharding the workflow is the best answer. Why have one MySQL instance responding to request when you could have 2,4,8...128, etc MySQL instances responding as a single database instance? They also have the ability to vertically scale each of the shards in that database as it's needed.
PlanetScale operates databases in AWS and GCP. There's no network latency penalty for choosing PlanetScale if you're hosting your app in one of those cloud providers (and in one of the many regions we operate in).
It depends a bit on your cloud provider but some of them have an offering that doesn't always match your needs or their pricing might be much more expensive at equal performance.
We run on the same instance types the larger PlanetScale Metal sizes offer as whole instances. For Intel that's r6id, i4i, i7i, i3en, and i7ie. For ARM that's r8gd, i8g, and i8ge. (Right now, at least. AWS is always cookin' up new instance types.) Same story will soon be true for GCP.
$50 bucks gets you an EIGHTH of a vCPU, 1GB RAM, and 10GB SSD??? This is quite frankly highway robbery. Not to mention the laughable bandwidth. Hetzner will give you 16 vCPU, 32GB RAM, and 640GB SSD for less than that. We're talking over an order of magnitude difference in value here.
1 GB of RAM for Postgres is really only useful for tinkering IMHO. Even for development, you’ll quickly need more memory, so HA doesn’t provide much value here. If you go with something even remotely reasonable (4 GB RAM, 200 GB SSD, 1/2 vCPU — and that’s still on the low end), the cost jumps to about $290/month. For that price, you could easily hire someone to set up HA Postgres for you on Hetzner or OVH and once configured, HA Postgres typically requires minimal ongoing maintenance.
Also, this is a shared server, not a truly dedicated one like you’d get with bare-metal providers. So, calling it "Metal" might be misleading marketing trick, but if you want someone to always blame and don’t mind overpaying for that comfort, then the managed option might be the right thing.
Considering they are charging an unfathomable $4529/mo for 256 GB databases, extrapolating that to a serious use case you can indeed just hire someone full-time with how much you'd save. And then you'll actually have someone who understands how databases work instead of treating it like an expensive black box.
Yeah per your edit that'd be for 256GB RAM which puts that into serious dollar category. For comparison I checked what AWS asks for for the same spec and that'd be $4616/month (for a db.m8gd.16xlarge), and that doesn't even yield you an actual NVMe. You can of course build the same for cheaper on Hetzner but again then you're on the hook also for the operations of the thing, which at that size is possibly non-trivial.
> $4529/month... can indeed just hire someone full-time
That's $54,348/year, not including the cost of benefits, not including stock compensation. Let's say you reserve 20% for benefits and that comes out to $43,478.40 in salary.
Besides the benefit of not needing the management / communication overhead of hiring somebody, do you know any DBAs willing to take a full-time job for $43,478.40 in salary?
This will be faster than an equivalent RDS instance and will handle more of the operational lifecycle around failover and high-availability with less downtime than RDS.
I'm stoked about this release too...
From what I can tell, the 'Metal' offering runs on nodes with directly attached NVMe rather than network-attached storage. That means there isn't a per-customer IOPS cap – they actually market it as 'unlimited I/O' because you hit CPU before saturating the disk. The new $50 M-class clusters are essentially smaller versions of those nodes with adjustable CPU and RAM in AWS and GCP .
RE: EC2 shapes, it's not a shared EBS volume but a dedicated instance with local storage. BUT you'll still want to monitor capacity since the storage doesn't autoscale.
ALSO this pricing makes high-throughput Postgres accessible for indie projects, which is pretty neat.
Correct you are.
Just want to add that you don't necessarily need to invest in fancy disk-usage monitoring as we always display it in the app and we start emailing database owners at 60% full to make sure no one misses it.
> 'unlimited I/O' because you hit CPU before saturating the disk.
So in the M-10 case, wouldn't this actually be somewhat misleading as I imagine hitting "1/8 vCPU" wouldn't be difficult at all?
Yes, you can certainly use up your CPU allocation on an M-10 database (at which point we offer online resizing as large as you want to go, all the way up to 192 CPUs and 1.5TiB RAM). Even still, I've been able to coax more than 10,000 IOPS from an M-10. (Actually, out of dozens of M-10s colocated on the same hardware all hammering away.)
You can get a lot more out of that CPU allocation with the fast I/O of a local NVMe drive than from the slow I/O of an EBS volume.
Doesnt "Metal" infer you get the whole box to yourself? Curious if my definitions are different to others here because I don't get what's "Metal" about sharing an instance with others.
You're still sharing nvme IO, cpu, memory bandwidth, etc. Not having a VM isn't really the point. (EDIT: and could have been done with non-metal aws instances with direct-attached nvme anyway)
Within PlanetScale's product lineup, Metal refers to the use of local NVMe drives. Nothing more. These extremely affordable sizes are indeed slices of larger boxen, though no resources are overcommitted.
I also think this naming is misleading - there is a very clear association with "bare metal", which is not what is being offered here
Planetscale support has been top-notch to work with, ++. Keep up the great work y'all!
It might be slightly off topic but I have a hard time understanding the layout of the website on mobile, it's not clear what is clickable and what's not.
Thank you for the feedback.
Really excited for more people to get to use Metal. Let me know if you have any questions.
Do such small caps on CPU/RAM mean that multiple customers are sharing the same server? Is there concern for noisy neighbors here, either IOPS or in case another customer's workload grows to take the full available storage on the NVMe? What kind of downtime would be needed to switch to a larger size?
We've engineered in protections from noisy neighbors in both CPU and I/O usage and we do not over-commit resources.
If your or another customer's workload grows and needs to size up we launch three whole new database servers of the appropriate size (whether that's more CPU+RAM, more storage, or both), restore the most recent backups there, catch up on replication, and then orchestrate changing the primary.
Downtime when you resize typically amounts to needing to reconnect i.e. it's negligible.
Why is Metal not offered for single instance deploys? Our app does not need this kind of uptime. We would be happy with a node going down once in a while (no data loss, of course) with a little bit of downtime to save 66% on the cost of running 2 additional nodes that will never see action.
It's a durability thing, we need to make sure writes are replicated off to at least one node. There might be avenues to get Metal down to single node in the future.
I definitely think there are use-cases out there which are fine with daily backups. Not every use-case requires high availability or high durability.
Even to take a case in point where durability is irrelevant - people building caches in Postgres (so as to only have one datastore / not need Redis as well). Not a big deal if the cache blows up - just force everyone to login again. Would love to see the vendor reduce complexity on their end and pass through the savings to the customer.
edit: per your other reply re. using replication to handle resizing, maybe being upfront with customers about additional latency / downtime being necessary with single-node discounts, then for resizing you could break connections, take a backup, then restore the backup on a resized node?
I haven't read HN for a while, this appears to just be an advertisement, did the rules change and advertisements for new products are promoted like product placement in movies?
asking for a friend that liked this space
It is a product / feature announcement. Much like blog post talking about their products or AWS announcing new features at their summit. Apple Announcing new MacBook Pro.
If anyone from Planetscale is reading this, please know I hate what you did to your website. I previously had it bookmarked as an example of excellent, usable website design. About a year ago it turned into a plaintext nightmare. The first time I saw the new design I genuinely thought that a CSS file had failed to load in my browser. It's awful.
*Edit:* It also fails to load other pages if you have JavaScript or XHR disabled.
Same. Love PlanetScale, love their previous website design. I struggle reading white text on black backgrounds, so I don't even try to read their product pages or blog posts since there's no light mode :( yes I know about reader mode
It feels it went from "professional Stripe level design that you admire and it inspires you" to just "hard to read black website", not sure what for.
(not fully functional) https://web.archive.org/web/20240811142248/https://planetsca...
There's definitely a light mode for planetscale.com (the docs, the blog, the changelog, and the UI). Should work on both desktop and mobile. Make sure your browser is requesting light mode. The browser doesn't always follow your OS-level preferences.
Was curious what it looks like now, and yea, not a fan of the fake hacker "we don't do CSS or styling". But then again maybe I was just used to their old design
Design is subjective of course. I love their new website and much prefer it to the old one.
Can you provide an example of a website you approve?
For the less experienced devs, how should I be thinking about choosing between this vs Amazon Aurora?
I don't think either is a bad choice, but Aurora has some advantages if you're not a DB expert. Starting with Aurora Serverless:
- Aurora storage scales with your needs, meaning that you don't need to worry about running out of space as your data grows. - Aurora will auto-scale CPU and memory based on the needs of your application, within the bounds you set. It does this without any downtime, or even dropping connections. You don't have to worry about choosing the right CPU and memory up-front, and for most applications you can simply adjust your limits as you go. This is great for applications that are growing over time, or for applications with daily or weekly cycles of usage.
The other Aurora option is Aurora DSQL. The advantages of picking DSQL are:
- A generous free tier to get you going with development. - Scale-to-zero and scale-up, on storage, CPU, and memory. If you aren't sending any traffic to your database it costs you nothing (except storage), and you can scale up to millions of transactions per second with no changes. - No infrastructure to configure or manage, no updates, no thinking about replicas, etc. You don't have to understand CPU or memory ratios, think about software versions, think about primaries and secondaries, or any of that stuff. High availability, scaling of reads and writes, patching, etc is all built-in.
They have a very nice comparison in terms of performance and price https://planetscale.com/benchmarks/aurora
It will be faster and a lot easier to use than Aurora.
Perhaps a naive question — but why would someone use a dedicated database provider and connect from another cloud provider's application service? ...as opposed to using the same provider's db + app service offering?
Wouldn't this introduce additional latency among other issues?
I had the same latency concerns when I heard about this PaaS DB trend, but you’ll note that this runs in the AWS (soon GCP) region of your choice, so if you’re hosted there, it should be about the same latency as using their managed DB service.
If you aren’t hosting the app in the same AWS/GCP region then I still have the same question.
> so if you’re hosted there, it should be about the same latency as using their managed DB service.
yes and no. In my AWS account I can explicitly pick an AZ (us-east-2a, us-east-2b or us-east-2c) but Availability Zones are not consistent between AWS accounts.
See https://docs.aws.amazon.com/ram/latest/userguide/working-wit...
But that's exactly why they introduced the AZ IDs (use1-az1 as opposed to us-east-1a), so you can tell whether you're really in the same zone, regardless of the name you see in a particular account.
Ah, thanks Internet stranger. TIL.
From the PlanetScale perspective keep in mind the ability to shard. What happens when the largest single node Aurora instance can no longer keep up with application/traffic demands?
I ask because we see it more often than not, and for that situation sharding the workflow is the best answer. Why have one MySQL instance responding to request when you could have 2,4,8...128, etc MySQL instances responding as a single database instance? They also have the ability to vertically scale each of the shards in that database as it's needed.
PlanetScale operates databases in AWS and GCP. There's no network latency penalty for choosing PlanetScale if you're hosting your app in one of those cloud providers (and in one of the many regions we operate in).
More importantly, no bandwidth charge penalty. As leaving AWS isn't inexpensive.
It depends a bit on your cloud provider but some of them have an offering that doesn't always match your needs or their pricing might be much more expensive at equal performance.
$60/TB of egress is quite a lot
but at least you get a fraction of CPU and 1GB memory.
Sounds amazing, but i would rather be able to run the database locally and use the same in dev as in production. Is this possible?
PlanetScale's Postgres offering is as close to plain-old-Postgres as we could possibly build.
It's just a hosted Postgres database, you could run this locally with Docker for example: https://hub.docker.com/_/postgres
Sounds cool!
Would be curious to know what the underlying aws ec2 instance is.
Is each DB on a dedicated instance?
If not, are there per-customer iops bounds?
We run on the same instance types the larger PlanetScale Metal sizes offer as whole instances. For Intel that's r6id, i4i, i7i, i3en, and i7ie. For ARM that's r8gd, i8g, and i8ge. (Right now, at least. AWS is always cookin' up new instance types.) Same story will soon be true for GCP.
there aren't per customer IOPs limits but the CPU will be the bottleneck.
Off-topic: when will postgres 18 be offered on metal?
You can expect to see it in early 2026 for both Metal and EBS backed databases.
Will these smaller instances be offered for Vitess / MySQL compatible users?
Soon!
$50 bucks gets you an EIGHTH of a vCPU, 1GB RAM, and 10GB SSD??? This is quite frankly highway robbery. Not to mention the laughable bandwidth. Hetzner will give you 16 vCPU, 32GB RAM, and 640GB SSD for less than that. We're talking over an order of magnitude difference in value here.
You're not paying for the infra, you're paying for not having to hire people who would have had to build/manage/test/operate/secure it.
On Hetzner you will be on the hook for managing the database though, and DBA is most certainly a full-time job if you have a serious use-case for it.
1 GB of RAM for Postgres is really only useful for tinkering IMHO. Even for development, you’ll quickly need more memory, so HA doesn’t provide much value here. If you go with something even remotely reasonable (4 GB RAM, 200 GB SSD, 1/2 vCPU — and that’s still on the low end), the cost jumps to about $290/month. For that price, you could easily hire someone to set up HA Postgres for you on Hetzner or OVH and once configured, HA Postgres typically requires minimal ongoing maintenance.
Also, this is a shared server, not a truly dedicated one like you’d get with bare-metal providers. So, calling it "Metal" might be misleading marketing trick, but if you want someone to always blame and don’t mind overpaying for that comfort, then the managed option might be the right thing.
Considering they are charging an unfathomable $4529/mo for 256 GB databases, extrapolating that to a serious use case you can indeed just hire someone full-time with how much you'd save. And then you'll actually have someone who understands how databases work instead of treating it like an expensive black box.
edit: my bad that's the price for 256GB RAM.
Yeah per your edit that'd be for 256GB RAM which puts that into serious dollar category. For comparison I checked what AWS asks for for the same spec and that'd be $4616/month (for a db.m8gd.16xlarge), and that doesn't even yield you an actual NVMe. You can of course build the same for cheaper on Hetzner but again then you're on the hook also for the operations of the thing, which at that size is possibly non-trivial.
Cloud databases have been pricey for a while.
The reality most databases are tiny as shit and most apps can tolerate the massive latency that the cloud provider dbs offer.
It is why it is sorta funny we are rediscovering non network attached storage is faster.
> $4529/month... can indeed just hire someone full-time
That's $54,348/year, not including the cost of benefits, not including stock compensation. Let's say you reserve 20% for benefits and that comes out to $43,478.40 in salary.
Besides the benefit of not needing the management / communication overhead of hiring somebody, do you know any DBAs willing to take a full-time job for $43,478.40 in salary?
Missed the 'extrapolating' part -- for 3x that, absolutely.
> $50 bucks gets you an EIGHTH of a vCPU, 1GB RAM, and 10GB SSD???
Apparently there are people who find this offering compelling. The lack of value is quite stunning to me.
how does this compare to an RDS offering?
This will be faster than an equivalent RDS instance and will handle more of the operational lifecycle around failover and high-availability with less downtime than RDS.
So what happens if you get a nvme failure? Is there automatic failover and restore?
How does cross data center nodes work?
These are all three-node clusters with PlanetScale's management handling backup, restore, failover, and replication.
Let's gooooo! Incredible deal for indiehackers.