Can we update the title to say Free trial? The minimum tier is still $40/month (though that is a huge improvement and makes this a reasonable option for startups).
Was really hoping for something that could scale to zero. Better than no change!
Yeah this article is just pure spam. Google Cloud has always had a max $300 free trial and the change here is now you do Spanner woooptie wooo. The offering is not at all like AWS free tier.
Yeah, we (Googler here) know, but it's not our post, so we can't edit it. :D Best we can do is hope people read the comments (weird thing to say on the internet).
You’re right, sorry about that. My non-english brain got it mixed up when I read the tweet here [1] as he previously talked about a free tier. Unfortunately I can‘t edit.
This is the next iteration to make Cloud Spanner more accessible for developers to start using a scalable relational database that values consistency AND availability.
If anyone wants to try out Spanner without cost, this is a great option. There's also a pretty cool starting experience in terms of in-console tutorial to set up a sample app on Spanner.
I'd love to get feedback on getting started with Spanner.
Disclaimer: as my username suggest, I work at Google, quite closely with the Cloud Spanner team.
Can you please as your PR and PMs to change the wording here so it’s so so misleading that the offering is NOT a free tier but rather a trial?
In general Ive had pretty ridiculous trouble with GCloud reps pushing their practices and mindesets on me, across several large contracts and companies. Your team dearly needs to work on basic communication. As Sundar has said publicly, Google as a whole needs to earn back user trust. As a corporate GCloud user (not my choice), you have obliterated it time and time again, and these sorts of fake free announcements really don’t help your case.
To be clear, Google Cloud always had a free $300 trial. The Spanner news is confusing at best, but most likely clickbait. If you really want Spanner in your default $300 free trial, you can just ask ops to enable / increase quota for it. The OP is a promotion advert disguised as a major service change and written as a PM. That goes to the core of the untrustworthy behavior that I've observed from GCloud reps over multiple contracts.
Not sure what you mean here. The Spanner trial is a completely separate thing from the $300 starting credits. So if you use the Spanner free trial, you aren't spending any part of the $300 credits during that trial period. You can also start the free Spanner trial regardless if you've spent the $300 starting credits in the past.
No but you wrote the article, and you have a lot of people who could email hn@ycombinator.com to get the advert corrected. Again, this is more about repeatedly poor customer service than accuracy of this specific article.
No extensions, no. Think of the interface more like a way for accessing Spanner features using Postgres, rather than using Postgres with a Spanner backend.
Technology behind spanner is incredible but unfortunately marketed horribly and positioned terribly. It was a solution built for large scale enterprise and no one else. It's not worth trying to sell this to developers. They're better off layering something else on top with a different brand and pricing model.
its a solution built for google where they need to track ad budget burn down globally. Most enterprises dont have this problem, or the latency/performance/uptime requirements of google.
Then you've got a google proprietary API for what should be similar to Aurora with PostgreSQL wire compatibility.
Add in Google's history of deprecating and non-backwards compatible changes to SDKs.
That makes you wonder, who exactly Spanner IS geared towards.
Spanner is used extensively in google for all sorts of things.
Many companies have the uptime requirements, a decent number have the performance requirements, and it's all going in one direction. There are a lot of things Google need and then all of a sudden 10 years later lots of others need them too.
Google employees don't pay for spanner. This is the big difference. You might have quota but you don't actually pay any money for it. No one on the outside world wants to get ripped off for database usage.
Neither, it's just not quite what folks expect. The postgres/spanner interface is about accessing SPANNER functionality via a PG interface. It's not about using Postgres with a Spanner back-end. That's all.
I think Spanner has always been targeted largely at mid-enterprise and bigger companies. Its potentially a VERY sticky product that may be tremendously difficult to offboard from. Small concessions like this 90 day trial doesn't really change that.
Especially with my understanding that a lot of GCP leadership is ex-Oracle, the whole vibe of Spanner seems to be a run at Oracle's marketshare.
I think spanner itself isn't really good at scaling down to be very small and cheap, so Google isn't happy to offer small instances for totally-free.
That's Google's mistake tho - if they had had a totally free option when they started, students would have built pet projects on it, and now they'd have a whole ecosystem of companies giving hundreds of millions of revenue for it.
They should have offered a totally free version (at a loss), knowing that in the future they could do the engineering work to make it cheap for them to run.
For example, stick 10,000 users into the same spanner database with some frontend to check no user can access rows owned by another. Then, when a user gets large and wants to pay for dedicated spanner, migrate them out of the shared instance.
I wish there was a true request per day free tier limit to play around with Spanner for small projects. Cloud Run and FiresStore are really awesome for hosting my hobby projects so far, at almost no cost.
If I'm reading that page right, the actual minimum cost (in us-west1) is $65.70/month, because a "node" is $657/month, but if you look elsewhere in the documentation, it says a single "instance" can be as small as 1/10 of a node.
I have no idea why they chose to structure the pricing model this way, but it makes the entry-level cost look a lot worse than it really is.
It might also be worth pointing out that 'node' does not reference underlying server.
In a PG or similar architecture, node might be used to specify how many "nodes" are in a replicated setup for high availability purposes.
In Spanner, "Node" is a measure of compute capacity (as is the finer-grain processing units) allowing you to scale up/down without impacting an application's workload.
All instances get high availability regardless of number of nodes or processing units, and the specifics of this are handled by the configuration.
That may have been true at one point, but now you can use Spanner closer to $60-70 a month factoring in storage (and even cheaper with committed use discounts) after upgrading from the free trial instance. https://cloud.google.com/blog/products/databases/use-spanner...
Same thing with BigQuery / Parquet: you can pay to load your data, and pay to have a giant BigQuery instance, but you can't get an equal with AWS Athena.
As an aside to any Google people here, not having error handling in your JavaScript is a bad idea - that blank page does not shout “you should trust your most sensitive data to us” even if I know those are separate teams.
Echoing other people, limiting it to 90 days / 10GB is an odd choice – if you need Spanner, you’re likely to bump into both of those limitations, especially if you’re not already a GCP user. Databases are a critical service and it takes time to evaluate one.
Yes, I’m just wondering who this announcement matters to - if you’re a tiny startup where the cost savings is what tips you over the edge, are you likely to be using enough data for it to matter? If you’re at a big enterprise, you probably can’t properly evaluate quickly and will need more than 10GB but are almost certainly going to have enough budget that the free trial won’t factor in to your decision. There are some people in the middle but I’m curious how many there actually are with these restrictions.
If you are a developer at a large company, you might be able to mess around with this and decide that it's worthwhile to push for a proper evaluation with a larger dataset.
I guess, but in my experience that either means you have enough budget to experiment anyway or are locked down to the point that you won’t be able to do anything which could cost money.
What I wish they had was a cap: allow me to tell the accountants that I can guarantee that a developer’s training will not leave the free tier.
The HN title is very misleading: It's a not actually a free tier, it's a time-limited, storage-capped free trial (separate from the normal $300 GCP free trial). Therefore, I don't think that there's a risk of it costing anything.
I guess I will say that I don't actually think that this is all that useful, personally, but I can see it as coming in handy in a very specific scenario.
I think the difference is that AlloyDB is basically PostgreSQL where they've modified how it's storing things to offer better performance via better integration with Google's infrastructure - like what Amazon did with AuroraDB. That means that AlloyDB is compatible with all the stuff you're used to using with PostgreSQL.
Spanner is a very different database. They've included a PostgreSQL query-language compatibility layer, but that isn't the same as being "100% compatible." You'll still be dealing with some Spanner concepts like not wanting to hot-spot your writes. Spanner does allow really scaling out and a truly distributed database while AlloyDB is an improvement via better integration with Google's infrastructure, but still mostly the same model.
Aurora/Alloy do have advantages by decoupling the storage from the server instances including performance and resiliency, but Spanner is really creating a horizontally scalable, distributed database - with some trade-offs.
Just from a glance at the AlloyDB documentation, it seems to be fundamentally a similar design to PostgreSQL. That is, you have a single primary instance that holds the entire dataset and handles all write traffic, plus some number of additional replicas within the system for failover and read-only queries.
Spanner has a very different implementation: your dataset is split into (possibly a large number of) shards, each of which is replicated across multiple regions. So you get better global availability and scalability, at the cost of more complicated performance characteristics when you have to do operations that involve multiple shards.
Do they mean free like legacy workspace accounts were free until all of the sudden they weren’t?
We won’t ever use a google product at my work because they burned me personally on that. Hope they like stepping over those dollars to pick up those pennies.
Was really hoping for something that could scale to zero. Better than no change!