Hacker News Reader

Comment by djha-skin | Hacker News Reader

Google could have (and many expected them to) eat Amazon's lunch for cloud compute back when it was new. However, Amazon actually had infrastructure around customer support while Google famously really doesn't. From what I gather Google cloud is technologically better, but no one wants to use a cloud from someone who doesn't do customer support as well.

View on HN

I've got a tale of two clouds. I complained about an issue with AWS on Twitter and two hours later was on a call with the EKS product manager who connected me with someone on their team who documented the issue, gave me a workaround, and then pushed out a fix to everyone a few days later. It was super impressive!

I complained about a separate issue with GCP. The product manager found my tweet and told me it was my fault for using a service marked as "preview". He then told me I should have used a different service, which was also marked as "preview". They were rude and defensive and made no attempt to resolve my issue. We decided to just not use GCP for the project as a result.

Neither of these are isolated stories. I have multiple examples of GCP staff just being super rude for no real reason, while AWS regularly goes above and beyond with customer service.

This matches my experience. GCP was like IBM a decade earlier: they assumed the name alone would close the deal and we didn’t feel like paying more for the privilege of doing their jobs for them.

I've experienced both:

One time two very competent guys shows up on site and let me reproduce the issue, agrees it is a serious bug and promises to take care of it.

Next time I get connected with what seems to be two part time students or something who doesn't seem to have a clue about the product.

Then in a third case some technical architect or something looks into my stack overflow question, verifies there is a known bug and it will be fixed in next monthly rollout, but I never saw a fix.

What's better is when their community forum "expert" redirects you to the first Stackoverflow link on Google - even though it is a band-aid solution that doesn't even solve your problem.

I had a similar experience, though it was years ago. Found a minor bug in Elastic Beanstalk and opened a ticket. They confirmed the bug and rolled out a fix in the afternoon. I was very impressed.

I've had very similar experiences with Google products. Their "support" for many products tends to be primarily community forums. And on those forums, "if you don't like it then don't use it" seems to be the standard type of answer.

> From what I gather Google cloud is technologically better, but no one wants to use a cloud from someone who doesn't do customer support as well

I have a hard time using any of google's stuff - cloud, whatever. Have had to deal with 'google search console' recently. "Give us your sitemap!". OK... "Could not fetch". The explanation for "could not fetch" might be "couldn't fetch" or "hasn't fetched yet" or "read but not processed" or "error reading" or.. anything else. On March 9, I had a sitemap entry listed as "couldn't fetch" updated to "last fetched date" of March 10 (9am UTC-4, so not even close to march 10 UTC time). It's just.... buggy. Colleagues currently moving stuff to google cloud (by edict of cto) are encountering bug after bug after slowness after flakiness. Google "support", to the extent they answer, says "we don't know".

Might it have been 'technically better' on day one? perhaps. But if it's buggy/flaky to deal with, and they have no support, not sure how you'd even verify the "technically better" part. How would you trust any numbers you see reported in their own tools?

Had to use GCP stuff about 6 years ago. It was flaky/slow and relatively unsupported then. Watching colleagues go through stuff today, in 2023, it seems no better.

I use both in my day to day work. My takes right now:

Roles/Permissions in GCP is just done better. The whole system of having to switch roles to be able to see stuff across multiple accounts in AWS is opaque and confusing. TrustPolicies are powerful but feel unnecessarily complex for almost every use case. Google has its own warts around permissioning (doing things often requires an opaque role that you have to look up, the error message is often unhelpful). However, it’s better than unraveling the series of permissions needed to, for instance, have an app pull from an S3 bucket in AWS.

AWS sucks at naming things. Everything is some nonsensical acronym that only makes sense to salespeople at Amazon. When you wonder what Google’s load balancer product is called, you look it up and it’s perhaps unsurprisingly called Elastic Load Balancer.

Another plus for Google: Having IAP as a first class citizen is a nice way to avoid having to set up bastions etc when prototyping.

On the other hand, we just spun up a Karpenter instance in EKS, and according to my colleague it’s much better than Google’s Autopilot product.

Also there is a whole industry around getting support from Google, lol. We use DoIT at my place of work, which is a company whose entire business is to pool together customer accounts for volume pricing and white glove support from Google. Interestingly, the cost savings from volume pricing are so significant there’s no cost to end users for using DoIT.

Elastic Load Balancer is in fact an AWS product. I believe GCP's equivalent is called Cloud Load Balancing.

Your point is valid and something I've felt many times as well (Route 53 means DNS somehow???), you just happened to choose literally the only AWS product that is well named.

DNS servers listen on port 53 :)

True. But if you asked me what Route 53 does I wouldn’t be able to tell you without checking. Cloud DNS, OTOH, is a pretty descriptive product name. I’m not saying you disagree, just reiterating the point that AWS naming is just a little too silly and clever for its own good ;)

I interviewed with Google Cloud a few years back and they straight up told me they were behind both AWS & Azure as a product and were lacking in user support. They said they were trying to build out an org to address it.

> From what I gather Google cloud is technologically better

It's a shitshow. Their design is ridiculous, overcomplicated and opinionated. It's clear they are run by engineers because no sane human would choose to use GCP knowing how bizarre and painful it is.

Creating an IAM role isn't strongly consistent. You have to build your own polling logic to wait for it to be done to attach a policy to it. Even using their own Terraform provider. If I worked there I'd be embarrassed.

As someone now in a position to make such decisions on a larger scale, GCP isn't even on my radar for this reason. I've had enough experiences with Google's lack of support when needed that it would take years of turnaround to undo that black mark.

AWS and Amazon as a whole has always had stellar support. And they'll continue to receive my business in both as long as that holds true, even though we rarely use it.

Google doesn't dogfood its own cloud. None of the popular services like Search, Gmail, or Drive actually use GCP.

Of course they don’t use layer upon any of their own services.

Otherwise what would they do when they inevitably get discontinued?

This is true and false. Google services, including GCP, run on its internal cloud. GCP is just a product on top.

Not so long ago (a few years), Amazon retail was not using AWS either (I think they do, at least partially, now).

Well, the problem with Google is that AFAIK they have no plans to onboard their important services like Search onto GCP. I believe a few years ago they have tried but found the GCP tools too lackluster for internal use.

Reply on news.ycombinator.com