Gitlab 13.10 Released

I haven't had any older yamls break and I wouldn't say the yaml schema is worse off because of backwards compatibility. Not sure a version would change much atm.

I first started thinking hard about this when they recently cancelled their bronze tier subscription. Gitlab is clearly ducking out of a brawl with Github for the individual/consumer $5/month tier; that makes sense as there is no way Gitlab can win by taking on the incumbent on their own turf. Instead Gitlab seems to be shifting focus to targeting larger enterprise customers; those who are fine paying $20/mo or ideally $100/mo for a one-stop solution to the full SDLC. (The counterargument here would be that they _are_ still targeting the $5/mo customer, they are just trying to replace a $5/mo Github subscription plus a $10/mo CircleCI sub plus a $10/mo Jira sub etc. -- I'm not sure I see that end of the userbase being as amenable to bundling though).

The largest enterprise customers will keep asking for more boxes to be ticked, because it's usually easier to add features onto your existing solution than to stitch multiple solutions together, and because the more features/config options you have, the more complex configurations/requirements you can satisfy. However that means you get feature bloat, and pricing becomes more challenging; you need to charge more for "all the features" tier, but as you broaden the offering, fewer customers actually want to pay for everything. "What do we keep in the $100/mo tier?" is a challenging question to get right as the feature-set grows.

As you get into enterprise sales, you start to need more customization/unbundling. Before I moved back to Github I paid Gitlab $20/mo per engineer on my team and would never dream of jumping up to $100/mo, but would absolutely have paid more for a la carte access to certain features from the $100/mo ultimate tier. (For example I have no interest in their issue tracker, but I'd love to have been able to use their DevOps / Kubernetes tooling).

I believe this sort of a la carte pricing is less developer-friendly because you tend to need to talk to a sales person vs. just having the developer sign up, but then I don't believe that "developer first" is your sales strategy in enterprise; see Okta vs. Auth0 for a good example:

https://auth0.com/pricing/ https://www.okta.com/pricing/#customer-identity-products

Auth0 keeps it as simple as possible. Even within customer-identity (their competitor to Auth0) Okta has way more configuration for add-ons like MFA, SSO etc.

(I know @sytse / Gitlab folks post on here regularly so I'd love to hear their feedback on whether I'm completely off-base in how I'm thinking about this stuff!)

Are you no longer concerned because memory usage dropped significantly?

How's the overall latency?

I remember seeing some benchmarks where ActionCable was using like 1-2gb of memory for around a thousand active connections with 95% percentile latency in the multi-second range to broadcast a message. I do see AnyCable makes huge improvements here and I love the idea of it but also not too sure how backwards compatible it is with Hotwire / Turbo Streams.

I've always thought Rails was good / fast enough but the ActionCable stuff worries me a little because a single low end VPS could happily handle a very solid amount of traffic (tens of thousands of daily page views, etc.) but once you start factoring in websocket connections that changes everything. Suddenly pages that have completed the request / response cycle are still on the hook to keep a ws connection open even if they're not doing anything that causes a state change.