The term "platform" gets wildly thrown about, but from a backend/infra pov only one definition matters:
How much of your users' chaos are you inviting to live on *your* service for *you* to deal with, instead of on theirs?
How much of your users' chaos are you inviting to live on *your* service for *you* to deal with, instead of on theirs?
Users are inherently chaotic entities, but the more flexibility and creativity your service allows, the more potential exists for users to generate unknown-unknowns that you will have to unpack and debug. And the more "platformy" your service is.
Like, at Parse we ran infrastructure for a million mobile apps. They wrote their own queries -- we just had to absorb the consequences, and try to protect them and their neighbors on shared replica sets. We let them write their own js functions -- we had to run those too.
If you think debugging one system is hard, debugging a platform of equally flexible systems is at least an n^2+n problem. At *least*. (Depending on your cotenancy model, maybe even harder.)
And you can't track down their change sets in git repos you don't have access to.
And you can't track down their change sets in git repos you don't have access to.
Lots of people seem to throw around the word "platform" when the word "service" or "product" would do just as well.
In a true platform, you have to care as much about the subjective performance of every individual app or customer, as well as the overall availability & capacity.
In a true platform, you have to care as much about the subjective performance of every individual app or customer, as well as the overall availability & capacity.
It's the ultimate high cardinality problem. What's the user experience like for this customer who's only doing 4 req/sec? If your system is doing millions of req/sec, they may be completely down -- but it will never show up on any of your time series aggregates.
Living through this nightmare of a problem at Parse -- and eventually coming out the other side battered but intact -- is literally what made @cyen and i decide to start honeycomb.
Platforms have these problems first. It's a function of pure complexity and possible outcomes. The rest of you'll catch up fast tho.
(God, the number of horrendous hacks I have seen or done trying to deal with this... Auto-generated dashboards for your top 10 users anyone? ✋)
(God, the number of horrendous hacks I have seen or done trying to deal with this... Auto-generated dashboards for your top 10 users anyone? ✋)
