I think introducing new tech is fine as long as it solves a new problem for your business. If you haven’t yet used a cache, but have a need for one, introducing Redis or Memcached makes sense. If you have a new need for pub/sub workflows, introducing Kafka or a webhook dispatcher makes sense. Same with Elasticsearch or whatever if you’re building out your first search feature, etc.
What rarely makes sense is introducing new tech that heavily overlaps with existing tech. You don’t wanna end up on MySQL and Postgres and Mongo. Python and PHP and Go. Django and Flask. EC2 autoscaling and Mesos and Kubernetes. Even worse, you definitely don’t wanna start building your own versions of tech that there’s a good off the shelf solution for. If someone tries to build their own UI framework, ORM, service discovery system, etc., kill that IMMEDIATELY.
Introducing new tech that overlaps with existing tech should only be done if you’re going to migrate OFF the existing tech. Obviously this is often a major effort, so the new tech has to be a HUGE win.
My current company was pretty immature for a long time in terms of tech decisions, and the costs are massive. For service-to-service communication we use REST, gRPC and this terrible in-house re-implementation of HTTP over ZeroMQ, complete with hand written clients and server frameworks in multiple languages. We’ve been trying to migrate off that for ages, at massive cost. We’re almost done, but even then will still be stuck with both HTTP and gRPC, when we only need one. Similarly, we use PHP, Scala, Go and Python on the backend. With multiple server frameworks for each. We use MongoDB and MySQL. Redis and Memcached.
The cost of all this is massive. Developers have to learn SO much to be productive in our stack(s). Every time we want to write some middleware, tooling, etc., we have to write like 15 versions of it, because we have about 15 unique combos of backend language/network protocol/framework. Losing a few immature devs who care more about using shiny new tech than solving business problems is a very, very small cost to pay to avoid this situation.
I think introducing new tech is fine as long as it solves a new problem for your business. If you haven’t yet used a cache, but have a need for one, introducing Redis or Memcached makes sense. If you have a new need for pub/sub workflows, introducing Kafka or a webhook dispatcher makes sense. Same with Elasticsearch or whatever if you’re building out your first search feature, etc.
What rarely makes sense is introducing new tech that heavily overlaps with existing tech. You don’t wanna end up on MySQL and Postgres and Mongo. Python and PHP and Go. Django and Flask. EC2 autoscaling and Mesos and Kubernetes. Even worse, you definitely don’t wanna start building your own versions of tech that there’s a good off the shelf solution for. If someone tries to build their own UI framework, ORM, service discovery system, etc., kill that IMMEDIATELY.
Introducing new tech that overlaps with existing tech should only be done if you’re going to migrate OFF the existing tech. Obviously this is often a major effort, so the new tech has to be a HUGE win.
My current company was pretty immature for a long time in terms of tech decisions, and the costs are massive. For service-to-service communication we use REST, gRPC and this terrible in-house re-implementation of HTTP over ZeroMQ, complete with hand written clients and server frameworks in multiple languages. We’ve been trying to migrate off that for ages, at massive cost. We’re almost done, but even then will still be stuck with both HTTP and gRPC, when we only need one. Similarly, we use PHP, Scala, Go and Python on the backend. With multiple server frameworks for each. We use MongoDB and MySQL. Redis and Memcached.
The cost of all this is massive. Developers have to learn SO much to be productive in our stack(s). Every time we want to write some middleware, tooling, etc., we have to write like 15 versions of it, because we have about 15 unique combos of backend language/network protocol/framework. Losing a few immature devs who care more about using shiny new tech than solving business problems is a very, very small cost to pay to avoid this situation.