#Serverless is still misunderstood.
It’s not about servers or lack of them.
It’s an operational construct 🧵
It’s not about servers or lack of them.
It’s an operational construct 🧵
#Serverless is NOT about:
1/Being cheaper - often you pay a premium for serverless💰but what you gain are fewer operational headaches
1/Being cheaper - often you pay a premium for serverless💰but what you gain are fewer operational headaches
It’s NOT about:
2/Being stateless - it’s more accurate to say that any state needs to be persisted externally to the serverless compute (e.g., when using FaaS) but if some service requires persistent state, it doesn't disqualify it from still being #Serverless
2/Being stateless - it’s more accurate to say that any state needs to be persisted externally to the serverless compute (e.g., when using FaaS) but if some service requires persistent state, it doesn't disqualify it from still being #Serverless
The best examples of being stateful AND #Serverless are:
- serverless data warehouse / query tools (BigQuery, @starburstdata)
- serverless data stores (@neondatabase, @fauna, DynamoDB, Aurora SLS, ...)
- serverless data warehouse / query tools (BigQuery, @starburstdata)
- serverless data stores (@neondatabase, @fauna, DynamoDB, Aurora SLS, ...)
Instead, #Serverless is about:
1/Responsibility for maintenance: who takes care of operations - you (not serverless) or the service provider (serverless)
2/Pricing: do you pay per node per hour (not serverless) or usage/consumption-based (serverless)
1/Responsibility for maintenance: who takes care of operations - you (not serverless) or the service provider (serverless)
2/Pricing: do you pay per node per hour (not serverless) or usage/consumption-based (serverless)
It's about:
3/ Operational risk: if something breaks, is it on you (not serverless) or on the service provider (#Serverless)
3/ Operational risk: if something breaks, is it on you (not serverless) or on the service provider (#Serverless)
It's about:
4/ ⏳Time & resource management: do you monitor and optimize capacity utilization of your compute (not serverless), or do you let the service provider ensure you don’t pay for idle capacity (#Serverless)
4/ ⏳Time & resource management: do you monitor and optimize capacity utilization of your compute (not serverless), or do you let the service provider ensure you don’t pay for idle capacity (#Serverless)
What #Serverless means in practice (1/5):
1. Compute is abstracted away: servers still exist, but you don’t have to provision them
1. Compute is abstracted away: servers still exist, but you don’t have to provision them
What #Serverless means in practice (2/5):
2. Maintenance: you don’t have to worry about upgrades and patches, the service provider takes care of it for you🧘♀️
2. Maintenance: you don’t have to worry about upgrades and patches, the service provider takes care of it for you🧘♀️
What #Serverless means in practice (3/5):
3. Resilience: service provider ensures high availability and fault tolerance👩🚒
3. Resilience: service provider ensures high availability and fault tolerance👩🚒
What #Serverless means in practice (4/5):
4. Elasticity: compute and/or storage scales automatically 📈
4. Elasticity: compute and/or storage scales automatically 📈
What #Serverless means in practice (5/5):
5. Consumption-based pricing 💰
- usage-based (per millisecond, second, minute, month)⏳
- resource-consumption-based (per CPU/memory units consumed)💿
rather than pricing per instance type per node per hour💿/⏳
5. Consumption-based pricing 💰
- usage-based (per millisecond, second, minute, month)⏳
- resource-consumption-based (per CPU/memory units consumed)💿
rather than pricing per instance type per node per hour💿/⏳
To determine if a service is serverless: check its pricing.
If they charge:
1/per instance type per hour: it's on-demand but NOT serverless ❌
2/per (milli)second of usage: #Serverless ☑️
3/a flat fee (e.g., per month) but handles Ops at any scale: it *might* be serverless
If they charge:
1/per instance type per hour: it's on-demand but NOT serverless ❌
2/per (milli)second of usage: #Serverless ☑️
3/a flat fee (e.g., per month) but handles Ops at any scale: it *might* be serverless
To read more: serverless.com/blog/why-many-…
And this thorough #Serverless post: martinfowler.com/articles/serve…
• • •
Missing some Tweet in this thread? You can try to
force a refresh
