Hetzner Cloud Documentation
In this technical retrospective (2024), @2-3-7-43-1807 shares practical architectural lessons learned from running, scaling, or migrating production software under lean constraints.
Core Principles for Independent Builders:
Defaulting to simpler architectures (such as monolithic frameworks, SQLite, or managed VPS) prevents early engineering bankruptcy.
Evaluating true compute and egress costs reveals that high cloud bills can often be slashed by 80%+ through judicious re-platforming.
Code you write at 2 AM as a solo engineer must be easy to debug six months later without complex microservice dependencies.
Community Debate Themes:
HN engineers debated the tradeoffs between cloud convenience and raw server economics, database indexing strategies, and resisting premature architectural complexity.
CATEGORY Frugal Engineering →
HISTORICAL ERA Post-ZIRP & AI (2023–2026)
COMMUNITY SCORE 20 points
TOTAL DISCUSSIONS 5 comments
It seems they have upgraded the specs on the block storage volumes, I have a support e-mail from December 22nd 2022 saying "You can expect a throughput of 250MB/s on average, and up to 2500 IOPS for the cloud blockstorage volumes." and that it's constant with size (which still seems to be the case). They write "It is not possible to RAID them, they are already in a RAID on the volume host." which is a mystery to me what it means exactly.
I guess they could write, if you must assume the volumes are served by the same cluster or not - e.g. for the case where you need more performance or storage space as one big filesystem/ file (and want to do a software RAID/ use ZFS/ btrfs/ bcachefs etc. to achieve that).
There is no note about the performance/ limits of the NVMe storage that the instance resides on - they generally seem to be much faster.
Also in 2021, I got a clarification by Markus Schade that:
"Instance disks are included in every server type and defined in the same way as physical drives (80 GB = 80000 MiB) rounded to the nearest 4MB. As we are offering dedicated as well as virtual servers, we use GB on all our pages for included drive sizes to reflect this. Volumes are different because it is a pay as you product. As such they are sized and invoiced on a per GiB basis."
Volumes seem to be base 2.
Unfortunately, Hetzner also have no log about internal migrations of VMs or relevant adjustments to the underlying host that would be visible to the user. What seems to be some kind of infrastructure adjust seems to have had a negative influence on Java GC at least once. It would be easier to debug these things if you didn't have to ask support first and make yourself look like a fool at times.
Is there anything noteworthy about this documentation?