Re:The first release of OpenVox, the OSS implementation of P...
by everforward on Wednesday January 22, 2025 @02:20PM
(#42796462)
The head node has to do all the processing of what changes to make, for each server it runs against. Pull based systems have the destination server do that, making them scale horizontally natively.
Presuming there's no need for orchestration between nodes (which pull-based configs make very hard/impossible), pull-based systems parallelize perfectly where push-based systems cannot parallelize.
SSH is also very slow as far as these things go. At some point the head node just gets saturated on doing SSH encryption or the NIC.
Re:The first release of OpenVox, the OSS implementation of P...
by citrin_ru on Saturday January 25, 2025 @04:13AM
(#42820502)
Puppet servers can be scaled horizontally - create N puppet servers and then allow a client (an agent) to select a random server (either using a load balancer or just a DNS RR). But even a single server on modern hardware with appropriate tuning should be able to serve thousands nodes.
I would have expected the opposite. Pushing changes as required seems easier when the head node is the only one with smarts deciding on what happens next. Unless Ansible has some hard cap on how many servers it can direct in parallel?