Look into multi instance installations

[poikilotherm]

Driving a Dataverse multi-instance installation has some edges and needs to be examined carefully. Obviously, scaling with Kubernetes is super-easy.

http://guides.dataverse.org/en/latest/installation/advanced.html#id1

[pdurbin]

I would suggest watching https://github.com/BU-NU-CLOUD-SP18/Dataverse-Scaling#our-project-video which I mentioned in a comment at IQSS/dataverse#4040 (comment)

This was the final video students at BU made after their efforts to scale Dataverse across multiple Glassfish and PostgreSQL servers.

We held weekly meetings during the class and recorded them and posted notes, if those are of interest as well.

[poikilotherm]

While looking into salvation for #65 I thought about how to sync the files from the docroot to other pods in case of a replication with a StatefulSet.

It looks like the easiest way to go is via a ReadWriteMany volume, backed by an NFS server (or similar). Ganesha NFSv4 seems to be worth a try, as it can be easily containerized in the K8s cluster. Running a kernel NFS server needs some hacking and priviliges.

[pdurbin]

@poikilotherm you should do whatever hacking you need to do to stay unblocked but I'm wondering if upstream Dataverse should evolve or change in some way to make it easier to run multiple (Glassfish) web servers. Should the logos used in the header of dataverses be stored in the the database, for example. (Heads up that logos in the footer are now supported as well, thanks to IQSS/dataverse#6219 .) I believe Harvard Dataverse uses some rsync scripts to keep the logos in sync across two web servers but I'm not sure how well that scales. What I'm trying to say is that I'm definitely open to ideas. 😄

[pdurbin]

Please see also IQSS/dataverse#6491