search

pelias-deprecated / kubernetes

Tools for running Pelias services on kubernetes
MIT License
54 stars 47 forks source link

Why using NFS Persistent Volumes for the importers? #109

Closed paulopez78 closed 4 years ago

paulopez78 commented 4 years ago

Hi Pelias Team,

First of all thanks for creating and mantaining that awesome project.

We are evaluating the possibility to use pelias in production running in GKE cluster, as I'm pretty new to the project I don't understand the need to use NFS Persistent Volumes for running the importers. What I though is that importers were independent so there is no need for sharing data between the different jobs running the import. Our idea was to use persistent volumes using the plugin of the cloud provider and not nfs, in that case GCP. Would it be possible?

Regards, Pau.

acondrat commented 4 years ago

We do not share data between importers and use emptyDir volumes for all of them, except whosonfirst. The latter requires a lot of disk space and inodes so we use a 1000G pd-ssd volume for it.

paulopez78 commented 4 years ago

Thanks for you quick reply,

I will do exactly what you suggest but this is confusing because latest version of the chart is showing a Persistent Volume (not emptyDir) mounted in all the exporters, for example here:

https://github.com/pelias/kubernetes/blob/master/build/templates/openaddresses-import-job.tpl#L63 https://github.com/pelias/kubernetes/blob/master/build/templates/openstreetmap-import-job.tpl#L63

acondrat commented 4 years ago

Yes, this is true. We created our own helm chart because of that.