The relforge cluster should be accessible through a single entry point. LVS seems the obvious choice as it mirror production setup, but other solutions could be used (nginx, haproxy, ...).
Description
Description
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Resolved | • Deskana | T139575 EPIC: Plan to enable BM25 on fulltext search | |||
| Resolved | Gehel | T137256 Setup two node elasticsearch cluster on relforge1001-1002 | |||
| Declined | Gehel | T142098 Setup load balancing for elasticsearch service on relforge servers |
Event Timeline
Comment Actions
Reading documentation, it seems that we only have the production LVS servers. There should be some isolation in place between those LVS servers and the labs support subnet, so we might not be able to configure LVS for relforge in the current context.
@BBlack could you confirm this? I'm not entirely clear on our network configuration and a bit lost in the different config files.
Comment Actions
The relforge cluster is not H/A, with 2 nodes we can only have a single master if we want to protect again a split-brain. This means that we can probably get away with not having load balancing and let elasticsearch routing take care of splitting the load across the cluster. @dcausse, @EBernhardson your input would be welcomed.
Comment Actions
@Gehel it's perfectly OK to run without H/A, this cluster is not meant to serve production traffic but only optimizations/evaluations of cirrus settings so I don't think LVS is very important here.
Comment Actions
Since we don't need load balancing, let's close this. If the need comes at some point, we will address it then.