deploy/gcp: support zonal and multi-zonal cluster #809
Conversation
|
@gregwebs @tennix @jlerche @aylei Known issues:
|
| description = "Operator helm values" | ||
| type = string | ||
| default = "" | ||
| } |
There was a problem hiding this comment.
declare it in root module to make it configurable
| description = "Kubernetes version to use for the GKE cluster" | ||
| type = string | ||
| default = "latest" | ||
| } |
There was a problem hiding this comment.
declare it in root module to make it configurable
|
|
||
| variable "tidb_replica_count" { | ||
| default = 3 | ||
| } |
There was a problem hiding this comment.
these three variables are not used anymore
| variable "gcp_region" { | ||
| description = "The GCP region" | ||
| } | ||
|
|
cofyc
force-pushed
the
improve_gcp
branch
2 times, most recently
from
145f378
to
e8ed70d
Compare
| } | ||
|
|
||
| variable "node_locations" { | ||
| description = "The list of zones in which the cluster's nodes should be located. These must be in the same region as the cluster zone for zonal clusters, or in the region of a regional cluster. In a multi-zonal cluster, the number of nodes specified in initial_node_count is created in all specified zones as well as the primary zone. If specified for a regional cluster, nodes will be created in only these zones." |
There was a problem hiding this comment.
What is the purpose of this for a single-zone cluster? Doesn't the above location variable already specify this properly?
There was a problem hiding this comment.
oh, I see, there are zonal clusters that are multi-zone clusters. I wonder why not just use the regional cluster though?
There was a problem hiding this comment.
yes, if you need to run a multi-zone cluster, it's better to use a regional cluster. However, if users want a small cluster for testing or for low network latency between work nodes, they can use a zonal cluster and run nodes in a single zone.
node_locations can also be used to choose node locations for the regional cluster too (at least two locations for a regional cluster).
Currently, in GCP, some types of machines are only available in a few zones, e.g. in region us-central1 c2-standard machines are only available in us-central1-b and us-central1-c zones.
If users want to use these compute-optimized machines, they must specify node_locations to deploy nodes in these zones.
cofyc
changed the title
|
/run-e2e-in-kind |
|
/run-e2e-in-kind |
|
@gregwebs @aylei |
What problem does this PR solve?
What is changed and how does it work?
Check List
Tests
Side effects
Related changes
Does this PR introduce a user-facing change?: