Update MIGRATIONS.md #3975
Update MIGRATIONS.md #3975
Conversation
Add description of the correct query node value
Codecov ReportPatch and project coverage have no change.
Additional details and impacted files@@ Coverage Diff @@
## master #3975 +/- ##
=======================================
Coverage 90.21% 90.21%
=======================================
Files 750 750
Lines 36397 36397
Branches 188 188
=======================================
Hits 32835 32835
Misses 3533 3533
Partials 29 29 Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here. ☔ View full report in Codecov by Sentry. |
| @@ -48,6 +48,8 @@ For each cluster, `single_node` can be set to either True or False. If False, `c | |||
|
|
|||
| For a single node cluster, do not change the default configuration. | |||
|
|
|||
| For multi node clusters (`single_node=False`), the `host` and `port` of the cluster mappings must point to a node that will have the distributed tables. Typically, this is a node that belongs to the `distributed_cluster_name` cluster in clickhouse. This node is used as the query node. In our production environment, the `host` value is a loadbalancer proxy over the nodes is the the `distributed_cluster_name` | |||
There was a problem hiding this comment.
the
hostvalue is a loadbalancer proxy over the nodes is the thedistributed_cluster_name
I'm not sure what this means.
There was a problem hiding this comment.
Typically, this is a node that belongs to the distributed_cluster_name cluster in clickhouse.
I question whether this is actually typical. At least it seems not at Sentry. Afaik we often have no distributed cluster at all (local, test, small self hosted envs) or multiple distributed tables with a loadbalancer in front, but not only one node in the distributed cluster like you described
There was a problem hiding this comment.
For cases where there's no distributed cluster, single_node = True and everything is in one host. What I was trying to communicate is that the query node has to have access to the dist tables in the non single node case because we don't make that clear anywhere.
There was a problem hiding this comment.
Does this make anything clearer?
For multi node clusters (
single_node=False), thehostandportof the cluster mappings must point to a node that will have the distributed tables. If distributed tables are separated onto their own node (query node) and you have more than one of these query nodes, you should have a ClickHouse cluster defined for those nodes. That cluster name is used for thedistributed_cluster_name. Thehostvalue is this scenario could either be the query node directly or a load balancer proxy in front of the query nodes.
Add description of the correct query node value