we hit the same issue with this topic. Our SLB system detect and determined
whether a broker is alive or not only by it's return HTTP Code.
Our Druid Cluster is quite huge, and a broker from start to ready to server
might have 10+ seconds time gap.
when our SLB system try call /druid/broker/v1/loadstatus
it might return
HTTP Code: 200
{
"inventoryInitialized": false
}
and assume that this broker could be served query, hens, some of the request
will fail according to this condition.
Proposal:
Have to look at the latest code of coordinator isLeader API
```
if (leading) {
return Response.ok(response).build();
} else {
return
Response.status(Response.Status.NOT_FOUND).entity(response).build();
}
```
i suggest that we can have a another broker http endpoint
/druid/broker/v1/isReady
if serverView is not inititalized, return 404, else return 200. and this will
make our SLB system work as expected.
@pdeva if you are with my proposal, i can do the pull request, and fix this
issue along with the documentation. FYI.
[ Full content available at:
https://github.com/apache/incubator-druid/issues/6172 ]
This message was relayed via gitbox.apache.org for [email protected]
