Upgrade Procedure (0.16.0) -->(0.17.0)

Hi all we are currently running an imply cluster, with 3 masters & query nodes, and 12 data nodes.

I see the following upgrade instructions, which are for zero downtime.

https://druid.apache.org/docs/latest/operations/rolling-updates.html

Which calls for this order:

Historical
Overlord
MiddleManager
Broker
Coordinator

My concern is since we are running the supervisor scripts it is difficult to only upgrade the historicals, followed by overlords etc…

seeing that we have so many nodes, and processes, I was wondering if there is a simpler way to upgrade, perhaps with with downtime?

i.e. could we not stop all our existing nodes, then deploy the new versions to all nodes, and start them up?

or what if we stopped all ingestion processes, then upgraded all the historical nodes 1 at time, then the master nodes, then the query nodes?

would this be ‘safe’?

Thanx

Dan

Hi Daniel. I have not performed a 16.0 to 17.0 upgrade specifically (I have performed a rolling upgrade to version 15.0) but in regards to your question:

“could we not stop all our existing nodes, then deploy the new versions to all nodes, and start them up?”

Definitely yes you can do this. You could even install and configure everything side by side with your running version. Stop all services on the 16.0 and start the 17.0

As long as the 17.0 is talking to the same meta data db and deep storage you will be A-OKAY.