#general
@yuanbenson1772: @yuanbenson1772 has joined the channel
@sunhee.bigdata: Hi, I checked Authorization in broker level. ```pinot.broker.access.control.principals.<user>.tables=test_table ``` It seems that restarting broker is required whenever adding a privileged table. Do you restart every time you grant table privileges in the production environment ? Or is there any other way?
@mayanks: Currently it is like that. In prod deployments, folks usually have their own ACL system outside of Pinot. Please file a GH issue and paste here so we can track the request
#random
@yuanbenson1772: @yuanbenson1772 has joined the channel
#troubleshooting
@yuanbenson1772: @yuanbenson1772 has joined the channel
#getting-started
@arekchmura: Hi, I have some questions about the internals of the StarTree index. I will put them below, each in a separate thread.
@arekchmura: 1. Can StarTree index be used only with offline tables?
@mayanks: Yes, most definitely.
@npawar: Afaik, you can set startree index for realtime table too. It will only apply to the completed segments
@mayanks: Oh I missed the “only” part. Yes consuming segments don’t have star tree index
@arekchmura: Thank you
@arekchmura: 2. If I have multiple segments on different nodes that belong to a single table, does every segment contains a separate StarTree index? Or is there always a single StarTree index for every table, no matter how many segments it has?
@mayanks: All indexes including starTree are per segment
@arekchmura: 3. If there is a separate StarTree index for every segment, how do Brokers aggregate the results from them? For example, I have a table containing movies ratings. One segment holds ratings from August and the second one from September. I configured my StarTree to calculate the average for the rating column. The first segment has a pre-computed value for August. Let's assume it's 4.7. For the second segment, it's 4.95. Now imagine that I execute a query to get the average rating from those two months (August and September). How can the Broker merge the results from these two segments?
@mayanks: Average is stored as sum + count internally, to keep it additive.
@arekchmura: It makes a lot of sense, thanks
@arekchmura: 4. Does the order of dimensions that we provide when configuring the StarTree index matter when it comes to the performance or storage? If so, are there some general heuristics to help us to choose which attributes should go first or last?
@mayanks: Depends on the query set, but generically speaking higher cardinality first.
@yuanbenson1772: @yuanbenson1772 has joined the channel
--------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@pinot.apache.org For additional commands, e-mail: dev-h...@pinot.apache.org
