Open samay-sharma opened 8 years ago
@samay-sharma, @begriffs, @sumedhpathak -- I wanted to add a note to this issue.
We talked about this issue in our product planning session yesterday. We have three sections in our docs, where we talk about real-time and task-tracker executors:
One challenge is how much we'd like to communicate router planner / executor to users. Do we for example want to go into router planner? Ideally, the simpler we can communicate this executor type's benefits to users, the better.
+1 for this.
I think query processing documentation is far behind the current product, missing the multi-tenant model (at all?).
For instance: The planning process for key-value lookups on the distribution column or modification queries is slightly different as they hit exactly one shard.
is misleading. We probably should not mention key-value
lookups, which connotates other systems and is primitive than what router planner can handle. Similarly, single shard
is a subset of router planner's capability.
(I remember @samay-sharma writing this documentation years back in Istanbul. Seems its time to update it 😄 )
We now have bigger sections for each executor in https://docs.citusdata.com/en/v7.3/performance/query_processing.html#distributed-query-executor
Although that "key-value" sentence still survives.
Should we close this issue?
In Citus 5.0, we used to have a section on Router Executor in our Citus Query Processing section. Currently, the page doesn't have that section since we removed it with 5.1. However, since the multi-tenant use case uses it heavily, we should probably re-add the section.