Web / Apache Solr Interview questions
How does Solr handle distributed search across shards?
When a collection has multiple shards, a query sent to any single replica must still search all shards to produce a complete result, since each shard only holds part of the data. Solr coordinates this transparently using a scatter-gather approach.
sequenceDiagram
participant Client
participant R as Replica (coordinator)
participant S1 as Shard 1 replica
participant S2 as Shard 2 replica
Client->>R: query request
R->>S1: forward query
R->>S2: forward query
S1-->>R: top doc IDs + scores
S2-->>R: top doc IDs + scores
R->>R: merge and re-rank
R->>S1: fetch full docs for final IDs
R->>S2: fetch full docs for final IDs
R-->>Client: merged response
The replica that received the request acts as a coordinator. It first asks every shard for just doc IDs and sort/score values, merges and sorts that combined list, then makes a second round of requests to fetch the full stored fields only for the final page of results, avoiding the cost of transferring full documents from shards whose results won't even be returned.
What does a coordinating replica request from each shard in the first round?
Why does Solr fetch full documents only in a second round?
More Related questions...
