[IMPALA-6843] Responses to prioritizedLoad() requests should be returned directly and not via the statestore

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: Impala 2.11.0
• Component/s: Catalog
• Labels: catalog frontend latency performance
• Target Version: Product Backlog
• Epic Color: ghx-label-9
• Assignee: Unassigned
• Reporter: Dimitris Tsirogiannis
• Created: 12/Apr/18 20:53
• Updated: 07/Feb/23 11:09

Description

Currently, when a statement (e.g. SELECT) needs to access some unloaded tables, it issues a prioritizedLoad() request to the catalog. The catalog loads the table metadata but does not respond directly to the coordinator that issued the request. Instead, the metadata for the newly loaded tables are broadcast via the statestore. The problem with this approach is that the latency of the response may vary significantly and may depend on the latencies of other unrelated metadata operations (e.g. REFRESH) that happen to be in the same topic update.

The response to a prioritizedLoad() request should come directly to the issuing coordinator. Other coordinators will receive the metadata of the newly loaded table via the statestore. 

Issue Links

• blocks: IMPALA-6671 Metadata operations that modify a table blocks topic updates for other unrelated operations
• relates to: IMPALA-4045 Catalog cache update should not tied to statestore update frequency