Being multi-region cluster here is very likely not the issue. The issue here is likely due to the fact that our query plan for Statement / Transaction Page is not optimal. Especially since we are using the virtual table as a proxy to query the system table.

Effectively, here's our query plan:

                       info
--------------------------------------------------
  distribution: local
  vectorized: true

  • filter
  │ filter: (aggregated_ts > '1970-01-01 00:00:00+00:00') AND (aggregated_ts < '1970-07-05 04:26:40+00:00')
  │ 
  └── • sort
      │ order: +column14,+transaction_fingerprint_id
      │
      └── • render
          │
          └── • scan
                missing stats
                table: statement_statistics@primary
                spans: FULL SCAN
(12 rows)

The sort on top of the full table scan is going to be very expensive. This is due to the massive amount of data coming from the full table scan and forces the SQL sorter to go into the disk spilling mode since we are definitely blowing up the in-memory sorter in the SQL Engine.

We need to pass down the predicate of aggregated_ts down into the system table in our query plan here, either by making the virtual table smarter, or by switching away from using virtual table and build our own interface here.

Statements page gives error in multi-region cluster #74680

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions