[DocDB] Analyze is running faster right after restore than subsequent attempts

[ttyusupov]

Jira Link: DB-11991

Description

I've loaded a data into fresh universe using https://github.com/yugabyte/benchbase/blob/retention_pd/config/yugabyte/regression_pipelines/retention_impact/yugabyte/retention_impact.yaml and then created a backup. Running analyze nets_1 on the loaded data takes about 20 minutes and "Disk Bytes / Node" read metric is about 17MB/sec, CPU load is less than 2% (and close to the background CPU load without analyze). But if I restore the data from this backup, the first time I am running analyze nets_1 it takes about 2.5 minutes and the "Disk Bytes / Node" read metric is up to ~160MB/sec with CPU load around 5-6%. Next executions of analyze nets_1 take around 20 minutes with 17 MB/node/sec disk read load and ~2% CPU load.

Issue Type

kind/bug

Warning: Please confirm that this issue does not contain any sensitive information

• [X] I confirm this issue does not contain any sensitive information.