[YSQL] Include the phrase disable_cost in the name bitmap_exceeded_work_mem_cost
#22388
Closed
1 task done
Assignees
Labels
area/ysql
Yugabyte SQL (YSQL)
kind/enhancement
This is an enhancement of an existing feature
priority/medium
Medium priority issue
Comments
timothy-e
added
area/ysql
yugabyte-ci
added
kind/enhancement
timothy-e
added a commit
that referenced
this issue
May 14, 2024
Summary: Because `bitmap_exceeded_work_mem_cost` doesn't contain the phrase `disable_cost`, it's easy to miss when someone try to find all the places that inflates the cost values, e.g. investigate into cost overflow, precision losses with complex plans, etc. Change the name to `bitmap_exceeded_work_mem_disable_cost`. Jira: DB-11285 Test Plan: Jenkins: compile only, build type: debug Reviewers: tnayak, amartsinchyk, mtakahara Reviewed By: mtakahara Subscribers: yql Differential Revision: https://phorge.dev.yugabyte.com/D35045
svarnau
pushed a commit
that referenced
this issue
May 25, 2024
Summary: Because `bitmap_exceeded_work_mem_cost` doesn't contain the phrase `disable_cost`, it's easy to miss when someone try to find all the places that inflates the cost values, e.g. investigate into cost overflow, precision losses with complex plans, etc. Change the name to `bitmap_exceeded_work_mem_disable_cost`. Jira: DB-11285 Test Plan: Jenkins: compile only, build type: debug Reviewers: tnayak, amartsinchyk, mtakahara Reviewed By: mtakahara Subscribers: yql Differential Revision: https://phorge.dev.yugabyte.com/D35045
svarnau
pushed a commit
that referenced
this issue
May 29, 2024
…of Bitmap Scan CBO Summary: Original commits: * D33861 / eac0c26 * D35044 / ca5b8e6 * D35045 / 49a8e47 Changes: * There are some differences in actual number of seqs and nexts in the table scan. Fix these by just adding `_IgnoreActualResults` to the tests in `testSeekNextEstimationBitmapScan` == Summary from D33861 == Implemented CBO for Bitmap Scans, using the same approach as PG: store a partially calculated index cost and use that to compute bitmap index access. If Bitmap Index Scans or Bitmap Ors are predicted to exceed work_mem, they'll have `disable_cost / 2` added to their cost. This discourages the planner from using them, but still costs them lower than other actually disabled scan types. Without this change, `/*+ BitmapScan(test) */ SELECT * FROM t1000000 ...` will use a sequential scan because it's cheaper, even though the sequential scan has been disabled by the hint. To better determine if a Bitmap Or will exceed work_mem, more work has been done to determine an estimate of how many rows a Bitmap Or will return. Also, a slight refactor was done: I put estimated seeks, nexts, width into a YbPlanInfo struct to reduce code duplication of printing / assigning those three values. Jira: DB-9574 == Summary from D35044 == [#22387] YSQL: Fix Bitmap Scan CBO tests 2338431 / D34646 introduced a flag `yb_enable_bitmapscan = false`, so YB bitmap scans are disabled by default, even when `enable_bitmapscan` is true. At the same time, eac0c26 / D33861 introduced CBO for bitmap scans. These tests passed before `yb_enable_bitmapscan` was added, but once it was added, they started failing because they stopped using bitmap scans. Fix the tests by setting `yb_enable_bitmapscan = true`. Jira: DB-11284 == Summary from D35045 == [##22388] YSQL: Change name of bitmap_exceeded_work_mem_cost Because `bitmap_exceeded_work_mem_cost` doesn't contain the phrase `disable_cost`, it's easy to miss when someone try to find all the places that inflates the cost values, e.g. investigate into cost overflow, precision losses with complex plans, etc. Change the name to `bitmap_exceeded_work_mem_disable_cost`. Jira: DB-11285 Test Plan: Latest: [[ https://jenkins.dev.yugabyte.com/job/optimizer/job/generate-report/231/artifact/taqo/report/20240501-185328/index_yb_vs_yb_cost-validation_.html | TAQO YB CBO with Bitmap Scans enabled / disabled ]] * Very few regressions in the default case - highest regression was 1.26 on query e57fd1caa18f59c4e40138d377ec2a01. I checked each regression above 1.1, and they were all run-to-run variances using the same execution plan. [[ https://jenkins.dev.yugabyte.com/job/optimizer/job/generate-report/224/artifact/taqo/report/20240418-134354/index_yb_vs_yb_cost-validation_bitmap_scan_yb_vs_pg.html | TAQO Run vs PG ]] [[ https://jenkins.dev.yugabyte.com/job/optimizer/job/generate-report/225/artifact/taqo/report/20240419-131907/index_yb_vs_yb_cost-validation_bitmap_scan_ybstatsonly_vs_pg.html | TAQO Run PG vs YB with only stats ]] ``` ./yb_build.sh --java-test 'org.yb.pgsql.TestPgBaseScansCostModel' ./yb_build.sh --java-test 'org.yb.pgsql.TestPgCostModelSeekNextEstimation' ``` Reviewers: gkukreja, mtakahara, amartsinchyk, tnayak Reviewed By: mtakahara Subscribers: yql, gkukreja Tags: #jenkins-ready Differential Revision: https://phorge.dev.yugabyte.com/D35136
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Jira Link: DB-11285
Description
bitmap_exceeded_work_mem_costis 5.0e9. It's easy to miss when someone try to find all the places that inflates the cost values, e.g. investigate into cost overflow, precision losses with complex plans, etc.Issue Type
kind/enhancement
Warning: Please confirm that this issue does not contain any sensitive information
The text was updated successfully, but these errors were encountered: