ddl: Fix adding vector index on empty table may blocked forever by JaySon-Huang · Pull Request #9470 · pingcap/tiflash

JaySon-Huang · 2024-09-25T10:26:56Z

What problem does this PR solve?

Issue Number: ref #9032

Problem Summary:

When creating vector index on empty table, the DDL may be blocked forever.

This is because when an empty table is synced to TiFlash, the StorageDeltaMerge is created without initializing the DeltaMergeStore instance. This is to avoid generating too many fragmented, small files that exhaust the filesystem inode.
However, if the DeltaMergeStore instance is not created, then system.dt_local_index will not get the local index stats. Which make tidb consider the vector index is still under building and the ddl does not end.

tiflash/dbms/src/Storages/System/StorageSystemDTLocalIndexes.cpp

Lines 88 to 94 in 1a781c7

    
           auto dm_storage = std::dynamic_pointer_cast<StorageDeltaMerge>(storage); 
        
           const auto & table_info = dm_storage->getTableInfo(); 
        
           auto table_id = table_info.id; 
        
           auto store = dm_storage->getStoreIfInited(); 
        
           if (!store) 
        
               continue;

What is changed and how it works?

If StorageDeltaMerge is created but DeltaMergeStore is not inited, then generate the local index info by the TiDB::TableInfo

Small refactor:

Remove a verbose logging when system table is read by tidb-server

Check List

Tests

Unit test
Integration test
Manual test (add detailed scripts or steps below)
No code

Side effects

Performance regression: Consumes more CPU
Performance regression: Consumes more Memory
Breaking backward compatibility

Documentation

Release note

None

JaySon-Huang · 2024-09-25T10:28:26Z

dbms/src/Interpreters/InterpreterSelectQuery.cpp

            || (managed_storage->engineType() != ::TiDB::StorageEngine::DT
                && managed_storage->engineType() != ::TiDB::StorageEngine::TMT))
        {
-            LOG_DEBUG(log, "{}.{} is not ManageableStorage", database_name, table_name);


Running queries on the system table will trigger this logging, which is verbose but not provide any information. Remove it.

Lloyd-Pottiger · 2024-09-25T11:50:13Z

Why not just insert a row even when store is not init?

breezewish · 2024-09-25T11:56:25Z

Not sure whether this idea is ok, we may also regard progress=1 and missing table in tiflash_indexes as index build is finished, because progress=1 already means TiFlash is aware of this table, and then no entry in tiflash_indexes means exactly this case?

JaySon-Huang · 2024-09-25T12:02:45Z

Not sure whether this idea is ok, we may also regard progress=1 and missing table in tiflash_indexes as index build is finished, because progress=1 already means TiFlash is aware of this table, and then no entry in tiflash_indexes means exactly this case?

@breezewish No, we can not tell the difference between these two cases

A table without any vector index but the tiflash replica is already built
A table with the tiflash replica is already built, tidb added a vector index definition to the table, but tiflash is not aware that the vector index has been added

JaySon-Huang · 2024-09-25T12:05:41Z

@Lloyd-Pottiger DeltaMergeStore::getLocalIndexStats requires a DeltaMergeStore instance to get the stats of local indexes. Without that result, we are hard to tell wihch index_id need to be added to dt_local_index

tiflash/dbms/src/Storages/System/StorageSystemDTLocalIndexes.cpp

Lines 91 to 98 in 1a781c7

    
           auto store = dm_storage->getStoreIfInited(); 
        
           if (!store) 
        
               continue; 
        
           if (dm_storage->isTombstone()) 
        
               continue; 
        
           auto index_stats = store->getLocalIndexStats();

dbms/src/Storages/StorageDeltaMerge.cpp

Lloyd-Pottiger · 2024-09-25T13:11:32Z

@Lloyd-Pottiger DeltaMergeStore::getLocalIndexStats requires a DeltaMergeStore instance to get the stats of local indexes. Without that result, we are hard to tell wihch index_id need to be added to dt_local_index

tiflash/dbms/src/Storages/System/StorageSystemDTLocalIndexes.cpp

Lines 91 to 98 in 1a781c7

auto store = dm_storage->getStoreIfInited();

if (!store)

continue;

if (dm_storage->isTombstone())

continue;

auto index_stats = store->getLocalIndexStats();

The index info can get from table info, and the rows indexed is 0.

wuhuizuo · 2024-09-27T10:57:18Z

/retest

This reverts commit 682892a.

JaySon-Huang · 2024-09-27T11:09:57Z

@Lloyd-Pottiger @JinheLin PTAL

Lloyd-Pottiger

lgtm

Lloyd-Pottiger · 2024-09-27T12:25:13Z

Need to update description.

JaySon-Huang · 2024-09-29T02:19:44Z

Need to update description.

Updated

ti-chi-bot · 2024-09-29T03:42:25Z

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: JinheLin, Lloyd-Pottiger

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Details

Needs approval from an approver in each of these files:

~~OWNERS~~ [JinheLin,Lloyd-Pottiger]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

ti-chi-bot · 2024-09-29T03:42:27Z

[LGTM Timeline notifier]

Timeline:

2024-09-27 12:06:40.027206939 +0000 UTC m=+10955.447419944: ☑️ agreed by Lloyd-Pottiger.
2024-09-29 03:42:26.686555973 +0000 UTC m=+153502.106768987: ☑️ agreed by JinheLin.

…cap#9470) ref pingcap#9032

ti-chi-bot bot added do-not-merge/work-in-progress Indicates that a PR should not merge because it is a work in progress. release-note-none Denotes a PR that doesn't merit a release note. size/L Denotes a PR that changes 100-499 lines, ignoring generated files. labels Sep 25, 2024

JaySon-Huang commented Sep 25, 2024

View reviewed changes

JaySon-Huang changed the title ~~[WIP] ddl: Fix adding vector index on empty table may blocked forever~~ ddl: Fix adding vector index on empty table may blocked forever Sep 25, 2024

ti-chi-bot bot removed the do-not-merge/work-in-progress Indicates that a PR should not merge because it is a work in progress. label Sep 25, 2024

JaySon-Huang requested review from JinheLin, Lloyd-Pottiger and breezewish September 25, 2024 11:21

JaySon-Huang commented Sep 25, 2024

View reviewed changes

dbms/src/Storages/StorageDeltaMerge.cpp Outdated Show resolved Hide resolved

ti-chi-bot bot added size/XL Denotes a PR that changes 500-999 lines, ignoring generated files. and removed size/L Denotes a PR that changes 100-499 lines, ignoring generated files. labels Sep 26, 2024

JaySon-Huang force-pushed the fix_store_not_init branch from 2337578 to 2b98e1f Compare September 26, 2024 07:46

JaySon-Huang force-pushed the fix_store_not_init branch from b60ba2a to 62b868d Compare September 26, 2024 08:28

ti-chi-bot bot added size/L Denotes a PR that changes 100-499 lines, ignoring generated files. and removed size/XL Denotes a PR that changes 500-999 lines, ignoring generated files. labels Sep 26, 2024

JaySon-Huang force-pushed the fix_store_not_init branch from 6f3a586 to aa597a9 Compare September 27, 2024 10:04

JaySon-Huang force-pushed the fix_store_not_init branch from aa597a9 to 56546e4 Compare September 27, 2024 10:53

JaySon-Huang added 7 commits September 27, 2024 19:05

Remove useless func

2f4a62b

Add fullstack test case

f3b2358

Generate index stats from TableInfo for non-inited store

5bbc783

Revert "Remove useless func"

eec731e

This reverts commit 682892a.

Pick some comment refine

b362ba3

Add unit test

051702d

Resolve conflict

679765d

JaySon-Huang force-pushed the fix_store_not_init branch from 56546e4 to 679765d Compare September 27, 2024 11:08

Lloyd-Pottiger approved these changes Sep 27, 2024

View reviewed changes

ti-chi-bot bot added needs-1-more-lgtm Indicates a PR needs 1 more LGTM. approved labels Sep 27, 2024

JinheLin approved these changes Sep 29, 2024

View reviewed changes

ti-chi-bot bot added lgtm and removed needs-1-more-lgtm Indicates a PR needs 1 more LGTM. labels Sep 29, 2024

ti-chi-bot bot merged commit cd204f5 into pingcap:feature/vector-index Sep 29, 2024

JaySon-Huang deleted the fix_store_not_init branch September 29, 2024 03:54

JaySon-Huang mentioned this pull request Sep 30, 2024

*: support vector search #9486

Merged

12 tasks

JaySon-Huang added a commit to JaySon-Huang/tiflash that referenced this pull request Oct 14, 2024

ddl: Fix adding vector index on empty table may blocked forever (ping…

698a9ef

…cap#9470) ref pingcap#9032

	auto dm_storage = std::dynamic_pointer_cast<StorageDeltaMerge>(storage);
	const auto & table_info = dm_storage->getTableInfo();
	auto table_id = table_info.id;
	auto store = dm_storage->getStoreIfInited();
	if (!store)
	continue;

Conversation

JaySon-Huang commented Sep 25, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What problem does this PR solve?

What is changed and how it works?

Check List

Release note

Uh oh!

JaySon-Huang Sep 25, 2024

Choose a reason for hiding this comment

Uh oh!

Lloyd-Pottiger commented Sep 25, 2024

Uh oh!

breezewish commented Sep 25, 2024

Uh oh!

JaySon-Huang commented Sep 25, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

JaySon-Huang commented Sep 25, 2024

Uh oh!

Uh oh!

Lloyd-Pottiger commented Sep 25, 2024

Uh oh!

wuhuizuo commented Sep 27, 2024

Uh oh!

JaySon-Huang commented Sep 27, 2024

Uh oh!

Lloyd-Pottiger left a comment

Choose a reason for hiding this comment

Uh oh!

Lloyd-Pottiger commented Sep 27, 2024

Uh oh!

JaySon-Huang commented Sep 29, 2024

Uh oh!

ti-chi-bot bot commented Sep 29, 2024

Uh oh!

ti-chi-bot bot commented Sep 29, 2024

[LGTM Timeline notifier]

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

JaySon-Huang commented Sep 25, 2024 •

edited

Loading

JaySon-Huang commented Sep 25, 2024 •

edited

Loading