Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Use file path as cache key (#1852) #1862

Merged
merged 3 commits into from
May 7, 2021
Merged

Use file path as cache key (#1852) #1862

merged 3 commits into from
May 7, 2021

Conversation

ti-srebot
Copy link
Collaborator

@ti-srebot ti-srebot commented May 6, 2021 •
edited
Loading

cherry-pick #1852 to release-5.0
You can switch your code base to this Pull Request by using git-extras:

# In tics repo:
git pr https://github.com/pingcap/tics/pull/1862

After apply modifications, you can push your change to this PR via:

git push [email protected]:ti-srebot/tics.git pr/1862:release-5.0-7bb9a84fe7f2

What problem does this PR solve?

Issue Number: close #1352

Problem Summary:

What is changed and how it works?

What's Changed:

Use file path as cache key instead of a 128 bits hash code.

Related changes

  • PR to update pingcap/docs/pingcap/docs-cn:
  • Need to cherry-pick to the release branch:

Check List

Tests

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

Side effects

Release note

@ti-srebot ti-srebot mentioned this pull request May 6, 2021
Merged
@ti-srebot ti-srebot added CHERRY-PICK cherry pick status/LGT1 Indicates that a PR has LGTM 1. type/bugfix This PR fixes a bug. labels May 6, 2021
@ti-srebot ti-srebot added this to the v5.0.2 milestone May 6, 2021
ilovesoup
ilovesoup reviewed May 6, 2021
View reviewed changes
dbms/src/Storages/MarkCache.h
{
private:
using Base = LRUCache<UInt128, MarksInCompressedFile, UInt128TrivialHash, MarksWeightFunction>;
using Base = LRUCache<String, MarksInCompressedFile, std::hash<String>, MarksWeightFunction>;
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Indeed it seems to me a silly design before, do you know why clickhouse designed like this first place? Saving memory? How much extra memory will it cost?

Copy link
Contributor

@JinheLin JinheLin May 6, 2021 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think it saves very little memory. The value of this cache is a file's content (I guess, maybe several hundred KB or several MB). Normally, the full path of the file is less than 1KB. So, in this LRU cache, memory consumed by keys are less than 1%.

I guess it is optimized for the speed of comparison.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

For this design, I think the main concern for ClickHouse is the performance. They don't need to compare the real path, only the hash code.

@JinheLin
Copy link
Contributor

JinheLin commented May 6, 2021

/run-all-tests

@JinheLin
Copy link
Contributor

JinheLin commented May 6, 2021

/build

@JaySon-Huang
Copy link
Contributor

Mark those hash_salt unused or delete them at all as #1866 does.

@JinheLin
Copy link
Contributor

JinheLin commented May 7, 2021

/run-all-tests

@JinheLin
Copy link
Contributor

JinheLin commented May 7, 2021

/merge

@ti-srebot ti-srebot added the status/can-merge Indicates a PR has been approved by a committer. label May 7, 2021
@ti-srebot
Copy link
Collaborator Author

/run-all-tests

@ti-srebot ti-srebot merged commit 5e3efcb into pingcap:release-5.0 May 7, 2021
JaySon-Huang pushed a commit to JaySon-Huang/tiflash that referenced this pull request Jun 8, 2021
@ti-srebot @JaySon-Huang
Fix split on empty segment (pingcap#1823)
468cadd 
Signed-off-by: ti-srebot <[email protected]>

Co-authored-by: Flowyi <[email protected]>

cherry pick pingcap#1821 to release-5.0 (pingcap#1827)

Signed-off-by: ti-srebot <[email protected]>

Co-authored-by: xufei <[email protected]>

Remove ReplicatedMergeTree family (pingcap#1805) (pingcap#1826)

cherry pick pingcap#1835 to release-5.0 (pingcap#1840)

Signed-off-by: ti-srebot <[email protected]>

Co-authored-by: xufei <[email protected]>

Use file path as cache key (pingcap#1852) (pingcap#1862)

Lazily initializing DeltaMergeStore  (pingcap#1423) (pingcap#1751) (pingcap#1868)

Add background gc check thread for DeltaTree storage (pingcap#1742) (pingcap#1828)

* cherry pick pingcap#1742 to release-5.0

Signed-off-by: ti-srebot <[email protected]>

* fix conflict

Co-authored-by: lidezhu <[email protected]>

Fix calculate stable property (pingcap#1839) (pingcap#1878)

* cherry pick pingcap#1839 to release-5.0

Signed-off-by: ti-srebot <[email protected]>

* fix compile and cherry pick a small fix

Co-authored-by: lidezhu <[email protected]>

Fix client-c mistake setting current_ts to 0 when resolving lock for async commit (pingcap#1856) (pingcap#1870)

support constant folding in dbg coprocessor (pingcap#1881) (pingcap#1884)

Optimize & Reduce time cost about ci test (pingcap#1849) (pingcap#1894)

* reduce time cost about ci test by parallel
* add `-DNO_WERROR=ON` to cmake config for release-darwin build
* Fix tidb_ghpr_tics_test fail (pingcap#1895)

Signed-off-by: Zhigao Tong <[email protected]>

Fix panic with feature `compaction filter` in release-5.0 (pingcap#1891)

Init store in background task. (pingcap#1843,pingcap#1896) (pingcap#1874)

Push down filter on timestamp type column to storage layer (pingcap#1875) (pingcap#1906)

* cherry pick pingcap#1875 to release-5.0

Signed-off-by: ti-srebot <[email protected]>

* remove irrevalant code

Co-authored-by: lidezhu <[email protected]>

Fix ExchangeSender: remove duplicated write stream operation (pingcap#1379) (pingcap#1883)

cherry pick pingcap#1917 to release-5.0 (pingcap#1923)

Signed-off-by: ti-srebot <[email protected]>

Co-authored-by: Fu Zhe <[email protected]>

Pre-handle SSTFiles to DTFiles when applying snapshots (pingcap#1439) (pingcap#1867)

* Add `raft.snapshot.method` in configuration file to control the method of applying snapshots
  - "block" decode snapshot data as a block
  - "file1" decode snapshot data as DTFiles (directory mode) [**By default**]
* Add a stream `SSTFilesToBlockInputStream` to decode snapshot data into blocks
* Add a stream `BoundedSSTFilesToBlockInputStream` to bound the blocks read from SSTFilesToBlockInputStream by column `_tidb_rowid`
* Add a stream `SSTFilesToDTFilesOutputStream` that accept `BoundedSSTFilesToBlockInputStream` as input stream to write blocks into DTFiles
* Make `STORAGE_FORMAT_CURRENT` to be `STORAGE_FORMAT_V2` by default to support ingest DTFile into DT engine
* Fix the bug that the block decoded from SSTFile is not sorted by primary key and version (pingcap#1888)
* Fix panic with feature compaction filter with DTFile
* Fix ingest write amplification after snapshot apply optimization (pingcap#1913)

Co-authored-by: Zhigao Tong <[email protected]>
Signed-off-by: JaySon-Huang <[email protected]>

Cast int/real as real (pingcap#1911) (pingcap#1928)

Fix the problem that old dmfile is not removed atomically (pingcap#1918) (pingcap#1925)

Eliminate the table lock between reading,writing and DDL operators for TiFlash (pingcap#1736) (pingcap#1858)

* Eliminate the table lock between reading,writing and DDL operators for TiFlash (pingcap#1736)

1. Port the latest `RWLock` for `IStorage` from Clickhouse. "phase-fair rwlocks have been shown to provide more predictable timings compared to task-fair rwlocks for certain workloads (for relatively low share of writes, <25%). Brandenburg B, Anderson J, (2010)"
2. Refactor the lock model for `IStorage` to eliminate the lock between reading, writing, and DDL operations.
3. Acquire table lock by QueryID/threadName instead of function name
4. Tombstone the database after receiving a "drop" DDL from TiDB. Physically drop the databases / tables after gc safe point.
5. Remove some outdated tests under `tests/mutable-test/txn_schema` (all those tests are ported into `tests/delta-merge-test/raft/schema`)

* Mark FIXME for conflict codes
Conflicts:
  dbms/src/Flash/Coprocessor/DAGQueryBlockInterpreter.cpp

Signed-off-by: JaySon-Huang <[email protected]>

cached tics code incase of clone same code repeatedly (pingcap#1994) (pingcap#2001)

Add row & bytes threshold config for CompactLog (pingcap#1997) (pingcap#2005)

* cherry pick pingcap#1997 to release-5.0

Signed-off-by: ti-srebot <[email protected]>
Co-authored-by: Zhigao Tong <[email protected]>

Fix the bug that incomplete write batches are not truncated (pingcap#1934) (pingcap#2003)

Revert "Push down filter on timestamp type column to storage layer (pingcap#1875) (pingcap#1906)" (pingcap#2010)

This reverts commit 630b612.

Revert Lazily Init Store (pingcap#2011)

* Revert "Init store in background task. (pingcap#1843,pingcap#1896) (pingcap#1874)"

This reverts commit 0882461.

Conflicts:
	dbms/src/Storages/StorageDeltaMerge.cpp

* format code

* Revert "Lazily initializing DeltaMergeStore  (pingcap#1423) (pingcap#1751) (pingcap#1868)"

This reverts commit bbce050.

Conflicts:
	dbms/src/Storages/DeltaMerge/tests/gtest_dm_storage_delta_merge.cpp
	dbms/src/Storages/StorageDeltaMerge.cpp
	dbms/src/Storages/StorageDeltaMerge.h

Co-authored-by: JinheLin <[email protected]>
Co-authored-by: Flowyi <[email protected]>

Revert "background gc thread" (pingcap#2015)

* Revert "Fix calculate stable property (pingcap#1839) (pingcap#1878)"

This reverts commit 6a9eb58.

* Revert "Add background gc check thread for DeltaTree storage (pingcap#1742) (pingcap#1828)"

This reverts commit cbbbd09.

* remove code

* format code

* format code

* fix test compile

* fix comment

Fix the potential concurrency problem when clone the shared delta index (pingcap#2030) (pingcap#2033)

* cherry pick pingcap#2030 to release-5.0

Signed-off-by: ti-srebot <[email protected]>

* Update dbms/src/Storages/DeltaMerge/DeltaIndex.h

Co-authored-by: JaySon <[email protected]>

Co-authored-by: lidezhu <[email protected]>
Co-authored-by: JaySon <[email protected]>

Fix bug for unexpected deleted ingest file (pingcap#2047) (pingcap#2048)

* Only enable GC for DTFiles after they get applied to all segments.
* Refine some loggings

Signed-off-by: JaySon-Huang <[email protected]>

Fix cast int as time (pingcap#1788) (pingcap#1893)

cherry pick pingcap#1903 to release-5.0 (pingcap#1915)

Signed-off-by: ti-srebot <[email protected]>

Fix ingesting file may add ref page to deleted page (pingcap#2054) (pingcap#2055)

1. Put the ingest file id into `storage_pool.data` before ingesting them into segments
2. Add ref pages to the ingest files for each segment
3. Delete the original page id after all

Signed-off-by: JaySon-Huang <[email protected]>

Fix compile error on Mac Clang 12.0.5 (pingcap#2058) (pingcap#2070)

cherry pick pingcap#2017 pingcap#2020 pingcap#2051 to release-5.0 (pingcap#2074)

Revert "cherry pick pingcap#2017 pingcap#2020 pingcap#2051 to release-5.0 (pingcap#2074)"

This reverts commit 8e0712c.

Revert "Fix compile error on Mac Clang 12.0.5 (pingcap#2058) (pingcap#2070)"

This reverts commit 3c31b92.

Revert "Fix ingesting file may add ref page to deleted page (pingcap#2054) (pingcap#2055)"

This reverts commit c773d61.

Revert "Fix bug for unexpected deleted ingest file (pingcap#2047) (pingcap#2048)"

This reverts commit b329618.

Revert "Pre-handle SSTFiles to DTFiles when applying snapshots (pingcap#1439) (pingcap#1867)"

This reverts commit c947fd6.

 Conflicts:
	dbms/src/Common/FailPoint.cpp
	dbms/src/Storages/DeltaMerge/File/DMFileBlockOutputStream.h
	dbms/src/Storages/DeltaMerge/File/DMFileWriter.h
	dbms/src/Storages/DeltaMerge/ReorganizeBlockInputStream.h
	dbms/src/Storages/DeltaMerge/SSTFilesToBlockInputStream.cpp
	dbms/src/Storages/DeltaMerge/SSTFilesToBlockInputStream.h
	dbms/src/Storages/DeltaMerge/SSTFilesToDTFilesOutputStream.cpp
	dbms/src/Storages/DeltaMerge/Segment.cpp
	dbms/src/Storages/DeltaMerge/StableValueSpace.cpp
	dbms/src/Storages/DeltaMerge/tests/gtest_dm_file.cpp
	dbms/src/Storages/DeltaMerge/tests/gtest_dm_storage_delta_merge.cpp
	dbms/src/Storages/StorageDeltaMerge.cpp
	dbms/src/Storages/StorageDeltaMerge.h
	dbms/src/Storages/Transaction/ApplySnapshot.cpp
	dbms/src/Storages/Transaction/PartitionStreams.cpp
	tests/delta-merge-test/raft/schema/drop_on_restart.test

Fix comment

Signed-off-by: JaySon-Huang <[email protected]>
@JaySon-Huang JaySon-Huang mentioned this pull request Jun 8, 2021
Merged
@LittleFall LittleFall deleted the release-5.0-7bb9a84fe7f2 branch June 19, 2021 05:45
flowbehappy pushed a commit that referenced this pull request Jun 19, 2021
@ti-srebot @flowbehappy
Use file path as cache key (#1852) (#1862)
7b5b229
flowbehappy added a commit that referenced this pull request Jun 20, 2021
@flowbehappy
Revert "Use file path as cache key (#1852) (#1862)"
4d52b2e 
This reverts commit 7b5b229.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@flowbehappy flowbehappy flowbehappy approved these changes

@ilovesoup ilovesoup ilovesoup left review comments

@JinheLin JinheLin JinheLin left review comments

@JaySon-Huang JaySon-Huang Awaiting requested review from JaySon-Huang

Assignees

@JinheLin JinheLin

Labels
CHERRY-PICK cherry pick status/can-merge Indicates a PR has been approved by a committer. status/LGT1 Indicates that a PR has LGTM 1. type/bugfix This PR fixes a bug.
Projects
None yet
Milestone
v5.0.2
Development

Successfully merging this pull request may close these issues.
5 participants
@ti-srebot @JinheLin @JaySon-Huang @flowbehappy @ilovesoup