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

Fix semaphore leak when EntryMemTable#addEntry accepts the same entries #2832

Merged
merged 1 commit into from
Oct 16, 2021
Merged

Fix semaphore leak when EntryMemTable#addEntry accepts the same entries #2832

merged 1 commit into from
Oct 16, 2021

Conversation

BewareMyPower
Copy link
Contributor

Motivation

If EntryMemTable#addEntry accepts the same entries (assuming size is N), which means they have the same ledger id and entry id, the semaphore will be acquired multiple times (assuming M times) but only one entry will be added to kvmap. When the EntryMemTable is flushed, only N permits will be released, while M * N permits are acquired. Then the semaphore leak happens.

Changes

In EntryMemTable#addEntry, release the acquired permits if internalAdd returns 0. Because it means the entry size is zero or the entry has already existed in kvmap. testAddSameEntries is added to verify when the same entries are added, flush can still release all acquired permits.

@BewareMyPower
Copy link
Contributor Author

Here's a block thread in production. We can see addEntry stuck at skipListSemaphore.acquireUninterruptibly(len), which means the semaphore is exhausted.

image

PTAL. @eolivelli @dlg99 @nicoloboschi @hangc0276 @codelipenghui

nicoloboschi
nicoloboschi approved these changes Oct 16, 2021
View reviewed changes
Copy link
Contributor

@nicoloboschi nicoloboschi left a comment

Choose a reason for hiding this comment

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

LGTM

@merlimat merlimat merged commit a9b576d into apache:master Oct 16, 2021
hsaputra pushed a commit that referenced this pull request Oct 20, 2021
@Vanlightly
BP-44 USE metrics
f6903b8 
BP-44: USE metrics. A proposal for improving BookKeeper metrics so that operators can employ the USE method for diagnosing performance issues.


Reviewers: Henry Saputra <hsaputra@apache.org>, Andrey Yegorov <None>, Enrico Olivelli <eolivelli@gmail.com>

This closes #2835 from Vanlightly/BP-44-use-metrics and squashes the following commits:

8d9baab [Jack Vanlightly] Added link to USE method and listed each term of USE
5a0f67d [Jack Vanlightly] BP-44 USE metrics
a9b576d [Yunze Xu] Release semaphore when addEntry accepts the same entries (#2832)
148bf22 [Yun Tang] Ensure to release cache during KeyValueStorageRocksDB#closec (#2821)
4dc4260 [gaozhangmin] Heap memory leak problem when ledger replication failed (#2794)
a522fa3 [Raúl Gracia] Issue 2815: Upgrade to log4j2 to get rid of CVE-2019-17571 (#2816)
0465052 [Nicolò Boschi] Upgrade httpclient from 4.5.5 to 4.5.13 (#2793)
594a056 [Raúl Gracia] Issue 2795: Bookkeeper upgrade using Bookie ID may fail due to cookie mismatch (#2796)
354cf37 [Raúl Gracia] Upgraded dependencies with CVEs (#2792)
e413c70 [Raúl Gracia] Issue 2728: Entry Log GC may get blocked when using entryLogPerLedgerEnabled option (#2779)
883231e [pradeepbn] Building bookkeeper with gradle on java11
zymap pushed a commit that referenced this pull request Oct 26, 2021
@BewareMyPower @zymap
Release semaphore when addEntry accepts the same entries (#2832)
82c4e1c 
(cherry picked from commit a9b576d)
hsaputra pushed a commit that referenced this pull request Oct 29, 2021
@nicoloboschi
Gradle build - some submodules are skipped
98ddf81 
### Motivation

In order to complete migration to Gradle we must build all the subprojects.

### Changes

- Enabled `sh` integration tests with gradle, located in `tests/scripts/src/test/bash/gradle`
- Added these modules to the build
    - `bookkeeper-http:servlet-http-server` 
    - `metadata-drivers:etcd`
    - `tests:backward-compat:*`
    - `tests:shaded:*`
    - `stream:bk-grpc-name-resolver`
- DL shading process is now performed (before it didn't build any jar)
- Groovy tests (`tests:backward-compat:*`) now are triggered by the build/tests itself; with Maven, there is a "runner" project (`tests/integration-tests-base-groovy`); in Gradle is useless so it is skipped


### Test

- Both `bin/bookkeper standalone` and `bin/bookkeper_gradle standalone` work locally
- Tests are passing locally  

Master Issue: #2849



Reviewers: Henry Saputra <hsaputra@apache.org>, Prashant Kumar <None>

This closes #2850 from nicoloboschi/fix/2849/gradle and squashes the following commits:

00b49f4 [Nicolò Boschi] Fix common_gradle.sh regex
bd739fd [Nicolò Boschi] fix sh tests
43230ba [Nicolò Boschi] revert sh files. Avoid to modify maven files, create gradle versions to faciltate migration
d1f95e4 [Nicolò Boschi] fix shaded deps
bcab40d [Nicolò Boschi] fix build
5fd0341 [Nicolò Boschi] fix build
0082e0e [Nicolò Boschi] fix build
2c32ac1 [Nicolò Boschi] fixes
3bc0b26 [Nicolò Boschi] bookkeeper-server-shaded-tests
ba89132 [Nicolò Boschi] shaded tests
6d39e33 [Nicolò Boschi] sh tests
e0032bc [Nicolò Boschi] actually run arquillian groovy tests
08dcc39 [Nicolò Boschi] backwards
2361f79 [Nicolò Boschi] hierarchical-ledger-manager
8388e11 [Nicolò Boschi] current-server-old-clients
6a24344 [Nicolò Boschi] bc-non-fips
2faca01 [Nicolò Boschi] bk-grpc-name-resolver
991bc11 [Nicolò Boschi] servlet-http-server
675ef7b [Nicolò Boschi] etcd
b1d5e14 [ZhangJian He] A empty implement in EtcdLedgerManagerFactory to let the project can compile (#2845)
bd5c50b [shustsud] Add error handling to readLedgerMetadata in over-replicated ledger GC (#2844)
746f9f6 [Matteo Merli] Remove direct ZK access for Auditor (#2842)
4117200 [ZhangJian He] the compare should be >= instead of > (#2782)
14ef56f [Prashant Kumar] BookieId can not be cast to BookieSocketAddress (#2843)
e10f3fe [ZhangJian He] Forget to close preAllocator log on shutdown (#2819)
53954ca [shustsud] Add ensemble check to over-replicated ledger GC (#2813)
919fdd3 [Prashant Kumar] Issue:2840 Create bookie shellscript for gradle (#2841)
031d168 [gaozhangmin] fix-npe-when-pulsar-ZkBookieRackAffinityMapping-getBookieAddressResolver (#2788)
3dd671c [Prashant Kumar] Migrate bookkeepr-server:test to gradle run unit tests excepts org.apache.bookkeeper.bookie. org.apache.bookkeeper.client org.apache.bookkeeper.replication org.apache.bookkeeper.tls. (#2812)
f6903b8 [Jack Vanlightly] BP-44 USE metrics
a4afaa4 [Matteo Merli] Eliminate direct ZK access in ScanAndCompareGarbageCollector (#2833)
a9b576d [Yunze Xu] Release semaphore when addEntry accepts the same entries (#2832)
148bf22 [Yun Tang] Ensure to release cache during KeyValueStorageRocksDB#closec (#2821)
4dc4260 [gaozhangmin] Heap memory leak problem when ledger replication failed (#2794)
a522fa3 [Raúl Gracia] Issue 2815: Upgrade to log4j2 to get rid of CVE-2019-17571 (#2816)
0465052 [Nicolò Boschi] Upgrade httpclient from 4.5.5 to 4.5.13 (#2793)
594a056 [Raúl Gracia] Issue 2795: Bookkeeper upgrade using Bookie ID may fail due to cookie mismatch (#2796)
354cf37 [Raúl Gracia] Upgraded dependencies with CVEs (#2792)
e413c70 [Raúl Gracia] Issue 2728: Entry Log GC may get blocked when using entryLogPerLedgerEnabled option (#2779)
883231e [pradeepbn] Building bookkeeper with gradle on java11
Ghatage pushed a commit to sijie/bookkeeper that referenced this pull request Jul 12, 2024
@BewareMyPower
Release semaphore when addEntry accepts the same entries (apache#2832)
aa298a2
Ghatage pushed a commit to sijie/bookkeeper that referenced this pull request Jul 12, 2024
@Vanlightly
BP-44 USE metrics
b7ca3cd 
BP-44: USE metrics. A proposal for improving BookKeeper metrics so that operators can employ the USE method for diagnosing performance issues.


Reviewers: Henry Saputra <hsaputra@apache.org>, Andrey Yegorov <None>, Enrico Olivelli <eolivelli@gmail.com>

This closes apache#2835 from Vanlightly/BP-44-use-metrics and squashes the following commits:

8d9baab [Jack Vanlightly] Added link to USE method and listed each term of USE
5a0f67d [Jack Vanlightly] BP-44 USE metrics
a9b576d [Yunze Xu] Release semaphore when addEntry accepts the same entries (apache#2832)
148bf22 [Yun Tang] Ensure to release cache during KeyValueStorageRocksDB#closec (apache#2821)
4dc4260 [gaozhangmin] Heap memory leak problem when ledger replication failed (apache#2794)
a522fa3 [Raúl Gracia] Issue 2815: Upgrade to log4j2 to get rid of CVE-2019-17571 (apache#2816)
0465052 [Nicolò Boschi] Upgrade httpclient from 4.5.5 to 4.5.13 (apache#2793)
594a056 [Raúl Gracia] Issue 2795: Bookkeeper upgrade using Bookie ID may fail due to cookie mismatch (apache#2796)
354cf37 [Raúl Gracia] Upgraded dependencies with CVEs (apache#2792)
e413c70 [Raúl Gracia] Issue 2728: Entry Log GC may get blocked when using entryLogPerLedgerEnabled option (apache#2779)
883231e [pradeepbn] Building bookkeeper with gradle on java11
Ghatage pushed a commit to sijie/bookkeeper that referenced this pull request Jul 12, 2024
@nicoloboschi
Gradle build - some submodules are skipped
ff821cb 
### Motivation

In order to complete migration to Gradle we must build all the subprojects.

### Changes

- Enabled `sh` integration tests with gradle, located in `tests/scripts/src/test/bash/gradle`
- Added these modules to the build
    - `bookkeeper-http:servlet-http-server` 
    - `metadata-drivers:etcd`
    - `tests:backward-compat:*`
    - `tests:shaded:*`
    - `stream:bk-grpc-name-resolver`
- DL shading process is now performed (before it didn't build any jar)
- Groovy tests (`tests:backward-compat:*`) now are triggered by the build/tests itself; with Maven, there is a "runner" project (`tests/integration-tests-base-groovy`); in Gradle is useless so it is skipped


### Test

- Both `bin/bookkeper standalone` and `bin/bookkeper_gradle standalone` work locally
- Tests are passing locally  

Master Issue: apache#2849



Reviewers: Henry Saputra <hsaputra@apache.org>, Prashant Kumar <None>

This closes apache#2850 from nicoloboschi/fix/2849/gradle and squashes the following commits:

00b49f4 [Nicolò Boschi] Fix common_gradle.sh regex
bd739fd [Nicolò Boschi] fix sh tests
43230ba [Nicolò Boschi] revert sh files. Avoid to modify maven files, create gradle versions to faciltate migration
d1f95e4 [Nicolò Boschi] fix shaded deps
bcab40d [Nicolò Boschi] fix build
5fd0341 [Nicolò Boschi] fix build
0082e0e [Nicolò Boschi] fix build
2c32ac1 [Nicolò Boschi] fixes
3bc0b26 [Nicolò Boschi] bookkeeper-server-shaded-tests
ba89132 [Nicolò Boschi] shaded tests
6d39e33 [Nicolò Boschi] sh tests
e0032bc [Nicolò Boschi] actually run arquillian groovy tests
08dcc39 [Nicolò Boschi] backwards
2361f79 [Nicolò Boschi] hierarchical-ledger-manager
8388e11 [Nicolò Boschi] current-server-old-clients
6a24344 [Nicolò Boschi] bc-non-fips
2faca01 [Nicolò Boschi] bk-grpc-name-resolver
991bc11 [Nicolò Boschi] servlet-http-server
675ef7b [Nicolò Boschi] etcd
b1d5e14 [ZhangJian He] A empty implement in EtcdLedgerManagerFactory to let the project can compile (apache#2845)
bd5c50b [shustsud] Add error handling to readLedgerMetadata in over-replicated ledger GC (apache#2844)
746f9f6 [Matteo Merli] Remove direct ZK access for Auditor (apache#2842)
4117200 [ZhangJian He] the compare should be >= instead of > (apache#2782)
14ef56f [Prashant Kumar] BookieId can not be cast to BookieSocketAddress (apache#2843)
e10f3fe [ZhangJian He] Forget to close preAllocator log on shutdown (apache#2819)
53954ca [shustsud] Add ensemble check to over-replicated ledger GC (apache#2813)
919fdd3 [Prashant Kumar] Issue:2840 Create bookie shellscript for gradle (apache#2841)
031d168 [gaozhangmin] fix-npe-when-pulsar-ZkBookieRackAffinityMapping-getBookieAddressResolver (apache#2788)
3dd671c [Prashant Kumar] Migrate bookkeepr-server:test to gradle run unit tests excepts org.apache.bookkeeper.bookie. org.apache.bookkeeper.client org.apache.bookkeeper.replication org.apache.bookkeeper.tls. (apache#2812)
f6903b8 [Jack Vanlightly] BP-44 USE metrics
a4afaa4 [Matteo Merli] Eliminate direct ZK access in ScanAndCompareGarbageCollector (apache#2833)
a9b576d [Yunze Xu] Release semaphore when addEntry accepts the same entries (apache#2832)
148bf22 [Yun Tang] Ensure to release cache during KeyValueStorageRocksDB#closec (apache#2821)
4dc4260 [gaozhangmin] Heap memory leak problem when ledger replication failed (apache#2794)
a522fa3 [Raúl Gracia] Issue 2815: Upgrade to log4j2 to get rid of CVE-2019-17571 (apache#2816)
0465052 [Nicolò Boschi] Upgrade httpclient from 4.5.5 to 4.5.13 (apache#2793)
594a056 [Raúl Gracia] Issue 2795: Bookkeeper upgrade using Bookie ID may fail due to cookie mismatch (apache#2796)
354cf37 [Raúl Gracia] Upgraded dependencies with CVEs (apache#2792)
e413c70 [Raúl Gracia] Issue 2728: Entry Log GC may get blocked when using entryLogPerLedgerEnabled option (apache#2779)
883231e [pradeepbn] Building bookkeeper with gradle on java11
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@nicoloboschi nicoloboschi nicoloboschi approved these changes

@dlg99 dlg99 dlg99 approved these changes

@merlimat merlimat merlimat approved these changes

Assignees

@BewareMyPower BewareMyPower

Labels
release/4.14.3
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
5 participants
@BewareMyPower @merlimat @dlg99 @nicoloboschi @zymap