[ZOOKEEPER-3642] Fix potential data inconsistency due to DIFF sync after partial SNAP sync by lvfangmin · Pull Request #1224 · apache/zookeeper

lvfangmin · 2020-01-17T01:45:25Z

No description provided.

…ter partial SNAP sync

lvfangmin · 2020-02-12T23:01:39Z

@eolivelli @hanm interested to take a look at this, the potential issue is mentioned in the description of the Jira.

hanm

LGTM, with two nits.

hanm · 2020-02-13T04:34:28Z

    public synchronized void shutdown(boolean fullyShutDown) {
        if (!canShutdown()) {
+            if (fullyShutDown && zkDb != null) {
+                zkDb.clear();


It's not obvious to me how we could end up in this code path - if we are here then the shutdown must have been called and that specific shut down must not be a full shutdown. Do we have a concrete example when we will hit this code path?

In case the follower exited before finish syncing with leader, the ZK server will not start, canShutdown will return false, but we still need to clear the db here.

hanm · 2020-02-13T04:36:05Z

+               new byte[1], Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT);
+
+            // wait some time to let this get written to disk
+            Thread.sleep(500);


I think in general we advocate not using sleep (at least, not using sleep alone) in unit tests as it's not reliable and proved to be a source of flaky-ness. Is there a better approach to wait for the log being flushed to disk (preferably also verify that fact in test)?

That's a reasonable concern, I'll try to find ways to check that in the test here.

hanm · 2020-10-19T03:26:10Z

@eolivelli do you want to give a look at this given you were tagged?

If no issues I will merge this once we get a green build on per-merge job (I think my nits are addressed - the sleeping in test is not ideal but I would not consider it's blocking).

eolivelli

LGTM

@hanm please go ahead and merge

hanm · 2020-10-20T01:06:48Z

ok, I will merge after CI is back to good shape.

eolivelli · 2020-10-20T10:38:16Z

@hanm I have sent a fix for CI
#1510

hanm · 2020-10-21T03:09:13Z

thanks for quick fix on CI issue @eolivelli ; looks like the test needs update to work with latest master. I'll do a rebase if @lvfangmin didn't have a chance to rebase.

eolivelli · 2020-10-21T15:49:22Z

thank you @hanm
the fix should be pretty simple, ping me when you are ready

[2020-10-21T02:26:03.778Z] [ERROR] Failed to execute goal org.apache.maven.plugins:maven-compiler-plugin:3.8.1:testCompile (default-testCompile) on project zookeeper: Compilation failure

[2020-10-21T02:26:03.778Z] [ERROR] /home/jenkins/jenkins-home/workspace/eper-precommit-github-pr_PR-1224/zookeeper-server/src/test/java/org/apache/zookeeper/server/quorum/QuorumPeerMainTest.java:[1733,17] no suitable method found for assertNotNull(java.lang.String,org.apache.zookeeper.data.Stat)

[2020-10-21T02:26:03.778Z] [ERROR]     method org.junit.jupiter.api.Assertions.assertNotNull(java.lang.Object,java.util.function.Supplier<java.lang.String>) is not applicable

[2020-10-21T02:26:03.778Z] [ERROR]       (argument mismatch; org.apache.zookeeper.data.Stat cannot be converted to java.util.function.Supplier<java.lang.String>)

[2020-10-21T02:26:03.778Z] [ERROR]     method org.junit.jupiter.api.Assertions.assertNotNull(java.lang.Object,java.lang.String) is not applicable

[2020-10-21T02:26:03.778Z] [ERROR]       (argument mismatch; org.apache.zookeeper.data.Stat cannot be converted to java.lang.String)

[2020-10-21T02:26:03.778Z] [ERROR] 

[2020-10-21T02:26:03.778Z] [ERROR] -> [Help 1]

hanm · 2020-10-24T01:16:06Z

@eolivelli #1515 is rebased version, thanks.

…er partial SNAP sync. Based on #1224 ; fixed unit test build issue. Author: Fangmin Lyu <fangmin@apache.org> Author: Michael Han <hanm@apache.org> Reviewers: Enrico Olivelli <eolivelli@apache.org>, Originally developed by Fangmin Lyu <fangmin@apache.org> Closes #1515 from hanm/ZOOKEEPER-3642

…er partial SNAP sync. Based on #1224 ; fixed unit test build issue. Author: Fangmin Lyu <fangmin@apache.org> Author: Michael Han <hanm@apache.org> Reviewers: Enrico Olivelli <eolivelli@apache.org>, Originally developed by Fangmin Lyu <fangmin@apache.org> Closes #1515 from hanm/ZOOKEEPER-3642 (cherry picked from commit a53cfeb)

…er partial SNAP sync. Based on apache#1224 ; fixed unit test build issue. Author: Fangmin Lyu <fangmin@apache.org> Author: Michael Han <hanm@apache.org> Reviewers: Enrico Olivelli <eolivelli@apache.org>, Originally developed by Fangmin Lyu <fangmin@apache.org> Closes apache#1515 from hanm/ZOOKEEPER-3642

kezhuw · 2024-08-12T14:18:16Z

Superceded by #1515.

[ZOOKEEPER-3642] Fix potential data inconsistency due to DIFF sync af…

f8eb68d

…ter partial SNAP sync

hanm approved these changes Feb 13, 2020

View reviewed changes

eolivelli approved these changes Oct 19, 2020

View reviewed changes

hanm mentioned this pull request Oct 24, 2020

[ZOOKEEPER-3642] Fix potential data inconsistency due to DIFF sync after partial SNAP sync. #1515

Closed

jonmv mentioned this pull request Sep 23, 2022

ZOOKEEPER-4541 Ephemeral znode owned by closed session visible in 1 of 3 servers #1925

Closed

ztzg force-pushed the master branch from 1c60545 to e2070be Compare October 3, 2023 12:57

kezhuw closed this Aug 12, 2024

Uh oh!

Conversation

lvfangmin commented Jan 17, 2020

Uh oh!

lvfangmin commented Feb 12, 2020

Uh oh!

hanm left a comment

Choose a reason for hiding this comment

Uh oh!

hanm Feb 13, 2020

Choose a reason for hiding this comment

Uh oh!

lvfangmin Feb 26, 2020

Choose a reason for hiding this comment

Uh oh!

hanm Feb 13, 2020

Choose a reason for hiding this comment

Uh oh!

lvfangmin Feb 26, 2020

Choose a reason for hiding this comment

Uh oh!

hanm commented Oct 19, 2020

Uh oh!

eolivelli left a comment

Choose a reason for hiding this comment

Uh oh!

hanm commented Oct 20, 2020

Uh oh!

eolivelli commented Oct 20, 2020

Uh oh!

hanm commented Oct 21, 2020

Uh oh!

eolivelli commented Oct 21, 2020

Uh oh!

hanm commented Oct 24, 2020

Uh oh!

kezhuw commented Aug 12, 2024

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants